The paper by authors Yu Dian Zhang, Chen Hao Xu, and others from the class of 2021 undergraduates at the College of Information Science and Technology, under the guidance of Professor Hai Jiang Zhu, has been accepted by the International Conference on Pattern Recognition (ICPR-2024), a B-class conference of the Chinese Association for Artificial Intelligence (CAAI).
Paper information: "Yudian Zhang, Chenhao Xu, Kaiye Xu, Haijiang Zhu. Mask-TS Net: Mask Temperature Scaling Uncertainty Calibration for Polyp Segmentation. International Conference on Pattern Recognition (ICPR-2024)". In the field of uncertainty analysis, existing methods focus more on classification problems and less on semantic segmentation. This paper introduces an uncertainty analysis method in the context of medical image segmentation, targeting the issue of small proportion of polyp lesion areas in medical images. It proposes a four-branch calibration network based on Mask-Loss and Mask-TS strategies. The core of this network architecture is its ability to not only analyze image information globally but also to focus on potential lesion areas through specific branches, achieving fine-tuning of the prediction probability for key areas. This strategy enhances the model's ability to capture lesion details, effectively reduces interference from background noise, and ensures the accuracy and reliability of uncertainty quantification results.
Figure 1. Structure of Mask-TS Network
ICPR (International Conference on Pattern Recognition) is one of the renowned conferences in the field of computer vision and pattern recognition internationally, hosted by the International Association for Pattern Recognition (IAPR). Since its inception in 1973, the conference has been held every two years and is classified as a B-class conference of the Chinese Association for Artificial Intelligence (CAAI-B).