I work on using deep learning to extract clinically useful information from medical images. I am jointly supervised by Irina Voiculescu in Computer Science and Sion Glyn-Jones in NDORMS, a department I am also affiliated with. My current project trains on a dataset of hip x-rays but I am interested in obtaining datasets containing other image modalities and body parts.
My specific areas of research are:
- Anatomical landmark detection from medical images, which is the process of automatically spotting points which are inherently clinically useful, in an image. For example, we can diagnose a hip impingement if we can spot the furthest point laterally of the acetabulum socket and see that it extends too far over the femur head.
- Repurposing pre-existing landmark detection networks that were created for the human pose detection challenge, such as stacked hourglass and high-resolution deep network, to the problem of finding anatomical landmarks.
- How problems in medicine can be encoded into a method which uses anatomical landmarks. For example, how do we spot a bump in the femur head by detecting point landmarks? We would like to do this so we can use landmark detection algorithms which have shown good results.
- Creating new measures for doctors to assess a patient. For example, the current alpha-angle measure of the hip does not capture small bumps in the femur head accurately, can we produce a new measure which does? And can it be calculated by computer automatically
Contour−Hugging Heatmaps for Landmark Detection
James McCouat and Irina Voiculescu
Conference on Computer Vision and Pattern Recognition (CVPR). June, 2022.
Automatically Diagnosing Hip Conditions from X−rays using Landmark Detection
James McCouat‚ Sion Glyn−Jones and Irina Voiculescu
International Symposium on Biomedical Imaging (ISBI). April, 2021.
Vertebrae Detection and Localization in CT with Two−Stage CNNs and Dense Annotations
James McCouat and Ben Glocker