Medical image acquisition
Active pixel sensors
We are partners in the biomedical imaging group within the multi-dimensional intelligent imaging + consortium (MI3+), a three-year extension of the original MI3 consortium. This project, aimed at developing intelligent imaging sensors for supporting the UK science base in the 21st century, is underwritten by £4.2M funding from the Research Council's basic technology programme. We work particularly closely with our medical imaging partners at University College of London (radiation physics group) and the Institute of Cancer Research/Royal Marsden NHS Trust Hospital (radiotherapy physics research team). This follows on from prior EPSRC funded work developing CCD imaging methods for autoradiography.
The work undertaken in our group using CCD and CMOS technology has been applied to digital autoradiography, an imaging modality widely used in biosciences to detect radiolabelled tracers in molecular pathways of thin tissue sections. Low energy (tritium) and medium energy radioisotopes (suplhur-35, carbon-14, iodine-125) are typically used in our laboratory to label specific neuroreceptors. Our group has published the first brain sections labelled with tritium acquired at room temperature, exposed to a CCD detector and to a CMOS detector.
Our group has established links with industry by collaborating with XCAM, a company the provides custom solutions with commercial CCDs for scientific applications, as consultants to obtain tritiated autoradiography samples using back-thinned commercial sensors under controlled temperature conditions.
Monte Carlo simulations using Geant4 have been developed to study different parameters of digital detectors and their effect on spatial resolution and sensitivity. These studies will potentially allow to design an optimum digital sensor for a specific radioisotope to maximize spatial resolution or sensitivity. An exemplar Monte Carlo simulation of 100 events emitted from a C14 point source on the surface of the detector is shown below, where the red tracks represent electron trajectories and the blue blobs represent energy deposited in the silicon.
These Monte Carlo simulations have been used to analyse the impact of some of the most important key parameters in digital detectors, such as pixel size and epitaxial layer thickness (active volume of the detector), on the spatial resolution and the amount of signal deposited by a specific radioisotope.
Collaborators: UCL, ICR, University of Surrey School of Biological and Biomedical Sciences and the MI3 consortium.
Coded apertures for scintimammography
Many groups are pursuing the development of dedicated gamma cameras for radioisotope imaging of the breast (scintimammography) as a complimentary imaging method to X-ray mammography. This is because it is difficult to gain close proximity to the breast with a standard clinical gamma camera. Our approach is that by offsetting the gamma camera, but compensating for loss of solid angle coverage by removing the collimator would allow efficient use of existing imaging systems. In order to produce efficient images we are currently pursuing the use of Uniformly Redundant Arrays (URAs) as coded apertures for this application. URAs have a number of useful imaging properties, including almost zero side lobes in their point response function as well as much greater sensitivity than using a collimator. Monte Carlo simulation has demonstrated an ability to visualise sub-cm lesions in breast equivalent material even in the presence of strong cardiac uptake. We are currently working on developing new imaging methodology to demonstrate this approach in realistic phantom studies.
Collaborator: Royal Surrey County Hospital NHS Trust.
Adaptive optimal thresholding in nuclear medicine
Radioisotope imaging methods such as PET, SPECT and planar scintigraphy, currently utilize a fixed energy (pulse height) acceptance window regardless of the volume of the subject being imaged. This is despite previous work suggesting that higher window settings may yield improved image quality for imaging larger objects. However, we speculate that non-standard energy windows have not been widely adopted because there has not, until now, been a method available for determining how and when to use such an approach. We are addressing this issue and propose a method for setting an adaptive photopeak acceptance window using a Bayes' Minimum Error Thresholding approach, which utilizes modelling the upper part of the observed energy spectrum as a two-class Gaussian mixture model.
Collaborator: Royal Surrey County Hospital NHS Trust.
Bias field correction
Analysis of serial MRI brain images is extremely important for the diagnosis of dementia, for example Alzheimer's disease. In order to obtain accurate diagnosis, it is essential that the serial scans are comparable, so that the amount of atrophy that has occurred between the two scans can be quantified.
Bias artifact is commonly seen in MRI scans, and is a smooth and slow unwanted, artificial change in intensities across the image. It confounds attempts to segment and quantify images; in particular it degrades longitudinal measurements, such as measurement of atrophy rates over time in patients with dementia. Previous work has shown excellent results correcting longitudinal bias using simple median filters; however, results are still poorer when there is atrophy present between the longitudinal scans. This project aims to build on this work by using structure and space-frequency filters to separate the bias component from the atrophy component.
Collaborator: Dementia Research Centre, UCL.