Aberdeen Biomedical Imaging Centre
Imaging Processing
Image Processing in Oncology
There are two broad research interests.
- Development of functional imaging as a monitoring tool for early indication of cancer response to therapy.
- Development of functional imaging biomarkers as indicators of cancer response potential pre-therapy.
Most of our publication history involves the development and application of pharmacokinetic multi-compartment models for the analysis of cancers, using breast, rectal and cervical cancer patients. Current aim is continued development of a multiparameter model of tumour response potential combining permeability, oxygenation, metabolism, perfusion, apoptosis (diffusion weighted imaging). Future aims include the development of spectroscopy in cancer outside of the head.
Automated detection of features of referable diabetic retinopathy in digital photographs
Screening programmes for diabetic retinopathy based on digital retinal photography are being set up across the UK with the aim of detecting the disease in people with diabetes before it causes blindness. Software has been developed at Aberdeen University for detection of abnormal images originating from a diabetic retinal screening programme and has been shown to be suitable for national implementation. Software has been developed to detect retinal haemorrhages using a standard image processing scheme involving, detection of lesion candidates, feature evaluation of each candidate and candidate classification into lesion and non-lesion. Feature evaluation includes the ability to distinguish haemorrhages from microaneurysms which are a feature of non-referrable retinopathy. An evaluation of the software using 200 images, 100 of which contained haemorrhages, has shown that detection of images containing haemorrhages is possible to 80% sensitivity and 72% specificity. Detection of haemorrhages has been achieved; further work is underway to detect the other features of referrable retinopathy including exudates and new vessel growth.
Neuroimage processing
Image processing in neurology centres around two related methodologies: structural and functional MRI. Structural MRI is the assessment of local variations in gray or white matter volume with pathology not visible to trained radiologists. Functional MRI is the assessment of the neuronal activity during a cognitive task. There are a number of packages available to aid in the interpretation of these large data sets. One on the most popular is Statistical Parametric Mapping written in matlab. This package includes functions to realign, co-register, normalise and segment 3D data sets. Although mainly used for analysis of the human brain it can be used with almost any 3D data set including animal brains and SPECT cardiac exams for example.
Results of two contrasts rendered onto a single, semitransparent, standard 3D structural MRI image. Activation related to related to joint attention vs rest is in red; that due to nonjoint attention vs rest is in green. Overlap is in yellow. Threshold at P ≤ 0.001 uncorrected uncorrected. (Williams et al: NeuroImage, 2005:25;133-140)
MR Cardiac perfusion
It has been possible for many years to detect regional variation in the blood flow to the myocardium using MRI contrast methods. It is also true that these methods have not become a mainstream diagnostic technique. The accuracy and ease of application of novel display and analysis methods which would provide significant added value to the clinical interpretation of dynamic gadolinium perfusion imaging is being undertaken. The results are being validated against new, state of the art, but intrusive cardiac catheterisation methods.
 |
 |
| patlak anlysis 1 | fermi anlysis 1 |
Lossy compression of medical images
The rapid proliferation of digital medical imaging, coupled with the relentless increase in resolution and frame-rates, is stretching storage media to their limits. Even where storage density is keeping pace with demand, data transfer to and from the media is not. While lossy compression is essential for digital photography and the internet, the requirements for medical imaging are more stringent; it's not sufficient just to "look right", it must "be right" -- no diagnostically important information should be sacrificed in the compression. We are currently investigating how we can measure lossy image degradation and how we may determine if a compression scheme is safe.
Image analysis of bone shape and structure
Bone and other musculoskeletal tissues are hierarchical structures and so present imaging challenges at every scale. Whether examining individual cells, bone microstructure, the shape of bones or even investigating how two or more bones fit together at a joint, image processing and analysis can help us understand how bone and other musculoskeletal tissues work. This is work in collaboration with the Musculoskeletal Research Programme.