Professor Kevin Wells

The advent of learning with noisy labels (LNL), multi-rater learning, and human-AI collaboration has revolutionised the development of robust classifiers, enabling them to address the challenges posed by different types of data imperfections and complex decision processes commonly encountered in real-world applications. While each of these methodologies has individually made significant strides in addressing their unique challenges, the development of techniques that can simultaneously tackle these three problems remains underexplored. This paper addresses this research gap by integrating noisy-label learning, multi-rater learning, and human-AI collaboration with new benchmarks and the innovative Learning to Complement with Multiple Humans (LECOMH) approach. LECOMH optimises the level of human collaboration during testing, aiming to optimise classification accuracy while minimising collaboration costs that vary from 0 to M, where M is the maximum number of human collaborators. We quantitatively compare LECOMH with leading human-AI collaboration methods using our proposed benchmarks. LECOMH consistently outperforms the competition, with accuracy improving as collaboration costs increase. Notably, LECOMH is the only method enhancing human labeller performance across all benchmarks.

Background Sleep disturbance is common among people living with dementia as well as their caregivers. Non‐contact video technology can be used to characterise such disturbances as well as quantifying sleep quality by measuring the number of sleep body positions (poses). Such an approach may be beneficial for home‐based longitudinal clinical monitoring of sleep pattern changes and disturbances at all stages of dementia. Here we present our pilot results of a personalised data‐driven method applied to video data for quantification of sleep disturbance comparing older and younger participants. Method Data were collected in two separate studies which included an overnight 10‐12 hour laboratory sleep recording from thirteen older (65‐80 years, 9 male:4 female) and eleven younger (18‐34 years, 7 male:4 female) participants in a dedicated sleep facility. A data‐driven analysis using Principal Component Analysis and k‐means clustering was applied to infrared video data extracted from a clinical polysomnography (PSG) system. The data‐driven analysis automatically determined statistically significant groupings or clusters of unique body poses for each individual. Pose number, number of pose transitions, pose duration, and pose transition duration were computed for each participant. Result The number of data‐driven poses in older and younger participants was remarkably similar with 15.2±3.9 and 15.8±1.8 (mean±SD) poses per participant, respectively. However, the older group had a higher number of pose transitions (33.0±8.82) compared to the younger group (23.9±6.83) (p = 0.03). A significant 20% difference (p = 0.03) in the average duration of each body position was observed, with 62.3±7.1 minutes and 78.9±15.9 minutes for the older and younger groups, respectively (see Figure 1 and Figure 2). Pose transition duration was 19.2±7.25 seconds and 14.6±4.32 seconds for the older and younger groups, respectively where they were not significantly different. Conclusion Although the number of body positions did not vary significantly between the two cohorts, the older group changed body position more frequently and it took them longer to do so. Data‐driven automated analysis of video‐based sleep monitoring holds significant promise for quantifying age‐related and inter‐individual differences in sleep behavior.

In this study, a novel hybrid tensor factorisation and deep learning approach has been proposed and implemented for sleep pose identification and classification of twelve different sleep postures. We have applied tensor factorisation to infrared (IR) images of 10 subjects to extract group-level data patterns, undertake dimensionality reduction and reduce occlusion for IR images. Pre-trained VGG-19 neural network has been used to predict the sleep poses under the blanket. Finally, we compared our results with those without the factorisation stage and with CNN network. Our new pose detection method outperformed the methods solely based on VGG-19 and 4-layer CNN network. The average accuracy for 10 volunteers increased from 78.1% and 75.4% to 86.0%.

Pruritus is a common clinical sign in dogs and is often underrecognized by dog owners and veterinarians. The Whistle FIT ® , a wearable accelerometer paired with analytics, can detect changes in pruritic activity in dogs, which can be reported to owners in a smartphone/tablet application. The objectives of this retrospective observational study were to investigate the impact of digital alerts for increased pruritic behaviors received by dog owners in a real-life setting, on (1) the initiation of veterinary clinic visits, and (2) if such visits resulted in initiation of therapy for pruritus. Whistle FIT ® data and electronic health records from 1,042 Banfield veterinary clinics in the United States were obtained for a 20-month period and reviewed retrospectively. Data on times of increased pruritic behaviors was calculated retrospectively by the investigators by applying the same algorithms used in the Whistle system. Data from the first 10-month interval was compared to the second 10 months, when reports on pruritic behaviors and alerts for increased pruritic behaviors were viewable by pet owners. Signalment of dogs with clinic visits in the first ( n  = 7,191) and second ( n  = 6,684) 10-month groups was similar. The total number of pruritic alerts was 113,530 in the first 10 months and 93,217 in the second 10 months. The odds of an ‘alert visit’ (the first veterinary clinic visit that occurred within 4 weeks after the time of a pruritus alert) was statistically significantly more likely (odds ratio, 1.6264; 95% CI, 1.57–1.69; p  

Necrosis seen in histopathology Whole Slide Images is a major criterion that contributes towards scoring tumour grade which then determines treatment options. However conventional manual assessment suffers from inter-operator reproducibility impacting grading precision. To address this, automatic necrosis detection using AI may be used to assess necrosis for final scoring that contributes towards the final clinical grade. Using deep learning AI, we describe a novel approach for automating necrosis detection in Whole Slide Images, tested on a canine Soft Tissue Sarcoma (cSTS) data set consisting of canine Perivascular Wall Tumours (cPWTs). A patch-based deep learning approach was developed where different variations of training a DenseNet-161 Convolutional Neural Network architecture were investigated as well as a stacking ensemble. An optimised DenseNet-161 with post-processing produced a hold-out test F1-score of 0.708 demonstrating state-of-the-art performance. This represents a novel first-time automated necrosis detection method in the cSTS domain as well specifically in detecting necrosis in cPWTs demonstrating a significant step forward in reproducible and reliable necrosis assessment for improving the precision of tumour grading.

The definitive diagnosis of canine soft-tissue sarcomas (STSs) is based on histological assessment of formalin-fixed tissues. Assessment of parameters, such as degree of differentiation, necrosis score and mitotic score, give rise to a final tumour grade, which is important in determining prognosis and subsequent treatment modalities. However, grading discrepancies are reported to occur in human and canine STSs, which can result in complications regarding treatment plans. The introduction of digital pathology has the potential to help improve STS grading via automated determination of the presence and extent of necrosis. The detected necrotic regions can be factored in the grading scheme or excluded before analysing the remaining tissue. Here we describe a method to detect tumour necrosis in histopathological whole-slide images (WSIs) of STSs using machine learning. Annotated areas of necrosis were extracted from WSIs and the patches containing necrotic tissue fed into a pre-trained DenseNet161 convolutional neural network (CNN) for training, testing and validation. The proposed CNN architecture reported favourable results, with an overall validation accuracy of 92.7% for necrosis detection which represents the number of correctly classified data instances over the total number of data instances. The proposed method, when vigorously validated represents a promising tool to assist pathologists in evaluating necrosis in canine STS tumours, by increasing efficiency, accuracy and reducing inter-rater variation.

Abstract for ISPOR Europe 2022 poster presentation. Social media are seldom explored in animal health despite the potential for insights into pet owners’ perceptions. Owners often seek information and advice online before seeking veterinary care. The aim was to investigate owners’ perceptions of feline allergic skin disease using Social Asset, a proof-of-concept social listening (SL) platform to create a dataset concerning information-seeking behaviours.

Respiratory motion correction degrades quantitatively and qualitatively Nuclear Medicine images. We propose that adaptive approaches are required to correct for the irregular breathing patterns often encountered in the clinical setting, which can be addressed within a Bayesian tracking formulation. This allows inference of the hidden organ configurations using only knowledge of an external observation such as a parametrized external surface. The flexible framework described provides a method to correct for organ motion whilst accommodating for irregular unseen respiratory patterns. In this work we utilize a Kalman filter and compare it with a Particle filter. A novel adaptive state transition model is also introduced to describe the evolution of organ configurations. The Kalman filter marginally outperforms the Particle filter, both approaches however offer an effective motion correction mechanism, correcting for motion with errors of around 1-3mm. We present results of simulated PET images derived from XCAT to demonstrate the efficacy of the approach. © 2012 IEEE.

A wafer scale CMOS Active Pixel Sensor has been designed employing design techniques of transistor enclosed geometry and P+ doped guard rings to offer ionizing radiation tolerance. The detector was irradiated with 160 kVp X-rays up to a total dose of 94 kGy(Si) and remained functional. The radiation damage produced in the device has been studied, resulting in a dark current density increase per decade of 96±5 pA/cm/decade and a damage threshold of 204 Gy(Si). The damage produced in the detector has been compared with a commercially available CMOS APS, showing a radiation tolerance about 100 times higher. Moreover Monte Carlo simulations have been performed to evaluate primary and secondary energy deposition in each of the detector stages. © 2012 IEEE.

In this present study, tumour (3D) locations are predicted via external surface motion, extracted from abdomen/ thoracic surface measurements that can be used to enhance dose targeting in external beam radiotherapy. Canonical Correlation Analysis (CCA) is applied to the surface and tumour motion data to maximise the correlation between them. This correlation is exploited for motion prediction [1]. Nine dynamic CT datasets were used to extract the surface and tumour motion and to create the Canonical Correlation model (CCM). Gaussian Mixture Regression (GMR) and Adaptive Kernel Density Estimation (AKDE) were trained on these nine datasets to predict the respiratory signal by updating the surface motion and CCM. A leave-one-out method was used to evaluate and compare the performance of GMR and AKDE in predicting the tumour motion. © 2012 IEEE.

