The Multi-Centre, Multi-Vendor & Multi-Disease Cardiac Image Segmentation Challenge 2020 (M&Ms) is a deep learning challenge run by MICCAI aimed at producing robust and accurate segmentation models that can be consistently used across data from different countries, clinics, and scanners. As my final year project for this year, I decided to challenge myself to become an ‘expert’ (sort of) at deep learning and to implement an accurate and working segmentation model using the data from the M&Ms challenge.

In medical image analysis, there is often a lack of data, especially labelled, given the time and effort demanded in manually labelling images. Whilst medical segmentation models have shown great accuracy with a large dataset from the same domain, with limited data from different domains, these models have shown lesser results. If we are able to generate consistent and accurate models with robust properties, translation into clinical practice may result in a greater physiological understanding of disease, diagnosis and treatment.