Learnable color space conversion and fusion for stain normalization in pathology images

Jing Ke, Yijin Zhou, Yiqing Shen, Yi Guo, Ning Liu, Xiaodan Han, Dinggang Shen

Abstract: Highlights•Introduces an automatic color-space variable optimization, eliminating manual adjustments and enhancing adaptability.•Employs a self-attention mechanism to optimally combine features from multiple color spaces, improving image interpretation accuracy.•Implements a dynamic template generation strategy from the source image set, ensuring flexibility and effectiveness across various analysis tasks.•Achieves substantial computational efficiency and faster inference by directly linking normalization to diagnostic networks without intermediate steps.