The most important factors that affect the image quality are contrast, spatial resolution and noise. These factors and their relationship are quantitatively described by the Contrast-to-Noise Ratio (CNR), Signal-to-Noise Ratio (SNR), Modulation Transfer Function (MTF), Noise Power Spectrum (NPS) and Detective Quantum Efficiency (DQE) parameters. The combination of SNR, MTF and NPS determines the DQE, which represents the ability to visualize object details of a certain size and contrast at a given dose. In this study the performance of a novel large area Complementary Metal-Oxide-Semiconductor (CMOS) Active Pixel Sensor (APS) X-ray detector, called DynAMITe (Dynamic range Adjustable for Medical Imaging Technology), was investigated and compared to other three digital mammography systems (namely a) Large Area Sensor (LAS), b) Hamamatsu C9732DK, and c) Anrad SMAM), in terms of physical characteristics and evaluation of the image quality. DynAMITe detector consists of two geometrically superimposed grids: a) 2560 × 2624 pixels at 50 μm pitch, named Sub-Pixels (SP camera) and b) 1280 × 1312 pixels at 100 μm pitch, named Pixels (P camera). The X-ray performance evaluation of DynAMITe SP detector demonstrated high DQE results (0.58 to 0.64 at 0.5 lp/mm). Image simulation based on the X-ray performance of the detectors was used to predict and compare the mammographic image quality using ideal software phantoms: a) one representing two three dimensional (3-D) breasts of various thickness and glandularity to estimate the CNR between simulated microcalcifications and the background, and b) the CDMAM 3.4 test tool for a contrast-detail analysis of small thickness and low contrast objects. The results show that DynAMITe SP detector results in high CNR and contrast-detail performance. © 2012 IEEE.

With the continual improvement in spatial resolution of Nuclear Medicine (NM) scanners, it has become increasingly important to accurately compensate for patient motion during image acquisition. Respiratory motion produced by normal lung ventilation is a major source of artefacts in NM emission imaging that can affect large parts of the abdominal thoracic cavity. As such, a particle filter (PF) is proposed as a powerful method for motion correction in emission imaging which can successfully account for previously unseen motion. This paper explores a basic PF approach and demonstrates that it is possible to estimate temporally non-stationary motion using training data consisting of only a single respiratory cycle. Evaluation using the XCAT phantom suggests that the PF is a highly promising approach, and can appropriately handle the complex data that arises in clinical situations.

Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) renography, in common with other medical imaging techniques, is influenced by respiratory motion. As a result, data quantification may be inaccurate. This work presents a novel on-line approach for motion correction by implementing a spatio-temporal independent component analysis method (STICA). This methodology firstly results in removal of motion artefacts and secondly provides independent components that have physiological characteristics. The STICA was applied to 10 healthy volunteers' renal DCE-MRI data. The results were evaluated using independent component curve gradients (ICGs) from different regions of interest and by comparing them with the Rutland-Patlak (RP) analysis. The r values for the ICGs were significantly higher compared to the RP curves. The standard deviations of the IC curve gradients also showed less dispersion with comparison to the RP curve gradients across all the ten volunteers' renal data. © 2012 IEEE.

The recent development of compact, cold cathode X-ray emitters presents the opportunity to use multiple emitters arranged in an array to perform medical imaging from a stationary source. This study considers some of the potential advantages in their application to digital breast tomosynthesis and breast specimen imaging. It presents modelling results from simulated images based on a validated toolkit that models the X-ray source, breast tissue and X-ray image receptor performance. It also presents results from physical experiments using flat panel source geometries. Stationary sources eliminate focal spot motion blur which can improve the visualization of small features such as microcalcifications. A rectangular array of emitters can provide enhanced depth resolution compared to a linear sweep of emitter positions, as the projection images are obtained over a 'sweep' in two directions. Moreover, a rectangular array gives the opportunity to reduce the source to image distance (SID). Halving this distance reduces the required beam current by four times, which could significantly reduce the size, weight, and cost of an overall system, albeit with an increase in the angular width of the focal spot. Both the simulation and physical experiments show that depth resolution is considerably improved when using a rectangular array in comparison to a linear array. If a rectangular array is used at half the SID, and a quarter of the power, then this benefit is maintained, even though the focal spot has a larger angular width. This could provide enhanced imaging from lower-cost, smaller devices compared with conventional systems.

Estimating population-level burden, abilities of pet-parents to identify disease and demand for veterinary services worldwide is challenging. The purpose of this study is to compare a feline pruritus survey with social media listening (SML) data discussing this condition. Surveys are expensive and labour intensive to analyse but SML data is freeform and requires careful filtering for relevancy. This study considers data from a survey of owner-observed symptoms of 156 pruritic cats conducted using Pet Parade® and SML posts collected through web-scraping, to gain insights into the characterisation and management of feline pruritus. SML posts meeting a feline body area, behaviour and symptom were captured and reviewed for relevance representing 1299 public posts collected from 2021 to 2023. The survey involved 1067 pet-parents who reported on pruritic symptoms in their cats. Among the observed cats, approximately 18.37% (n=196) exhibited at least one symptom. The most frequently reported symptoms were hair loss (9.2%), bald spots (7.3%) and infection, crusting, scaling, redness, scabbing, scaling, or bumpy skin (8.2%). Notably, bald spots were the primary symptom reported for short-haired cats, while other symptoms were more prevalent in medium and long-haired cats. Affected body areas, according to pet-parents, were primarily the head, face, chin, neck (27%), and the top of the body, along the spine (22%). 35% of all cats displayed excessive behaviours consistent with pruritic skin disease. Interestingly, 27% of these cats were perceived as non-symptomatic by their owners, suggesting an under-identification of itch-related signs. Furthermore, a significant proportion of symptomatic cats did not receive any skin disease medication whether prescribed or over the counter (n=41). These findings indicate a higher incidence of pruritic skin disease in cats than recognized by pet owners, potentially leading to a lack of medical intervention for clinically symptomatic cases. The comparison between the survey and social media listening data revealed bald spots were reported in similar proportions in both datasets (25% in the survey and 28% in SML). Infection, crusting, scaling, redness, scabbing, scaling, or bumpy skin accounted for 31% of symptoms in the survey, whereas it represented 53% of relevant SML posts (excluding bumpy skin). Abnormal licking or chewing behaviours were mentioned by pet-parents in 40% of SML posts compared to 38% in the survey. The consistency in the findings of these two disparate data sources, including a complete overlap in affected body areas for the top 80% of social media listening posts, indicates minimal biases in each method, as significant biases would likely yield divergent results. Therefore, the strong agreement across pruritic symptoms, affected body areas, and reported behaviours enhances our confidence in the reliability of the findings. Moreover, the small differences identified between the datasets underscore the valuable insights that arise from utilising multiple data sources. These variations

Digital breast tomosynthesis (DBT) is currently under consideration for replacement of, or combined use with 2D-mammography in national breast screening programmes. To investigate the potential benefits that DBT can bring to screening, the threshold detectable lesion diameters were measured for different forms of DBT in comparison to 2D-mammography. The aim of this study was to compare the threshold detectable mass diameters obtained with narrow angle (15°/15 projections) and wide angle (50°/25 projections) DBT in comparison to 2Dmammography. Simulated images of 60mm thick compressed breasts were produced with and without masses using a set of validated image modelling tools for 2D-mammography and DBT. Image processing and reconstruction were performed using commercial software. A series of 4-alternative forced choice (4AFC) experiments was conducted for signal detection with the masses as targets. The threshold detectable mass diameter was found for each imaging modality with a mean glandular dose of 2.5 mGy. The resulting values of the threshold diameter for 2D-mammography (10.2 ± 1.4 mm) were found to be larger (p < 0.001) than those for narrow angle DBT (6.0 ± 1.1 mm) and wide angle DBT (5.6 ± 1.2 mm). There was no significant difference between the threshold diameters for wide and narrow angle DBT. Implications for the introduction of DBT alone or in combination with 2D-mammography in breast cancer screening are discussed.

Respiratory motion degrades quantitative and qualitative analysis of medical images. Estimation and hence correction of motion commonly uses static correspondence models between an external surrogate signal and internal motion. This work presents a patient specific respiratory motion model with the ability to adapt in the presence of irregular motion via a Kalman filter with Expectation Maximisation for parameter estimation. The adaptive approach introduces generalizability allowing the model to account for a broader variety of motion. This may be required in the presence of irregular breathing and with different sensors monitoring the external surrogate signal. The motion model framework utilizing an adaptive Kalman filter approach is tested on dynamic MRI data of nine volunteers and compared to a state-of-the-art static total least squares approach. Results demonstrate the framework is capable of reducing motion to the order of < 3mm and is significantly (p < 0:001) more effective in the presence of irregular motion, assessed using the F test for model comparison. Utilizing the total sum of squares of estimated vector field error from the calculated ground truth, we observe approximately a fifty percent reduction in root mean square error and thirty percent reduction in standard deviation utilizing the Kalman model (EKF) in comparison to a static counterpart.

