Rhodri Davis - CMIC/WEISS joint seminar series

Rhodri Davies, PhD, MRCP
Senior clinical data scientist and honorary consultant cardiologist, Barts Heart Centre and UCL

Title - Improving cardiac MR image analysisAbstract:

Many critical decisions in clinical cardiology depend on biomarkers derived from images. Cardiac MR is established as the best way of measuring cardiac structure and function, but its analysis currently involves manually delineating cardiac anatomy on 25-30 images per patient. Not only is this onerous and time-consuming (>15 mins/patient) but human analysis also introduces error, which can lead to the wrong clinical decision with profound consequences.

We have used convolutional neural networks to fully automate the segmentation process. Our focus was on building a system that improves the performance of a clinical expert. I will describe how we achieved this and how we measure performance in the absence of a truth standard. I will talk about the challenges that we have overcome and the many that we still face.

Link to the Moodle page here: https://eu.bbcollab.com/guest/1F8A77AEFA6374A7DA87FA45940F516B