Supervisors:
Dr Mauro Manassi - University of Aberdeen, School of Psychology - [Email Address Removed]
Dr. Gerald Lip - Honorary Lecturer at University of Aberdeen - NHS Grampian, North East of Scotland Breast Screening Centre - [Email Address Removed].uk
Cancer diagnosis in medical images is crucial for the health of millions of people, but it is still far from perfect. For example, within mammography, false negative and false positive rates have been reported to be 0.15% and 9%, respectively (Nelson et al., 2016). One critical reason for misdiagnoses is the misinterpretation of radiographs by clinicians.
When looking at radiographs, clinicians are typically asked to localize lesions (if present), and then to classify them by judging their size, class, and so on. Importantly, radiographs can contain various kinds of structures, from actual masses, to microcalcifications, architectural distortions, parenchymal deformities, and other morphologies. It is thus crucial for radiologists to correctly recognize lesions in cluttered radiographs, discarding non-relevant masses. In everyday life, psychophysical research showed that our percept in cluttered environments is biased by contextual effects. For example, visual perception is limited by ‘crowding’, which is
the deleterious influence of nearby objects on object recognition (Manassi et al. 2018). Crowding is a major bottleneck for object recognition, impeding reading and visually guided actions. Given its importance and ubiquity, we will investigate the impact of crowding in cancer screening.
We will first test how tumor recognition in radiologists and untrained observers is affected by crowded radiographs. Second, we will characterize what kind of masses are mostly impacted by crowding. Third, we will investigate how AI algorithms in breast cancer screening perform in crowding situations. Finally, we will compare recognition performance in untrained observers, radiologists, and AI algorithms (de Vries et al. 2021).
The outcome of the research will advance our knowledge regarding factors underlying the interpretation of medical images. Importantly, it will provide a more integrative understanding of the visual search mechanisms of radiologists and artificial intelligence systems used for medical screening, highlighting critical advantages and disadvantages. The long-term goal is to improve cancer image screening by improving our understanding of medical image perception and AI algorithms.
The PhD candidate will learn (1) psychophysical techniques, (2) breast cancer imaging techniques and (3) will acquire knowledge regarding AI algorithms implemented for medical image screening. This project will involve developing a high level of expertise in digital signal processing and programming (Matlab, Python, R), skills highly useful in both industry and academia. The project is suitable for candidates with a background in psychology, neuroscience, radiology, computer science or engineering who have interests in objects recognition or medical imaging.
It may be possible to undertake this project part-time, in discussion with the lead supervisor, however, please note that part-time study is unavailable to students who require a Student Visa to study within the UK.
Application Procedure:
Please visit this page for full application information: http://www.eastscotbiodtp.ac.uk/how-apply-0
Please send your completed EASTBIO application form, along with academic transcripts to Alison Innes at [Email Address Removed]
Two references should be provided by the deadline using the EASTBIO reference form.
Please advise your referees to return the reference form to [Email Address Removed]
Unfortunately, due to workload constraints, we cannot consider incomplete applications