MSc Thesis: Temporal Contrastive Learning of Cardiac Heartbeat

Description

Contrastive learning is currently the most effective way to learn representations in a self-supervised manner. Contrastive learning is strong because it uses multiple views of an object to learn which information to capture/encode. If the views are smartly chosen through proper augmentations, then information which is useful for the targeted downstream task is learned and the resulting representations are strong.

Contrastive learning has been explored extensively in the natural image domain but there still remain many unanswered questions and untapped potential in the medical domain. The unique nature of medical data offers many avenues to explore to try and determine how best to adapt contrastive learning to medical imaging.

One such area is cardiac MR imaging, where instead of having a static image of an object that we wish to encode, we have an entire time series spanning at least one heart beat. The goal of this thesis is to explore how to best select views from a time-series of cardiac MR data to maximize the information learned through contrastive learning. You will use full cycle cardiac MR data from the UKBB which contains over 45k subjects.

To accomplish this work successfully, we expect you to have:

• Strong coding skills and familiarity with Pytorch
• Basic knowledge of contrastive learning
• A strong spirit of independent work and desire to solve interesting research questions

What we offer

• An exciting research project with many possibilities to bring in your own ideas.
• Close supervision and access to state-of-the-art computer hardware.
• The chance to work in a team of experts in image processing, deep learning, biomedical engineering and medicine.
• Support in bringing your finished project to publication

References

[1] A Simple Framework for Contrastive Learning of Visual Representations

[2] What Makes for Good Views for Contrastive Learning?

[3] Spatiotemporal Contrastive Video Representation Learning

How to apply:

Send an email to paul.hager@tum.de with your CV and transcript. We promise to get back to you within a couple of days.