Boundary-Aware Spectral and Morphological Guidance Method for Feature-Driven Colorectal Cancer Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Accurate medical image segmentation is vital for clinical decisions.
• Deep learning methods struggle with complex lesion variability, limiting generalization.
• Challenges include difficult data acquisition and diverse morphological features.

Purpose of the Study:

• To develop an advanced feature-driven segmentation model for improved medical image analysis.
• To overcome limitations of traditional deep learning methods in segmenting variable and complex lesions.
• To enhance segmentation accuracy and boundary perception in challenging medical imaging tasks.

Main Methods:

• A spectrum-prior-boundary triple modeling paradigm was developed.
• Frequency domain reconstruction and modulation identified ambiguous signals.
• Level set-based segmentation incorporated anatomical distance fields and morphological priors.
• An auxiliary edge branch integrated deep and shallow features for boundary awareness.

Main Results:

• The proposed model significantly improved segmentation accuracy and boundary perception for colorectal cancer.
• Experiments demonstrated superior performance compared to state-of-the-art methods.
• The model showed strong generalization capabilities on lung and breast cancer segmentation tasks.

Conclusions:

• The feature-driven model effectively addresses limitations in complex medical image segmentation.
• The spectrum-prior-boundary triple modeling paradigm enhances performance by mining intrinsic data information.
• The method shows promise for diverse clinical applications requiring precise lesion segmentation.