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

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 student model that combines knowledge distillation cross teaching, consistency regularization based pseudo-labels, efficient semi-supervised learning. The proposed method is demonstrated the abdominal multi-organ segmentation task in CT images under MICCAI FLARE 2022 challenge, mean Dice scores 0.8429 0.8520 validation test sets, respectively. code available at https://github.com/jwc-rad/MISLight .