Knowledge Distillation from Cross Teaching Teachers for Efficient Semi-Supervised Abdominal Organ Segmentation in CT
This addresses the need for more efficient semi-supervised segmentation in medical imaging, particularly for abdominal organs in CT, with incremental improvements in performance.
The study tackled the problem of high demands on labeled data and computational resources for deep learning models in medical image segmentation by proposing a coarse-to-fine framework with knowledge distillation and cross teaching for efficient semi-supervised learning, achieving mean Dice scores of 0.8429 and 0.8520 on validation and test sets for abdominal organ segmentation in CT images.
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.