A new method of generating realistic three dimensional simulated breast lesions known as diffusion limited aggregation (DLA) is presented, and compared with the random walk (RW) method. Both methods of lesion simulation utilize a physics-based method for inserting these simulated lesions into 2D clinical mammogram images that takes into account the polychromatic x-ray spectrum, local glandularity and scatter. DLA and RW masses were assessed for realism via a receiver operating characteristic (ROC) study with nine observers. The study comprised 150 images of which 50 were real pathology proven mammograms, 50 were normal mammograms with RW inserted masses and 50 were normal mammograms with DLA inserted masses. The average area under the ROC curve for the DLA method was 0.55 (95% confidence interval 0.51-0.59) compared to 0.60 (95% confidence interval 0.56-0.63) for the RW method. The observer study results suggest that the DLA method produced more realistic masses with more variability in shape compared to the RW method. DLA generated lesions can overcome the lack of complexity in structure and shape in many current methods of mass simulation.

Nuclear Medicine (NM) imaging serves as a powerful diagnostic tool for imaging of biochemical and physiological processes in vivo. The degradation in spatial image resolution caused by the often irregular respiratory motion must be corrected to achieve high resolution imaging. In order perform motion correction more accurately, it is proposed that patient motion obtained from 4D MRI can be used to analyse respiratory motion. To extract motion from the dynamic MRI dataset an organ wise intensity based affine registration framework is proposed and evaluated. Comparison of the resultant motion obtained within selected organs is made against an open source free form deformation algorithm. For validation, the correlation of the results of both techniques to a previous study of motion in 20 patients is found. Organwise affine registration correlates very well (r = 0:9) with a previous study (Segars et al., 2007)1 whilst free form deformation shows little correlation (r = 0:3). This increases the confidence of the organ wise affine registration framework being an effective tool to extract motion from dynamic anatomical datasets. © 2013 SPIE.

CCD and CMOS imaging technologies can be applied to thin tissue Autoradiography as potential imaging alternatives to using conventional film. In this work, we compare two particular devices; a CCD operating in slow scan mode and a CMOS-based Active Pixel sensor, operating at near video rates. Both imaging sensors have been operated at room temperature with images produced from calibrated microscales and radiolabelled tissue samples. We also compare these digital imaging technologies with the use of conventional film. We show first comparative results obtained with 14C calibrated microscales and 35S radiolabelled tissue sections. We also present first results of 3H images produced under direct irradiation of a CCD sensor operating at room temperature. Compared to film, silicon-based imaging technologies exhibit enhanced sensitivity, dynamic range and linearity.

Nuclear Medicine (NM) imaging is currently the most sensitive approach for functional imaging of the human body. However, in order to achieve high-resolution imaging, one of the factors degrading the detail or apparent resolution in the reconstructed image, namely respiratory motion, has to be overcome. All respiratory motion correction approaches depend on some assumption or estimate of respiratory motion. In this paper, the respiratory motion found from 4D MRI is analysed and characterised. The characteristics found are compared with previous studies and will be incorporated into the process of estimating respiratory motion. © 2013 SPIE.

Digital breast tomosynthesis (DBT) is under consideration to replace or to be used in combination with 2D-mammography in breast screening. The aim of this study was the comparison of the detection of microcalcification clusters by human observers in simulated breast images using 2D-mammography, narrow angle (15°/15 projections) and wide angle (50°/25 projections) DBT. The effects of the cluster height in the breast and the dose to the breast on calcification detection were also tested. Simulated images of 6 cm thick compressed breasts were produced with and without microcalcification clusters inserted, using a set of image modelling tools for 2D-mammography and DBT. Image processing and reconstruction were performed using commercial software. A series of 4-alternative forced choice (4AFC) experiments was conducted for signal detection with the microcalcification clusters as targets. Threshold detectable calcification diameter was found for each imaging modality with standard dose: 2D-mammography: 2D-mammography (165 ± 9 µm), narrow angle DBT (211 ± 11 µm) and wide angle DBT (257 ± 14 µm). Statistically significant differences were found when using different doses, but different geometries had a greater effect. No differences were found between the threshold detectable calcification diameters at different heights in the breast. Calcification clusters may have a lower detectability using DBT than 2D imaging.

Planar 2D x-ray mammography is generally accepted as the preferred screening technique used for breast cancer detection. Recently, digital breast tomosynthesis (DBT) has been introduced to overcome some of the inherent limitations of conventional planar imaging, and future technological enhancements are expected to result in the introduction of further innovative modalities. However, it is crucial to understand the impact of any new imaging technology or methodology on cancer detection rates and patient recall. Any such assessment conventionally requires large scale clinical trials demanding significant investment in time and resources. The concept of virtual clinical trials and virtual performance assessment may offer a viable alternative to this approach. However, virtual approaches require a collection of specialized modelling tools which can be used to emulate the image acquisition process and simulate images of a quality indistinguishable from their real clinical counterparts. In this paper, we present two image simulation chains constructed using modelling tools that can be used for the evaluation of 2D-mammography and DBT systems. We validate both approaches by comparing simulated images with real images acquired using the system being simulated. A comparison of the contrast-to-noise ratios and image blurring for real and simulated images of test objects shows good agreement ( 

Compensation for respiratory motion has been identified as a crucial factor in achieving high resolution Nuclear Medicine (NM) imaging. Many motion correction approaches have been studied and they are seen to have advantages over simpler approaches such as respiratory gating. However, all motion correction approaches rely on an assumption or estimation of respiratory motion. This paper builds upon previous work in recursive Bayesian estimation of respiratory motion assuming a stereo camera observation of the motion of the external torso surface. This paper compares the performance of a modified autoregressive transition model against the previously presented linear transition model used when estimating motion within a 4D dataset generated from the XCAT phantom. © 2013 SPIE.

The steady rise of the breast cancer screening population, coupled with data expansion produced by new digital screening technologies (tomosynthesis/CT) motivates the development of new, more efficient image screening processes. Rapid Serial Visual Presentation (RSVP) is a new fast-content recognition approach which uses electroencephalography to record brain activity elicited by fast bursts of image data. These brain responses are then subjected to machine classification methods to reveal the expert's 'reflex' response to classify images according to their presence or absence of particular targets. The benefit of this method is that images can be presented at high temporal rates (∼10 per second), faster than that required for fully conscious detection, facilitating a high throughput of image (screening) material. In the present paper we present the first application of RSVP to medical image data, and demonstrate how cortically coupled computer vision can be successfully applied to breast cancer screening. Whilst prior RSVP work has utilised multichannel approaches, we also present the first RSVP results demonstrating discriminatory response on a single electrode with a ROC area under the curve of 0.62-0.86 using a simple Fisher discriminator for classification. This increases to 0.75-0.94 when multiple electrodes are used in combination. © 2013 SPIE.

Background: Recent studies including an innovative machine learning technique indicated Chiari-like malformation (CM) is influenced by brachycephalic features. Objectives: Morphometric analysis of facial anatomy and dysmorphia in CM-associated pain (CM-P) and syringomyelia (SM) in the Cavalier King Charles Spaniel (CKCS). Animals:Sixty-six client-owned CKCS. Methods:Retrospective study of anonymized T2W sagittal magnetic resonance imaging of 3 clinical groups: (1) 11 without central canal dilation (ccd) or SM (CM-N),(2) 15 with CM-P with no SM or

A novel method has been developed for generating quasi-realistic voxel phantoms which simulate the compressed breast in mammography and digital breast tomosynthesis (DBT). The models are suitable for use in virtual clinical trials requiring realistic anatomy which use the multiple alternative forced choice (AFC) paradigm and patches from the complete breast image. The breast models are produced by extracting features of breast tissue components from DBT clinical images including skin, adipose and fibro-glandular tissue, blood vessels and Cooper’s ligaments. A range of different breast models can then be generated by combining these components. Visual realism was validated using a receiver operating characteristic (ROC) study of patches from simulated images calculated using the breast models and from real patient images. Quantitative analysis was undertaken using fractal dimension and power spectrum analysis. The average areas under the ROC curves for 2D and DBT images were 0.51 ± 0.06 and 0.54 ± 0.09 demonstrating that simulated and real images were statistically indistinguishable by expert breast readers (7 observers); errors represented as one standard error of the mean. The average fractal dimensions (2D, DBT) for real and simulated images were (2.72 ± 0.01, 2.75 ± 0.01) and (2.77 ± 0.03, 2.82 ± 0.04) respectively; errors represented as one standard error of the mean. Excellent agreement was found between power spectrum curves of real and simulated images, with average β values (2D, DBT) of (3.10 ± 0.17, 3.21 ± 0.11) and (3.01 ± 0.32, 3.19 ± 0.07) respectively; errors represented as one standard error of the mean. These results demonstrate that radiological images of these breast models realistically represent the complexity of real breast structures and can be used to simulate patches from mammograms and DBT images that are indistinguishable from patches from the corresponding real breast images. The method can generate about 500 radiological patches (~30mm × 30mm) per day for AFC experiments on a single workstation. This is the first study to quantitatively validate the realism of simulated radiological breast images using direct blinded comparison with real data via the ROC paradigm with expert breast readers

Purpose Virtual clinical trials (VCT) are a powerful imaging tool that can be used to investigate digital breast tomosynthesis (DBT) technology. In this work, a fast and simple method is proposed to estimate the two‐dimensional distribution of scattered radiation which is needed when simulating DBT geometries in VCTs. Methods Monte Carlo simulations are used to precalculate scatter‐to‐primary ratio (SPR) for a range of low‐resolution homogeneous phantoms. The resulting values can be used to estimate the two‐dimensional (2D) distribution of scattered radiation arising from inhomogeneous anthropomorphic phantoms used in VCTs. The method has been validated by comparing the values of the scatter thus obtained against the results of direct Monte Carlo simulation for three different types of inhomogeneous anthropomorphic phantoms. Results Differences between the proposed scatter field estimation method and the ground truth data for the OPTIMAM phantom had an average modulus and standard deviation of over the projected breast area of 2.4 ± 0.9% (minimum −17.0%, maximum 27.7%). The corresponding values for the University of Pennsylvania and Duke University breast phantoms were 1.8 ± 0.1% (minimum −8.7%, maximum 8.0%) and 5.1 ± 0.1% (minimum −16.2%, maximum 7.4%), respectively. Conclusions The proposed method, which has been validated using three of the most common breast models, is a useful tool for accurately estimating scattered radiation in VCT schemes used to study current designs of DBT system.

