Data-Driven Band Optimization and Frequency-Aware Modeling in Medical Hyperspectral Image Segmentation.

Area of Science:

• Medical Imaging
• Computational Pathology
• Biomedical Engineering

Background:

• Hyperspectral imaging offers rich spectral-spatial data for molecular tissue characterization.
• Current segmentation methods struggle with suboptimal band selection and limited frequency modeling, hindering clinical utility.

Purpose of the Study:

• To develop a unified framework, AMBS-SF2Net, for enhanced spectral representation and hierarchical frequency modeling in hyperspectral image segmentation.
• To improve the accuracy and efficiency of molecular tissue characterization using hyperspectral imaging.

Main Methods:

• Proposed AMBS-SF2Net framework with three key modules: Adaptive Mask-based Band Selection (AMBS), Adaptive Spectral-Frequency Integration (ASFI), and Multi-Axis Frequency Enhanced (MAFE).
• AMBS dynamically selects informative spectral channels.
• ASFI fuses multi-scale spatial edges with frequency-aware spectral features.
• MAFE captures complementary spectral and spatial frequency patterns.

Main Results:

• AMBS-SF2Net demonstrated superior performance compared to state-of-the-art methods across diverse datasets (cholangiocarcinoma, pig organs, human placenta).
• The framework exhibits significant robustness across varying spectral resolutions and spatial modalities.
• Validated generalizability on microscopic and macroscopic tissue imaging scales.

Conclusions:

• AMBS-SF2Net effectively addresses limitations in existing hyperspectral segmentation techniques.
• The proposed method shows strong potential for diverse clinical applications requiring accurate molecular tissue characterization.
• Enhanced spectral representation and hierarchical frequency modeling are crucial for advancing hyperspectral imaging utility.