Multi-Task, Multi-Domain Deep Segmentation with Shared Representations and Contrastive Regularization for Sparse Pediatric Datasets

TL;DR

This paper introduces a multi-task, multi-domain deep learning model with contrastive regularization for pediatric MR image segmentation, effectively addressing data scarcity and improving generalization across different anatomical regions.

Contribution

It proposes a novel shared representation framework with domain-specific components and contrastive regularization, enhancing segmentation accuracy on sparse pediatric datasets.

Findings

Outperforms state-of-the-art methods on ankle and shoulder datasets

Improves generalization through contrastive regularization

Effectively handles multi-domain pediatric MRI segmentation

Abstract

Automatic segmentation of magnetic resonance (MR) images is crucial for morphological evaluation of the pediatric musculoskeletal system in clinical practice. However, the accuracy and generalization performance of individual segmentation models are limited due to the restricted amount of annotated pediatric data. Hence, we propose to train a segmentation model on multiple datasets, arising from different parts of the anatomy, in a multi-task and multi-domain learning framework. This approach allows to overcome the inherent scarcity of pediatric data while benefiting from a more robust shared representation. The proposed segmentation network comprises shared convolutional filters, domain-specific batch normalization parameters that compute the respective dataset statistics and a domain-specific segmentation layer. Furthermore, a supervised contrastive regularization is integrated to further improve generalization capabilities, by promoting intra-domain similarity and impose inter-domain margins in embedded space. We evaluate our contributions on two pediatric imaging datasets of the ankle and shoulder joints for bone segmentation. Results demonstrate that the proposed model outperforms state-of-the-art approaches.