Dynamic Strip Convolution and Adaptive Morphology Perception Plugin for Medical Anatomy Segmentation.

Area of Science:

• Medical imaging analysis
• Computer-aided diagnosis
• Anatomical segmentation

Background:

• Medical anatomy segmentation is crucial for computer-aided diagnosis and lesion localization.
• Existing methods overlook morphological heterogeneity, struggling with orientation-varying structures like ribs and clavicles.

Purpose of the Study:

• To propose a novel convolution plugin, DSC-AMP, for enhanced medical anatomy segmentation.
• To address the limitations of current methods in handling diverse anatomical shapes and orientations.

Main Methods:

• Developed a dynamic strip convolution (DSC) operator for customized receptive fields.
• Integrated an adaptive morphology perception (AMP) strategy using various shape-aware kernels.
• Combined DSC and AMP into a novel plugin (DSC-AMP) for medical anatomy segmentation.

Main Results:

• The proposed DSC-AMP approach demonstrated superior performance in segmenting heterogeneous medical anatomies.
• Extensive experiments on two large-scale datasets validated the effectiveness of the method.
• The approach successfully handles complex structures like ribs and clavicles with varying orientations.

Conclusions:

• The DSC-AMP plugin significantly improves medical anatomy segmentation accuracy.
• This novel method offers a robust solution for computer-aided diagnosis and medical reporting.
• The approach effectively addresses the challenge of morphological heterogeneity in medical images.