© Published under licence by IOP Publishing Ltd.It is well known that respiratory motion affects image acquisition and also external beam radiotherapy (EBRT) treatment planning and delivery. However often the existing approaches for respiratory motion management are based on a generic view of respiratory motion such as the general movement of organ, tissue or fiducials. This paper thus aims to present a more in depth analysis of respiratory motion based on 4D MRI for further integration into motion correction in image acquisition or image based EBRT. Internal and external motion was first analysed separately, on a per-organ basis for internal motion. Principal component analysis (PCA) was then performed on the internal and external motion vectors separately and the relationship between the two PCA spaces was analysed. The motion extracted from 4D MRI on general was found to be consistent with what has been reported in literature.

There are many applications for which sparse, or partial sampling of dynamic image data can be used for articulating or estimating motion within the medical imaging area. In this new work, we propose a generalized framework for dense motion propagation from sparse samples which represents an example of transfer learning and manifold alignment, allowing the transfer of knowledge across imaging sources of different domains which exhibit different features. Many such examples exist in medical imaging, from mapping 2D ultrasound or fluoroscopy to 3D or 4D data or monitoring dynamic dose delivery from partial imaging data in radiotherapy. To illustrate this approach we animate, or articulate, a high resolution static MR image with 4D free breathing respiratory motion derived from low resolution sparse planar samples of motion. In this work we demonstrate that sparse sampling of dynamic MRI may be used as a viable approach to successfully build models of freebreathing respiratory motion by constrained articulation. Such models demonstrate high contrast, and high temporal and spatial resolution that have so far been hitherto unavailable with conventional imaging methods. The articulation is based on using a propagation model, in the eigen domain, to estimate complete 4D motion vector fields from sparsely sampled free-breathing dynamic MRI data. We demonstrate that this approach can provide equivalent motion vector fields compared to fully sampled 4D dynamic data, whilst preserving the corresponding high resolution / high contrast inherent in the original static volume. Validation is performed on three 4D MRI datasets using 8 extracted slices from a fast 4D acquisition (0.7sec per volume). The estimated deformation fields from sparse sampling are compared to the fully sampled equivalents, resulting in an rms error of the order of 2mm across the entire image volume. We also present exemplar 4D high contrast, high resolution articulated volunteer datasets using this methodology. This approach facilitates greater freedom in the acquisition of free breathing respiratory motion sequences which may be used to inform motion modelling methods in a range of imaging modalities and demonstrates that sparse sampling of free breathing data may be used within a manifold alignment and transfer learning paradigm to estimate fully sampled motion. The method may also be applied to other examples of sparse sampling to produce dense motion propagation.

This is the first study of partial volume effect in quantifying renal function on dynamic contrast enhanced magnetic resonance imaging. Dynamic image data were acquired for a cohort of 10 healthy volunteers. Following respiratory motion correction, each voxel location was assigned a mixing vector representing the 'overspilling' contributions of each tissue due to the convolution action of the imaging system's point spread function. This was used to recover the true intensities associated with each constituent tissue. Thus, non-renal contributions from liver, spleen and other surrounding tissues could be eliminated from the observed time-intensity curves derived from a typical renal cortical region of interest. This analysis produced a change in the early slope of the renal curve, which subsequently resulted in an enhanced glomerular filtration rate estimate. This effect was consistently observed in a Rutland-Patlak analysis of the time-intensity data: the volunteer cohort produced a partial volume effect corrected mean enhancement of 36% in relative glomerular filtration rate with a mean improvement of 7% in r(2) fitting of the Rutland-Patlak model compared to the same analysis undertaken without partial volume effect correction. This analysis strongly supports the notion that dynamic contrast enhanced magnetic resonance imaging of kidneys is substantially affected by the partial volume effect, and that this is a significant obfuscating factor in subsequent glomerular filtration rate estimation. (C) 2009 Elsevier Ireland Ltd. All rights reserved.

CCD (charged coupled device) and CMOS imaging technologies can be applied to thin tissue autoradiography as potential imaging alternatives to using conventional film. In this work, we compare two particular devices: a CCD operating in slow scan mode and a CMOS-based active pixel sensor, operating at near video rates. Both imaging sensors have been operated at room temperature using direct irradiation with images produced from calibrated microscales and radiolabelled tissue samples. We also compare these digital image sensor technologies with the use of conventional film. We show comparative results obtained with (14)C calibrated microscales and (35)S radiolabelled tissue sections. We also present the first results of (3)H images produced under direct irradiation of a CCD sensor operating at room temperature. Compared to film, silicon-based imaging technologies exhibit enhanced sensitivity, dynamic range and linearity.

Images of the kidneys using dynamic contrast-enhanced magnetic resonance renography (DCE-MRR) contains unwanted complex organ motion due to respiration. This gives rise to motion artefacts that hinder the clinical assessment of kidney function. However, due to the rapid change in contrast agent within the DCE-MR image sequence, commonly used intensity-based image registration techniques are likely to fail. While semi-automated approaches involving human experts are a possible alternative, they pose significant drawbacks including inter-observer variability, and the bottleneck introduced through manual inspection of the multiplicity of images produced during a DCE-MRR study. To address this issue, we present a novel automated, registration-free movement correction approach based on windowed and reconstruction variants of dynamic mode decomposition (WR-DMD). Our proposed method is validated on ten different healthy volunteers’ kidney DCE-MRI data sets. The results, using block-matching-block evaluation on the image sequence produced by WR-DMD, show the elimination of 99%99% of mean motion magnitude when compared to the original data sets, thereby demonstrating the viability of automatic movement correction using WR-DMD.

This work investigates the detection performance of specialist and non-specialist observers for different targets in 2D-mammography and digital breast tomosynthesis (DBT) using the OPTIMAM virtual clinical trials (VCT) Toolbox and a 4-alternative forced choice (4AFC) assessment paradigm. Using 2D-mammography and DBT images of virtual breast phantoms, we compare the detection limits of simple uniform spherical targets and irregular solid masses. Target diameters of 4 mm and 6 mm have been chosen to represent target sizes close to the minimum detectable size found in breast screening, across a range of controlled contrast levels. The images were viewed by a set of specialist observers (five medical physicists and six experienced clinical readers) and five non-specialists. Combined results from both observer groups indicate that DBT has a significantly lower detectable threshold contrast than 2D-mammography for small masses (4 mm: 2.1% [DBT] versus 6.9% [2D]; 6 mm: 0.7% [DBT] versus 3.9% [2D]) and spheres (4 mm: 2.9% [DBT] versus 5.3% [2D]; 6 mm: 0.3% [DBT] versus 2.2% [2D]) (p  ˂  0.0001). Both observer groups found spheres significantly easier to detect than irregular solid masses for both sizes and modalities (p  ˂  0.0001) (except 4 mm DBT). The detection performances of specialist and non-specialist observers were generally found to be comparable, where each group marginally outperformed the other in particular detection tasks. Within the specialist group, the clinical readers performed better than the medical physicists with irregular masses (p  ˂  0.0001). The results indicate that using spherical targets in such studies may produce over-optimistic detection thresholds compared to more complex masses, and that the superiority of DBT for detecting masses over 2D-mammography has been quantified. The results also suggest specialist observers may be supplemented by non-specialist observers (with training) in some types of 4AFC studies.

Background: Chiari-like malformation (CM) is a complex malformation of the skull and cranial cervical vertebrae potentially resulting in pain and secondary syringomyelia (SM). CM associated pain can be challenging to diagnose [35]. We propose a machine learning approach to characterize morphological changes in dogs that may/may not be apparent to human observers. This data driven approach can remove potential bias (or blindness) that may be produced by a hypothesis driven expert observer approach. Hypothesis/Objectives: Using a novel machine learning approach to understand neuromorphological change and to identify image-based biomarkers in dogs with CM associated pain (CM-P) and symptomatic SM (SM-S), with the aim of deepening the understanding on these disorders. Animals: 32 client owned Cavalier King Charles Spaniels (CKCS) (11 controls, 10 CM-P, 11 SM-S) Methods: Retrospective study using T2W midsagittal DICOM anonymized images which were mapped to a images of a average clinically normal CKCS reference using Demons image registration. Key deformation features were automatically selected from the resulting deformation maps. A kernelized Support Vector Machine was used for classifying characteristic localized changes in morphology. Results: Candidate biomarkers were identified with receiver operating characteristic (ROC) curves with area under the curve (AUC) of 0.78 (sensitivity = 82%; specificity = 69%) for the CM-P biomarkers collectively, and an AUC of 0.82 (sensitivity = 93%; specificity = 67%) for the SM biomarkers collectively. Conclusions and clinical importance: Machine learning techniques can assist CM/SM diagnosis and understand abnormal morphology location with the potential to be applied to a variety of breeds and conformational diseases.

