Liam Burrows

I am a postdoctoral research associate at the University of Bath. My main research interests lie in medical image segmentation and image registration using novel deep learning methods as well as more mathematical, variational methods. I am particularly interested in the intersection of these two areas.

My current work is on the development of an AI screening tool for pulmonary hypertension (PH), working in collaboration with the Royal United Hospitals Bath NHS Trust. Funding is provided by the AI award from the National Institute for Health and Care Research (NIHR)

The screening tool for PH involves segmentation of various cardiovascular anatomies, including: the pulmonary arterial tree, ascending aorta, and chambers of the heart. The segmented anatomy is then analysed using tools from vmtk, in which various morphometric features are extracted from fine vessel analysis. In addition, statistical shape modelling is done to extract anatomical shape modes describing variation in the patient population, many of which are found to correlate well with clinical variables. Finally, combining morphometrics and shape modes, we use regression and classification machine learning models to predict clinical variables and phenotype disease.

I received a masters in Mathematics 2017 from the University of Liverpool, and subsequently completed a PhD in 2022 on topics of medical image segmentation by way of variational and deep learning methods. Since I have worked as a postdoctoral research associate at the University of Liverpool and University of Bath.

During my experience in academia I have collaborated with various NHS trusts and industry partners on a wide variety of projects, including:

• Segmenting abdominal aortic aneurysms (CT) and tracking progression.
• Segmentation of brain tumours (MR) (meningioma, glioma).
• Segmentation of lung tumours in mouse models (MR).
• Segmenting stromal compartment and stromal stain biomarkers for colorectal cancer digital pathology images (TMAs).
• Segmenting pulmonary arteries and ascending aorta, and applying statistical shape methods to inform diagnosis of pulmonary hypertension.