Few-shot brain segmentation from weakly labeled data with deep heteroscedastic multi-task networks

TL;DR

This paper demonstrates that few weakly labeled brain MRI volumes, combined with heteroscedastic multi-task networks, can achieve high segmentation accuracy, reducing the need for extensive manual annotations.

Contribution

It introduces a novel heteroscedastic network that directly learns label uncertainty, improving brain segmentation performance with minimal training data.

Findings

Achieved a mean Dice coefficient of 0.84 with only 9 weakly labeled examples.

Heteroscedastic networks improved segmentation accuracy over existing methods.

The proposed method significantly outperformed prior approaches across most brain structures.

Abstract

In applications of supervised learning applied to medical image segmentation, the need for large amounts of labeled data typically goes unquestioned. In particular, in the case of brain anatomy segmentation, hundreds or thousands of weakly-labeled volumes are often used as training data. In this paper, we first observe that for many brain structures, a small number of training examples, (n=9), weakly labeled using Freesurfer 6.0, plus simple data augmentation, suffice as training data to achieve high performance, achieving an overall mean Dice coefficient of $0.84 \pm 0.12$ compared to Freesurfer over 28 brain structures in T1-weighted images of $\approx 4000$ 9-10 year-olds from the Adolescent Brain Cognitive Development study. We then examine two varieties of heteroscedastic network as a method for improving classification results. An existing proposal by Kendall and Gal, which uses Monte-Carlo inference to learn to predict the variance of each prediction, yields an overall mean Dice of $0.85 \pm 0.14$ and showed statistically significant improvements over 25 brain structures. Meanwhile a novel heteroscedastic network which directly learns the probability that an example has been mislabeled yielded an overall mean Dice of $0.87 \pm 0.11$ and showed statistically significant improvements over all but one of the brain structures considered. The loss function associated to this network can be interpreted as performing a form of learned label smoothing, where labels are only smoothed where they are judged to be uncertain.