Voluntary inspiration breath hold (VIBH) for left breast cancer patients has been shown to be a safe and effective method of reducing radiation dose to the heart. Currently, VIBH protocol compliance is monitored visually. In this work, we establish whether it is possible to gate the delivery of radiation from an Elekta linac using the Microsoft Kinect version 2 (Kinect v2) depth sensor to measure a patient breathing signal. This would allow contactless monitoring during VMAT treatment, as an alternative to equipment–assisted methods such as active breathing control (ABC). Breathing traces were acquired from six left breast radiotherapy patients during VIBH. We developed a gating interface to an Elekta linac, using the depth signal from a Kinect v2 to control radiation delivery to a programmable motion platform following patient breathing patterns. Radiation dose to a moving phantom with gating was verified using point dose measurements and a Delta4 verification phantom. 60 breathing traces were obtained with an acquisition success rate of 100%. Point dose measurements for gated deliveries to a moving phantom agreed to within 0.5% of ungated delivery to a static phantom using both a conventional and VMAT treatment plan. Dose measurements with the verification phantom showed that there was a median dose difference of better than 0.5% and a mean (3% 3 mm) gamma index of 92.6% for gated deliveries when using static phantom data as a reference. It is possible to use a Kinect v2 device to monitor voluntary breath hold protocol compliance in a cohort of left breast radiotherapy patients. Furthermore, it is possible to use the signal from a Kinect v2 to gate an Elekta linac to deliver radiation only during the peak inhale VIBH phase.

C autoradiography is a well established technique for structural and metabolic analysis of cells and tissues. The most common detection medium for this application is film emulsion, which offers unbeatable spatial resolution due to its fine granularity but at the same time has some limiting drawbacks such as poor linearity and rapid saturation. In recent years several digital detectors have been developed, following the technological transition from analog to digital-based detection systems in the medical and biological field. Even so such digital systems have been greatly limited by the size of their active area (a few square centimeters), which have made them unsuitable for routine use in many biological applications where sample areas are typically ∼ 10-100 cm. The Multidimensional Integrated Intelligent Imaging (MI3-Plus) consortium has recently developed a new large area CMOS Active Pixel Sensor (12.8 cm × 13.1 cm). This detector, based on the use of two different pixel resolutions, is capable of providing simultaneously low noise and high dynamic range on a wafer scale. In this paper we will demonstrate the suitability of this detector for routine beta autoradiography in a comparative approach with widely used film emulsion. © 2012 IOP Publishing Ltd and Sissa Medialab srl.

In recent times, the security focus for civil aviation has shifted from hijacking in the 1980s, towards deliberate sabotage. X-ray imaging provides a major tool in checked baggage inspection, with various sensitive techniques being brought to bear in determining the form, and density of items within luggage as well as other material dependent parameters. This review first examines the various challenges to X-ray technology in securing a safe system of passenger transportation. An overview is then presented of the various conventional and less conventional approaches that are available to the airline industry, leading to developments in state-of-the-art imaging technology supported by enhanced machine and observer-based decision making principles. © 2012 Elsevier Ltd.

Industrial regulation to protect privacy, intellectual property and proprietary information often restricts data sharing ─ an important prerequisite for developing services in the digital economy. Social media data is publicly available for data mining but requires intensive cleaning and harmonisation before analysis. This paper reveals the process of semantic sensing to convert social network tweets into meaningful insights. Our research question is: how to realise semantic sensing for data innovation? We use design science research to develop an artefact-ontology that collects tweets by pet owners talking about their itchy pet into knowledge graphs, including symptoms, location, breed, timestamp and potential cause and converts them into a thematic map of the regional occurrence of symptoms and potential treatment needs, providing vital information for data innovation. The semantic engine can predict potential causes of itching from the tweet, so a Chatbot may contact the pet owner, inviting them to a veterinary screening. Animal health and pharma companies can use this information to position their services. Our theoretical contribution is a process of semantic sensing, which is a vital part of dynamic capability. Although limited to animal health, the results could be transferred to other contexts.

In this study, a novel sleep pose identification method has been proposed for classifying 12 different sleep postures using a two-step deep learning process. For this purpose, transfer learning as an initial stage retrains a well-known CNN network (VGG-19) to categorise the data into four main pose classes, namely: supine, left, right, and prone. According to the decision made by VGG-19, subsets of the image data are next passed to one of four dedicated sub-class CNNs. As a result, the pose estimation label is further refined from one of four sleep pose labels to one of 12 sleep pose labels. 10 participants contributed for recording infrared (IR) images of 12 pre-defined sleep positions. Participants were covered by a blanket to occlude the original pose and present a more realistic sleep situation. Finally, we have compared our results with (1) the traditional CNN learning from scratch and (2) retrained VGG-19 network in one stage. The average accuracy increased from 74.5% & 78.1% to 85.6% compared with (1) & (2) respectively.

Sleep quality is an important determinant of human health and wellbeing. Novel technologies that can quantify sleep quality at scale are required to enable the diagnosis and epidemiology of poor sleep. One important indicator of sleep quality is body posture. In this paper, we present the design and implementation of a non-contact sleep monitoring system that analyses body posture and movement. Supervised machine learning strategies applied to noncontact vision-based infrared camera data using a transfer learning approach, successfully quantified sleep poses of participants covered by a blanket. This represents the first occasion that such a machine learning approach has been used to successfully detect four predefined poses and the empty bed state during 8-10 hour overnight sleep episodes representing a realistic domestic sleep situation. The methodology was evaluated against manually scored sleep poses and poses estimated using clinical polysomnography measurement technology. In a cohort of 12 healthy participants, we find that a ResNet-152 pre-trained network achieved the best performance compared with the standard de novo CNN network and other pre-trained networks. The performance of our approach was better than other video-based methods for sleep pose estimation and produced higher performance compared to the clinical standard for pose estimation using a polysomnography position sensor. It can be concluded that infrared video capture coupled with deep learning AI can be successfully used to quantify sleep poses as well as the transitions between poses in realistic nocturnal conditions, and that this non-contact approach provides superior pose estimation compared to currently accepted clinical methods.

The open data market size is estimated at €184 billion and forecast to reach between €199.51 and €334.21 billion in 2025. In this paper, we conceptualise the semantic data innovation platform, which will be able to answer inter-disciplinary questions via semantic reasoning over open data. We use 750 open animal healthcare datasets to exemplify this work, covering mainly poultry, swine, ruminants, and other livestock, which are complemented by open data from complementary domains, such as geographic location, medicine and virology. We aggregate the domain knowledge (classes) and enable the logical links (properties) between these classes. The prototype encapsulates the complexity of animal healthcare knowledge into ontology, which can answer complex questions using semantic reasoning on the datasets (answer-as-a-service).

BACKGROUND Social media are seldom explored in animal health despite the potential for insights into pet owners’ perceptions and information seeking behaviours before and after accessing veterinary care [1]. A study in Feline Pruritus was conducted using social listening to investigate owners’ perceptions of feline allergic skin disease using a thematic analysis technique. OBJECTIVES • To apply thematic analysis to social listening (SL) data and thereby create a unique dataset concerning pet owner perceptions of feline pruritus and online information-seeking behaviours. METHODS • Fifty dynamic (frequently updated) content sources applicable to cats and feline pruritus were chosen, keywords were defined by a veterinary expert panel and organised into topics. • Keywords were augmented by reference to academic literature, a baseline survey of 1000 cat owners in the United States, the addition of synonyms and further iterations using Google Trends analytics keywords and sources. • Six bespoke topic filters were developed: body areas, behaviours, symptoms, disease diagnosis, solutions and treatments. • Content from the selected sources was collected using a social intelligence solution developed by ATC, tagged using both keywords (with stemming) and topic filters. • The data was aggregated, duplicates removed, and sentiment calculated by algorithm. • Content matching topic(s) in the body areas, behaviours and symptoms filters were reviewed manually, relevancy criteria developed, and posts marked relevant if: posted by a pet owner, identifying an itchy cat and not duplicated e.g. previous versions of a post, similar posts or cross posting to different sources. • A sub-set of 493 posts (title and text only) marked relevant and published between 2009 and 2022 were used for reflexive thematic analysis in NVIVO (Burlington, MA) to extract the key themes. RESULTS Qualitative thematic analysis was conducted on 493 relevant posts collected up to 30th May 2022 producing five top level themes: allergy, pruritus, additional behaviours, unusual or undesirable behaviours, diagnosis and treatment. The analytical method used the most recent ‘reflexive thematic analysis’ approach developed by Braun and Clarke [2] and adapted from [3]. The newly developed reflexive thematic analysis approach is not bound to one specific theoretical framework but allows for the flexibility to return to a previous phase, as the analysis develops, guiding the research based on the researcher’s level of interpretation and design of the study. The data was published between 2009 and 2022, met the body areas, behaviours and symptoms topic filters, met the relevancy criteria, had been manually reviewed and marked relevant for feline pruritus. Internet forums and Twitter were the most likely sources of relevant posts: Reddit (198/493), Catsite (110/493), Twitter (97/493) and Quora (59/493). Relevant posts were most frequently from the United States (188/493), United Kingdom (12/493), Canada (9/493), Greece (7/493), Australia (3/493) and Italy (2/493). A single post came from each of 11 countries and 260/493 posts had no location. The total number of responses coded was 493; the total number of themes was 5, total codes was 47 and the total number of references coded was 880. CONCLUSIONS • SL provides unique insights into verbatim owner perceptions on health and veterinary care. • This study shows there is a need for an increased awareness by veterinarians to pet owner frustrations with treatment options to tackle feline pruritus. • The data analysis could be scaled up using machine learning for topic modelling. • The data could enable data-driven decisions such as assessing demand for veterinary services by location and impact on quality of life. • These findings will be validated by comparison with thematic analysis of a direct pet owner survey.

