Medical image segmentation method based on confidence learning and collaborative training

Confidence learning plays an important role in the training of low-quality annotated data of medical images. However, the current applications of confidence learning are based on the average teacher model, and the possibility of confidence learning on other networks has not been explored. To address this problem, a segmentation model based on confidence learning and co-training is proposed. The model uses two different networks to encourage the output results of the two networks to be consistent. Then, the output of one of the networks is compared with the original low-quality annotations using confidence learning, and the low-quality annotated data is corrected to provide an effective training reference. The proposed model is compared on three medical image datasets of different modalities. Experimental results show that the segmentation effect of the proposed model is better than that of the existing confidence learning model.