Self-Supervised Alignment Learning for Medical Image Segmentation

TL;DR

This paper introduces a novel self-supervised alignment learning framework that leverages spatial correspondence in 3D medical images to improve segmentation, using local and global alignment losses to enhance feature learning.

Contribution

The paper proposes a new self-supervised alignment learning framework with local and global losses, exploiting spatial correspondence in 3D medical images for better pre-training.

Findings

Competitive performance on CT and MRI datasets

Effective in limited annotation settings

Outperforms existing SSL methods

Abstract

Recently, self-supervised learning (SSL) methods have been used in pre-training the segmentation models for 2D and 3D medical images. Most of these methods are based on reconstruction, contrastive learning and consistency regularization. However, the spatial correspondence of 2D slices from a 3D medical image has not been fully exploited. In this paper, we propose a novel self-supervised alignment learning framework to pre-train the neural network for medical image segmentation. The proposed framework consists of a new local alignment loss and a global positional loss. We observe that in the same 3D scan, two close 2D slices usually contain similar anatomic structures. Thus, the local alignment loss is proposed to make the pixel-level features of matched structures close to each other. Experimental results show that the proposed alignment learning is competitive with existing self-supervised pre-training approaches on CT and MRI datasets, under the setting of limited annotations.