The continual improvement in spatial resolution of Nuclear Medicine (NM) scanners has made accurate compensation of patient motion increasingly important. A major source of corrupting motion in NM acquisition is due to respiration. Therefore a particle filter (PF) approach has been proposed as a powerful method for motion correction in NM. The probabilistic view of the system in the PF is seen as an advantage that considers the complexity and uncertainties in estimating respiratory motion. Previous tests using XCAT has shown the possibility of estimating unseen organ configuration using training data that only consist of a single respiratory cycle. This paper augments application specific adaptation methods that have been implemented for better PF estimates with an iterative model update step. Results show that errors are further reduced to an extent up to a small number of iterations and such improvements will be advantageous for the PF to cope with more realistic and complex applications. © 2011 Copyright Society of Photo-Optical Instrumentation Engineers (SPIE).

Autoradiography is a well established imaging modality in Biology and Medicine. This aims to measure the location and concentration of labelled molecules within thin tissue sections. The brain is the most anatomically complex organ and identification of neuroanatomical structures is still a challenge particularly when small animals are used for pre-clinical trials. High spatial resolution and high sensitivity are therefore necessary. This work shows the performance and ability of a prototype commercial system, based on a Charged-Couple Device (CCD), to accurately obtain detailed functional information in brain Autoradiography. The sample is placed in contact with the detector enabling direct detection of β- particles in silicon, and the system is run in a range of quasi-room temperatures (17-22 °C) under stable conditions by using a precision temperature controller. Direct detection of β- particles with low energy down to ~5 keV from 3[H] is possible using this room temperature approach. The CCD used in this work is an E2V CCD47-20 frame-transfer device which removes the image smear arising in conventional full-frame imaging devices. The temporal stability of the system has been analyzed by exposing a set of 14[C] calibrated microscales for different periods of time, and measuring the stability of the resultant sensitivity and background noise. The thermal performance of the system has also been analyzed in order to demonstrate its capability of working in other life science applications, where higher working temperatures are required. Once the performance of the system was studied, a set of experiments with biological samples, labelled with typical β- radioisotopes, such as 3[H], has been carried out to demonstrate its application in life sciences.

We propose a method for Partial Volume correction and intensity recovery that models blood vessels as small cylinders of known diameter. We use a Bayesian classifier that explicitly models the effects of the point spread function on these cylinders. Although the method requires prior knowledge of the cylinder/arterial width, there is no requirement for any registration. A further advantage is that Region Of Interest (ROI) definition can be limited to only a few axial slices, thus minimizing time averaging. Furthermore, ROI selection requires only approximate placement around the target artery, encompassing both artery and background tissue, so that recovered data values are not operator-dependent. We present results for classifier performance on simulated phantom data of hot cylindrical inserts in a warm background with different contrast to noise ratios. © 2006 IEEE.

This paper present preliminary work in developing a method of using a marker-less tracking system to analyze the natural temporal variations in chest wall configuration during breathing, thus avoiding reliance on a limited number of fiducial markers. This involves using a marker-less video capture of the motion of the abdominal-chest surface and the development of a B-spline model to parameterize this motion. The advantage of the marker-less system that is non-invasive and non-ionizing, thus facilitating high throughput without the need for marker-based patient set-up time.

The use of conventional clinical trials to optimise technology and techniques in breast cancer screening carries with it issues of dose, high cost and delay. This has motivated the development of Virtual Clinical Trials (VCTs) as an alternative in-silico assessment paradigm. However, such an approach requires a set of modelling tools that can realistically represent the key biological and technical components within the imaging chain. The OPTIMAM image simulation toolbox provides a complete validated end-to-end solution for VCTs, wherein commonly-found regular and irregular lesions can be successfully and realistically simulated. As spiculated lesions are the second most common form of solid mass we report on our latest developments to produce realistic spiculated lesion models, with particular application in Alternative Forced Choice trials. We make use of sets of spicules drawn using manually annotated landmarks and interpolated by a fitted 3D spline for each spicule. Once combined with a solid core, these are inserted into 2D and tomosynthesis image segments and blended using a combination of elongation, rotational alignment with background, spicule twisting and core radial contraction effects. A mixture of real and simulated images (86 2D and 86 DBT images) with spiculated lesions were presented to an experienced radiologist in an observer study. The latest observer study results demonstrated that 88.4% of simulated images of lesions in 2D and 67.4% of simulated lesions in DBT were rated as definitely or probably real on a six-point scale. This presents a significant improvement on our previous work which did not employ any background blending algorithms to simulate spiculated lesions in clinical images.

A wide variety of digital mammography systems are available for breast cancer imaging, each varying in physical performance. However, the relationship between physical performance assessment and clinical outcome is not clear. Thus, a means of simulating technically and clinically realistic images from different systems would represent a first step towards elucidating the impact of physical performance on clinical outcome. To this end, a framework for simulating technically realistic images has been developed. A range of simulated test objects, including CDMAM have been used to determine whether the simulation chain correctly reproduces these objects thus validating the simulation framework. Results evaluated for two digital mammography systems have been promising, with simulated images proving similar to experimental images for Modulation Transfer Function and Normalised Noise Power Spectrum measurements differing by approximately 3%.

With the continual improvement in spatial resolution of Nuclear Medicine (NM) scanners, it has become increasingly important to accurately compensate for patient motion during acquisition. Respiratory motion produced by lung ventilation is a major source of artefacts in NM that can affect large parts of the abdominal-thoracic cavity. As such, a particle filter (PF) is proposed as a powerful method for motion correction in NM imaging. This paper explores a basic PF approach and demonstrates that it is possible to estimate non-stationary motion using a single respiratory cycle as training data. Using the XCAT phantom, 7 test datasets that vary in depth and rate of respiration were generated. The results using these datasets show that the PF has an average Euclidean distance error over all voxels of only 1.7 mm, about half of the typical dimensions of an NM voxel for clinical applications. The conclusion is that use of the PF is promising, and can be adapted to handle more sophisticated data such as those that arise in clinical situations.

In external beam radiotherapy, patient misalignment during set-up and motion during treatment may result in lost dose to target tissue and increased dose to normal tissues, reducing therapeutic benefit. The most common method for initial patient setup uses room mounted lasers and surface marks on the skin. We propose to use the Microsoft Kinect which can capture a complete patient skin surface representing a multiplicity of 3D points in a fast reproducible, marker-less manner. Our first experiments quantitatively assess the technical performance of Kinect technology using a planar test object and a precision motion platform to compare the performance of Kinect for Xbox and Kinect for Windows. Further experiments were undertaken to investigate the likely performance of using the Kinect during treatment to detect respiratory motion, both in supine and prone positions. The Windows version of the Kinect produces superior performance of less than 2mm mean error at 80-100 cm distance. © 2013 IEEE.

The aim of this study was to compare the detection of microcalcification clusters by human observers in breast images using 2D-mammography and narrow (15°/15 projections) and wide (50°/25 projections) angle digital breast tomosynthesis (DBT). Simulated microcalcification clusters with a range of microcalcification diameters (125 μm-275 μm) were inserted into 6 cm thick simulated compressed breasts. Breast images were produced with and without inserted microcalcification clusters using a set of image modelling tools, which were developed to represent clinical imaging by mammography and tomosynthesis. Commercially available software was used for image processing and image reconstruction. The images were then used in a series of 4-alternative forced choice (4AFC) human observer experiments conducted for signal detection with the microcalcification clusters as targets. The minimum detectable calcification diameter was found for each imaging modality: (i) 2D-mammography: 164±5 μm (ii) narrow angle DBT: 210±5 μm, (iii) wide angle DBT: 255±4 μm. A statistically significant difference was found between the minimum detectable calcification diameters that can be detected by the three imaging modalities. Furthermore, it was found that there was not a statistically significant difference between the results of the five observers that participated in this study. In conclusion, this study presents a method that quantifies the threshold diameter required for microcalcification detection, using high resolution, realistic images with observers, for the comparison of DBT geometries with 2D-mammography. 2Dmammography can visualise smaller detail diameter than both DBT imaging modalities and narrow-angle DBT can visualise a smaller detail diameter than wide-angle DBT.

Virtual clinical trials are a promising new approach increasingly used for the evaluation and comparison of breast imaging modalities. A key component in such an assessment paradigm is the use of simulated pathology, in particular, simulation of lesions. Breast mass lesions can be generally classified into two categories based on their appearance; nonspiculated masses and spiculated masses. In our previous work, we have successfully simulated non-spiculated masses using a fractal growth process known as diffusion limited aggregation. In this new work, we have extended the DLA model to simulate spiculated lesions by using features extracted from patient DBT images containing spiculated lesions. The features extracted included spicule length, width, curvature and distribution. This information was used to simulate realistic looking spicules which were attached to the surface of a DLA mass to produce a spiculated mass. A batch of simulated spiculated masses was inserted into normal patient images and presented to an experienced radiologist for review. The study yielded promising results with the radiologist rating 60% of simulated lesions in 2D and 50% of simulated lesions in DBT as realistic.

