Semi-Supervised Segmentation via Embedding Matching

TL;DR

This paper introduces a semi-supervised segmentation approach that combines uncertainty assessment and embedding matching to effectively utilize unlabeled medical images, reducing the need for extensive annotations.

Contribution

The novel method integrates uncertainty-based pseudo-labeling with embedding matching, improving segmentation accuracy with minimal labeled data in medical imaging.

Findings

Achieved HD95 of 3.30 and IoU of 0.929 with only 4 CT scans.

Outperformed existing methods in segmentation accuracy.

Demonstrated effectiveness in hip bone segmentation in CT images.

Abstract

Deep convolutional neural networks are widely used in medical image segmentation but require many labeled images for training. Annotating three-dimensional medical images is a time-consuming and costly process. To overcome this limitation, we propose a novel semi-supervised segmentation method that leverages mostly unlabeled images and a small set of labeled images in training. Our approach involves assessing prediction uncertainty to identify reliable predictions on unlabeled voxels from the teacher model. These voxels serve as pseudo-labels for training the student model. In voxels where the teacher model produces unreliable predictions, pseudo-labeling is carried out based on voxel-wise embedding correspondence using reference voxels from labeled images. We applied this method to automate hip bone segmentation in CT images, achieving notable results with just 4 CT scans. The proposed approach yielded a Hausdorff distance with 95th percentile (HD95) of 3.30 and IoU of 0.929, surpassing existing methods achieving HD95 (4.07) and IoU (0.927) at their best.