Medical imaging is routinely used in clinical practice for evaluating and diagnosing disease. Typically, particularly with MRI, several different looking images are obtained (known as modalities), since each one of them portrays different information about the underlying anatomy and pathology. As a result, the amount of available medical imaging data collected is increasing, with several experts now agreeing that we can use machine learning to create artificial intelligence systems that can help the clinicians in their work. To train these systems we require significant amount of annotated images and in the context of many modalities, annotations for the same subject (patient) in all available modalities (a complete set).
However, this is not what is currently practiced by clinicians, since typically they ascertain the presence and extent of disease taking into account all of this information in a qualitative fashion. Then when quantitative measurements are necessary for diagnosis, they usually manually annotate (for example outlining the anatomy and pathology) a few out of all those images. While they primarily do this to save time, it really complicates the training of classical machine learning algorithms because we lack annotated datasets that are complete.
This project will investigate and develop machine learning algorithms that can deal with the above problem. Specifically, we will base our solutions drawing inspiration from areas of machine learning such as weakly supervised, semi-supervised, multiple instance, and transfer learning. These areas all come into play to aggregate information across the variety of imaging data and annotations available, since both qualitative (e.g., presence of disease) and quantitative information (e.g., annotations of a pathology) can be available but not for all datasets. Furthermore, since images of different modalities are available, to represent this information in a manner suitable for the machine learning algorithm, proper features have to be extracted. Thus, this project will also emphasize on how to best learn features relying on techniques stemming from the field of representation learning.
To achieve this, we are looking for an enthusiastic and strongly motivated student to join our group. He/she will have the opportunity to collaborate with our partners (in Edinburgh and USA) and participate in exciting projects particularly in cardiac imaging were medical image computing helps us understand physiology and provide solutions that aid diagnosis.
This position is fully funded for 42 months (3.5 years) and is open to all students with a preference for UK/EU nationals.
Candidates should have a Master’s level education (or in exceptional cases an excellent (upper second class and higher) Bachelor’s degree) in electronic/electrical engineering, computer science (informatics), physics or closely related subjects. Any prior experience in medical image analysis is desirable but not necessary. Prior exposure to machine learning is desirable but not necessary. The candidate will be expected to have a high level of analytical and investigative skills. The candidate should have good programming skills (e.g., Matlab) and a good mathematical background.
Enquiries can be made to Dr Tsaftaris at [email protected]
Please note that the position remains open till January 2017 but we are looking for applicants who can start in the Fall of 2016. Thus, please apply immediately since reviewing of application is ongoing and offers will be made as soon as a suitable candidate is found.