Virtual clinical trials are an emergent approach for the rapid evaluation and comparison of various breast imaging technologies and techniques using computer-based modeling tools. A fundamental requirement of this approach for mammography is the use of realistic looking breast anatomy in the studies to produce clinically relevant results. In this work, a biologically inspired approach has been used to simulate realistic synthetic breast phantom blocks for use in virtual clinical trials. A variety of high and low frequency features (including Cooper’s ligaments, blood vessels and glandular tissue) have been extracted from clinical digital breast tomosynthesis images and used to simulate synthetic breast blocks. The appearance of the phantom blocks was validated by presenting a selection of simulated 2D and DBT images interleaved with real images to a team of experienced readers for rating using an ROC paradigm. The average areas under the curve for 2D and DBT images were 0.53±.04 and 0.55±.07 respectively; errors are the standard errors of the mean. The values indicate that the observers had difficulty in differentiating the real images from simulated images. The statistical properties of simulated images of the phantom blocks were evaluated by means of power spectrum analysis. The power spectrum curves for real and simulated images closely match and overlap indicating good agreement.

Virtual clinical trials (VCTs) are increasingly being seen as a viable pre-clinical method for evaluation of imaging systems in breast cancer screening. The CR-UK funded OPTIMAM project is aimed at producing modelling tools for use in such VCTs. In the initial phase of the project, modelling tools were produced to simulate 2D-mammography and digital breast tomosynthesis (DBT) imaging systems. This paper elaborates on the new tools that have recently been developed for the current phase of the OPTIMAM project. These new additions to the framework include tools for simulating synthetic breast tissue, spiculated masses and variable-angle DBT systems. These tools are described in the paper along with the preliminary validation results. Four-alternative forced choice (4-AFC) type studies deploying these new tools are underway. The results of the ongoing 4AFC studies investigating minimum detectable contrast/size of masses/microcalcifications for different modalities and system designs are presented.

Although X-ray mammography is the gold standard technique for breast cancer detection, it suffers from limitations due to tissue superposition which could either obscure or mimic a breast lesion. Dedicated breast computed-tomography (BrCT) represents an alternative technology with the potential to overcome these limitations. However, this technology is still under investigation in order to study and improve certain parameters (e.g. dose, scattered radiation, etc.). In this work, an image simulation framework is proposed to generate realistic BrCT images and spectral imaging analysis is explored to enhance the contrast of breast lesions. Results illustrated an improvement in contrast between 5 and 10% when the final image is reconstructed using X-ray photons with energies between 21 and 30 keV, in comparison with the reconstructed image from the polychromatic energy spectrum recorded within the image receptor. © 2013 IEEE.

A method to convert digital mammograms acquired on one system to appear as if acquired using another system is presented. This method could be used to compare the clinical efficacy of different systems. The signal transfer properties modulation transfer function (MTF) and noise power spectra (NPS) were measured for two detectors - a computed radiography (CR) system and a digital radiography (DR) system. The contributions to the NPS from electronic, quantum and structure sources were calculated by fitting a polynomial at each spatial frequency across the NPS at each dose. The conversion process blurs the original image with the ratio of the MTFs in frequency space. Noise with the correct magnitude and spatial frequency was added to account for differences in the detector response and dose. The method was tested on images of a CDMAM test object acquired on the two systems at two dose levels. The highest dose images were converted to lower dose images for the same detector, then images from the DR system were converted to appear as if acquired at a similar dose using CR. Contrast detail curves using simulated CDMAM images closely matched those of real images.

Autoradiography is a widely extended pre-clinical nuclear imaging modality used in life sciences to investigate and localise radiolabelled biological pathways in thin ex-vivo tissue sections. After the tissue section has been exposed to an ionising radiation detector the resulting labelled regions are subsequently analysed. Typically, the resulting autoradiograms are analysed manually by an expert life scientists using a visual template as reference to measure the different radioligand uptake levels in the different areas of, in our case, mouse brain. This process is extremely time consuming and error prone, with the expertise of the life scientist playing a significant role. In this paper we describe a semi-automatic method to register a template brain atlas on to the brain autoradiogram making the analysis process more efficient, repeatable and independent of the expertise of the life scientist. The method first identifies those regions with high and low level of radioligand uptake by region growing segmentation. Subsequently, the counterpart regions in the corresponding atlas image are manually identified. Finally a set of control points is extracted from each region contour in the autoradiogram and the atlas image to apply a scattered data interpolator. ©2009 IEEE.

In nuclear medicine, there is a significant research focus in developing a new approach in monitoring, tracking and compensating respiratory motion during image acquisition. We address this by attempting to model the respiratory cycle pattern and finding a method that describes the configuration of the anterior surface which then correlates with the internal position/configuration of the internal organ as a foundation for motion compensation. This paper presents novel work in parameterizing external respiratory motion using a method based on the variation of abdominal vs. thoracic surface markers to investigate inter- and intra-subject variation. The dominant mode of variation of the Abdominal and Thoracic surfaces during respiration using Principle Component Analysis (PCA) is studied. This demonstrates that pattern of TS vs AS motion appears temporally at a global level stable. Thus although breathing style is consistent within a given subject, we there observe temporal changes in the amplitude and density of the PDF in intra-subject data.

Segmentation in medical imaging plays a critical role easing the delineation of key anatomical functional structures in all the imaging modalities. However, many segmentation approaches are optimized with the assumption of high contrast, and then fail when segmenting poor contrast to noise objects. The number of approaches published in the literature falls dramatically when functional imaging is the aim. In this paper a feature extraction based approach, based on region growing, is presented as a segmentation technique suitable for poor quality (low Contrast to Noise Ratio CNR) images, as often found in functional images derived from Autoradiography. The region growing combines some modifications from the typical region growing method, to make the algorithm more robust and more reliable. Finally the algorithm is validated using synthetic images and biological imagery.

Traditional Autoradiography is an imaging modality used in life sciences where thin ex-vivo tissue sections are placed in direct contact with autoradiographic film. High resolution autoradiograms can be obtained using low energy radioisotopes, such as H where an intrinsic 0.1-1 μm spatial resolution can be achieved due to limited β- path length. Several digital alternatives have been presented in recent years to replace conventional film as the imaging medium, but the spatial resolution of film remains unmatched. Although silicon-based imaging technologies have demonstrated higher sensitivity compared to conventional film, the main issue that remains is spatial resolution. We address this here with an investigation into the design parameters that impact on spatial resolution when imaging uncollimated β-found in Autoradiography. The study considers Monte Carlo simulation of the energy deposition process, the charge diffusion process in silicon and the detector noise, and this is applied to a range of radioisotope β energies typically used in Autoradiography. Finally an optimal detector geometry to obtain the best possible spatial resolution for a specific technology and a specific radioisotope is suggested. ©2009 IEEE.

One of the current major challenges in clinical imaging is modeling and prediction of respiratory motion, for example, in nuclear medicine or external-beam radio therapy. This paper presents preliminary work in developing a method for modeling and predicting the temporal behavior of the anterior surface position during respiration. This is achieved by tracking the anterior surface during respiration and projecting the captured motion sequence data into a lower dimensional space using Principle Component Analysis and extracting the variation in the Abdominal surface and Thoracic surface separately. Modeling is based on learning the multivariate probability distribution of the motion sequence using a joint Probability Distribution Function (PDF) between the variation of the Thoracic surface and Abdomen surface in the Eigen space. Moreover, the prediction model encodes the amplitude of the variation in the Eigen space for both Thoracic surface and Abdominal surface and the derivative of the variation which reflects the motion path (velocity). The joint Probability Distribution Function (PDF) of the prediction model covers the likelihood of each position/phase configuration and the associated maximum-likelihood motion path. Moreover, feeding the real-time tracking data into the model during nuclear medicine acquisition or external-beam radio therapy will facilitate adjusting the model for any changes and overcome irregularities in the observed respiration cycle.

About 1.7 million new cases of breast cancer were estimated by the World Health Organization (WHO) in 2012, accounting for 23 percent of all female cancers. In the UK 33 percent of women aged 50 and above were diagnosed in the same year, thus positioning the UK as the 6th highest in breast cancer amongst the European countries. The national Screening programme in the UK has been focused on the procedure of early detection and to improve prognosis by timely intervention to extend the life span of patients. To this end, the National Health Service Breast Screening Programme (NHSBSP) employs 2-D planar mammography because it is considered to be the gold standard technique for early breast cancer detection worldwide. Breast tomosynthesis has shown great promise as an alternative method for removing the intrinsic overlying clutter seen in conventional 2D imaging. However, preliminary work in breast CT has provided a number of compelling aspects that motivates the work featured in this thesis. These advantages include removal of the need to mechanically compress the breast which is a source of screening non-attendances, and that it provides unique cross sectional images that removes almost all the overlying clutter seen in 2D. This renders lesions more visible and hence aids in early detection of malignancy. However work in Breast CT to date has been focused on using scaled down versions of standard clinical CT systems. By contrast, this thesis proposes using a photon counting approach. The work of this thesis focuses on investigating photoncounting detector technology and comparing it to conventional CT in terms of contrast visualization. Results presented from simulation work developed in this thesis has demonstrated the ability of photoncounting detector technology to utilize data in polychromatic beam where contrast are seen to decrease with increasing photon energy and compared to the conventional CT approach which is the standard clinical CT system.

