Noisy-to-Clean Label Learning for Medical Image Segmentation
Zihao Bu, Xiaoxiao Wang, Chengjian Qiu, Zhixuan Wang, Kai Han, Xiuhong Shan, Zhe Liu · 2023
In the field of medical image processing, accurate segmentation is of great importance to assist doctors in diagnosis. However, existing machine learning methods are hardly effective for medical image segmentation in the absence of large and accurate datasets. Existing methods of learning with noisy labels rarely try to explore the correlation between noisy and clean labels. We found that some error corrections are learnable in the process of noisy labels corrected by medical experts. In this work, we propose a novel method to improve the performance of medical image segmentation. The method consists of two main networks: segmentation network segments the image and label correction network records and learns the denoising process of noisy labels, denoises the noisy labels. In addition, we introduce a feature fusion branch between the two networks. We compare with several state-of-the-art methods which learning with noisy label on the gastric wall dataset and notice that our method has strong competitiveness.