Interactive Medical Image Segmentation with Self-Adaptive Confidence Calibration

• DOI: 10.48550/arxiv.2111.07716
• Publication Date: 2021-01-01
• Authors: Wenhao Li; Qisen Xu; Chuyun Shen; Bin Hu; Fengping Zhu; Yuxin Li; Bo Jin; Xiangfeng Wang
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Machine Learning (cs.LG)

Medical image segmentation is one of the fundamental problems for artificial intelligence-based clinical decision systems. Current automatic medical methods are often failed to meet requirements. As such, a series interactive algorithms proposed utilize expert correction information. However, existing suffer from some refining failure after long-term interactions and cost annotation, which hinder applications. This paper proposes an framework, called MEdical with self-adaptive Confidence CAlibration (MECCA), by introducing corrective action evaluation, combines action-based confidence learning multi-agent reinforcement (MARL). The evaluation established through novel network, actions obtained MARL. Based on confidential information, reward function designed provide more detailed feedback, simulated label generation mechanism unsupervised data reduce over-reliance labeled data. Experimental results various datasets have shown significant performance algorithm.