Virtual clinical trials (VCTs) represent an alternative assessment paradigm that overcomes issues of dose, high cost and delay encountered in conventional clinical trials for breast cancer screening. However, to fully utilize the potential benefits of VCTs requires a machine-based observer that can rapidly and realistically process large numbers of experimental conditions. To address this, a Deep Learning Model Observer (DLMO) was developed and trained to identify lesion targets from normal tissue in small (200 x 200 pixel) image segments, as used in Alternative Forced Choice (AFC) studies. The proposed network consists of 5 convolutional layers with 2x2 kernels and ReLU (Rectified Linear Unit) activations, followed by max pooling with size equal to the size of the final feature maps and three dense layers. The class outputs weights from the final fully connected dense layer are used to consider sets of n images in an n-AFC paradigm to determine the image most likely to contain a target. To examine the DLMO performance on clinical data, a training set of 2814 normal and 2814 biopsy-confirmed malignant mass targets were used. This produced a sensitivity of 0.90 and a specificity of 0.92 when presented with a test data set of 800 previously unseen clinical images. To examine the DLMOs minimum detectable contrast, a second dataset of 630 simulated backgrounds and 630 images with simulated lesion and spherical targets (4mm and 6mm diameter), produced contrast thresholds equivalent to/better than human observer performance for spherical targets, and comparable (12 % difference) for lesion targets.

The continual improvement in spatial resolution of Nuclear Medicine (NM) scanners has made accurate compensation of patient motion increasingly important. A major source of corrupting motion in NM acquisition is due to respiration. Therefore a particle filter (PF) approach has been proposed as a powerful method for motion correction in NM. The probabilistic view of the system in the PF has an advantage in that it considers the complexity and uncertainties of respiratory motion. Tests using the XCAT phantom have previously shown the possibility of estimating unseen organ configurations using training data that only consist of a single respiratory cycle. This paper builds upon previous work in two ways: (i) this is the first evaluation of a PF framework using clinical 4D thoracic CT data; and, (ii) this implementation uses a kernel density estimation (KDE) representation for the transition model, thus taking advantage of the PF's ability to use a wider range of stochastic models. The results show some improvement with the use of a KDE-based transition model and indicates that the PF should be applicable to clinical data. © 2011 IEEE.

CMOS imaging technology can be applied to Chemiluminescence Western Blotting as a potential imaging alternative technology to using conventional film-emulsion. In this work the authors present a through investigation on the performance of CMOS Active Pixel Sensors for using in western blotting. Chemiluminescence labeling is a well established technique to detect proteins and presents several advantages compared with the fluorescence labeling. In fact it requires neither external illumination nor filtering optics and does not produce an inherently label-related background to correct. In this paper the first results of imaging a secondary antibody labeled with chemiluminescence reagents obtained with a CMOS sensor operating at room temperature are presented

Wafer scale detector technology represents an alternative approach for biomedical imaging where currently Flat Panel Imagers (FPIs) are the most common option. However, FPIs possess several key drawbacks such as large pixels, high noise, low frame rates, and excessive image artefacts. Recently Active Pixel Sensors have gained popularity overcoming such issues and are now scalable up to wafer size by appropriate reticule stitching. Detectors for biomedical imaging applications require high spatial resolution, low noise and high dynamic range. These figures of merit are related to pixel size and as the pixel size is fixed at the time of the design, spatial resolution, noise and dynamic range cannot be further optimized. The authors report on a new rad-hard monolithic APS, named DynAMITe (Dynamic range Adjustable for Medical Imaging Technology), developed by the UK MI-3 Plus consortium. This large area detector (12.8 x 12.8 cm2) is based on the use of two different diode geometries within the same pixel array with different size pixels (50 um and100 um). Hence the resulting camera can possess two inherently different resolutions each with different noise and saturation performance. The small pixels and the large pixels can be reset at different voltages, resulting in different depletion widths. The larger depletion width for the small pixels allows the initial generated photo-charge to be collected by the small pixels, which ensures an intrinsically lower noise and higher spatial resolution. After these pixels reach near saturation, the larger pixels start collecting so offering a higher dynamic range whereas the higher noise floor is not important as at higher signal levels performance is set by Poisson noise. Further different reset voltage can selectively choose the operating resolution of the detector leading to a true pixel binning.

Due to the increasing amount of data available from medical imaging procedures and also the increase in computing power, there has been a rise in the automation of the analysis of such data. A crucial step in the automation of such procedures is accurate segmentation of anatomy. Popular approaches include model based segmentation. However, these approaches require an atlas which may not be generic enough. This paper proposes a semi-automated data-driven segmentation framework of thoracic CT scans. The preliminary results of the framework is presented and discussed with proposals for future work. © 2013 IEEE.

This paper describes a method of using a tracking system to track the upper part of the anterior surface during scanning for developing patient-specific models of respiration. In the experimental analysis, the natural variation in the anterior surface during breathing will be modeled to reveal the dominant pattern in the breathing cycle. The main target is to produce a patient-specific set of parameters that describes the configuration of the anterior surface for all respiration phases. These data then will be linked to internal organ motion to identify the effect of the morphology of each on motion using particle filter to account for previously unseen patterns of motion. In this initial study, a set of volunteers were imaged using the Codamotion infrared marker-based system. In the marker-based system, the temporal variation of the respiratory motion was studied. This showed that for the 12 volunteer cohort, the mean displacement of the thorax surface TS (abdomen surface AS) region is 10.7±5.6 mm (16.0±9.5mm). Finally, PCA was shown to capture the redundancy in the data set with the first principal component (PC) accounting for more than 96% of the overall variance in both AS and TS datasets. A fitting to the dominant modes of variation using a simple piecewise sinusoid has suggested a maximum error of about 1.1mm across the complete cohort dataset.

Compensation of respiratory motion has been identified as a crucial factor in achieving high resolution Nuclear Medicine (NM) imaging. Many motion correction approaches have been studied and they are seen to have advantages over simpler motion compensation approaches such as respiratory gating. However, all motion correction approaches rely on an assumption or estimation of respiratory motion. This paper builds upon previous work in recursive Bayesian estimation (RBE) of respiratory motion assuming stereo camera observation of external respiratory motion. In this paper, additional stereo-camera derived XCAT cycles are used to evaluate the robustness of RBE with inter-cycle variation. © 2013 IEEE.

In many biomedical imaging applications there is a strong demand for large area sensors. Nowadays the most common detectors in this field are Flat Panel imagers which offer a reasonably large area, typically greater than 20 cm×20 cm. Even so such detectors present severe drawbacks such as large pixels, high noise, low frame rate and excessive image artefacts. In the last two decades Active Pixel Sensors (APSs) have gained popularity because of a potential for overcoming such issues. Furthermore, in recent years, improvements in design and fabrication techniques have made available fabricative processes for wafer scale imagers, which can be now seamlessly scaled from a few centimetres square up to the whole wafer size. A suitable detector for biomedical imaging application needs to fulfil specific requirements: it should have a high spatial resolution, a low noise and a high dynamic range. These figures of merit are connected with the pixel size. Since the pixel size is normally fixed at the time of the design, spatial resolution, noise and dynamic range cannot be further optimized. The authors propose a novel edge-buttable wafer scale APS (12.8 cm×12.8 cm), named the Dynamic range Adjustable for Medical Imaging Technology or DynAMITe, developed by the Multidimensional Integrated Intelligent Imaging Plus (MI-3 Plus) consortium. This APS is based on the use of two different diode geometries in the same pixel array and with different size active pixels. As the effective pixel size is no longer fixed, but two different pixel sizes are used for the whole detector matrix, this detector can deliver two inherently different resolutions each with different noise and saturation performance in the same pixel array. The DynAMITe design has great potential for use in a variety of biomedical imaging applications. In its initial deployment the authors will be developing demonstrators in radiotherapy portal imaging, breast mammography and diffraction imaging and also in sequencing methods for the life sciences.

Western blotting electrophoretic sequencing is an analytical technique widely used in Functional Proteomics to detect, recognize and quantify specific labelled proteins in biological samples. A commonly used label for western blotting is Enhanced ChemiLuminescence (ECL) reagents based on fluorescent light emission of Luminol at 425nm. Film emulsion is the conventional detection medium, but is characterized by non-linear response and limited dynamic range. Several western blotting digital imaging systems have being developed, mainly based on the use of cooled Charge Coupled Devices (CCDs) and single avalanche diodes that address these issues. Even so these systems present key drawbacks, such as a low frame rate and require operation at low temperature. Direct optical detection using Complementary Metal Oxide Semiconductor (CMOS) Active Pixel Sensors (APS)could represent a suitable digital alternative for this application. In this paper the authors demonstrate the viability of direct chemiluminescent light detection in western blotting electrophoresis using a CMOS APS at room temperature. Furthermore, in recent years, improvements in fabrication techniques have made available reliable processes for very large imagers, which can be now scaled up to wafer size, allowing direct contact imaging of full size western blotting samples. We propose using a novel wafer scale APS (12.8 cm×13.2 cm), with an array architecture using two different pixel geometries that can deliver an inherently low noise and high dynamic range image at the same time representing a dramatic improvement with respect to the current western blotting imaging systems.