Deep learning for Medical Images with limited labels

Project Motivation

Supervised deep learning methods require a large amount of labelled data to achieve human-level performance. Getting labels for medical image datasets is an arduous and time-consuming task generally requiring medical expertise. Therefore, our motivation is to design data-efficient deep learning methods which leverage large unlabelled dataset and provides state-of-the-art performance with limited labelled data.  

Project Description

The objective is to bridge the gap between fully supervised methods developed using large, labelled datasets and data-efficient self-supervised learning methods developed using large unlabelled datasets.

Status: Ongoing

Team Members

Partners