Knowledge Distillation from Cross Teaching Teachers for Efficient Semi-Supervised Abdominal Organ Segmentation in CT

TL;DR

This paper introduces a semi-supervised learning framework using cross-teaching and knowledge distillation with teacher-student models to improve abdominal organ segmentation in CT images, reducing labeled data and computational needs.

Contribution

It presents a novel coarse-to-fine semi-supervised framework combining cross teaching and knowledge distillation for medical image segmentation.

Findings

Achieved mean Dice scores of 0.8429 (validation) and 0.8520 (test).

Demonstrated effectiveness on MICCAI FLARE 2022 challenge dataset.

Improved segmentation accuracy with semi-supervised approach.

Abstract

For more clinical applications of deep learning models for medical image segmentation, high demands on labeled data and computational resources must be addressed. This study proposes a coarse-to-fine framework with two teacher models and a student model that combines knowledge distillation and cross teaching, a consistency regularization based on pseudo-labels, for efficient semi-supervised learning. The proposed method is demonstrated on the abdominal multi-organ segmentation task in CT images under the MICCAI FLARE 2022 challenge, with mean Dice scores of 0.8429 and 0.8520 in the validation and test sets, respectively.