M-Net: Integrating Spectral Features and Physical Field Operators into Deep Learning for Medical Image Segmentation

The M-Net (Math-Augmented Network) is a novel deep learning framework that enhances U-Net for medical image segmentation by incorporating explicit mathematical inductive biases. This method utilizes continuous spectral features based on the condition number of centered local pixel matrices, alongside physical field operators, including divergence and a discrete curl-like boundary irregularity operator derived from image gradient fields. Additionally, a Math-Attention Gate (MAG) adaptively merges these mathematical features. Published on arXiv (2608.12196), the research explores the potential of matrix spectral analysis and vector calculus operators to improve segmentation beyond traditional data-driven approaches. This work is part of broader initiatives to integrate mathematical principles into deep learning for medical imaging and is classified as a cross-type announcement, suggesting it may have been showcased in various forums.