Experts' cognition-driven safe noisy labels learning for precise segmentation of residual tumor in breast cancer

2023-04-13

arXiv.org

AI Summary

To address the low segmentation accuracy of post-neoadjuvant chemotherapy residual tumor in breast cancer (PSRTBC) caused by morphological heterogeneity and severe annotation noise, this paper proposes a **safe weakly supervised learning framework** that integrates pathological expertise with deep learning. For the first time, expert prior knowledge regarding tumor morphology is explicitly encoded as a **safe noisy label prior**, which is embedded into UNet-based architectures to enable trustworthy collaborative modeling under noisy annotations. The method achieves both robustness and interpretability. Quantitatively, it improves the recall lower bound by 2.42% and the foreground Intersection-over-Union (fIoU) lower bound by 4.1%, while also enhancing mean and upper-bound performance. This advancement provides reliable AI support for precise clinical assessment of treatment response.

Abstract

Precise segmentation of residual tumor in breast cancer (PSRTBC) after neoadjuvant chemotherapy is a fundamental key technique in the treatment process of breast cancer. However, achieving PSRTBC is still a challenge, since the breast cancer tissue and tumor cells commonly have complex and varied morphological changes after neoadjuvant chemotherapy, which inevitably increases the difficulty to produce a predictive model that has good generalization with usual supervised learning (SL). To alleviate this situation, in this paper, we propose an experts' cognition-driven safe noisy labels learning (ECDSNLL) approach. In the concept of safe noisy labels learning, which is a typical type of safe weakly supervised learning, ECDSNLL is constructed by integrating the pathology experts' cognition about identifying residual tumor in breast cancer and the artificial intelligence experts' cognition about data modeling with provided data basis. Experimental results show that, compared with usual SL, ECDSNLL can significantly improve the lower bound of a number of UNet variants with 2.42% and 4.1% respectively in recall and fIoU for PSRTBC, while being able to achieve improvements in mean value and upper bound as well.

Problem

• Breast Cancer
• Neoadjuvant Chemotherapy
• Tumor Recognition

Innovation

• ECDSNLL
• Expert Cognitive Guidance
• Label Noise Handling