CMCNet:Cross-directional morphology-aware convolution network for chest X-ray anatomy segmentation

Lili Huang, Yuhan Feng, Xiaowei Zhao, Chenglong Li, Jin Tang

Abstract: Highlights•We propose the Cross-Directional Morphological-Aware Module (CMAM) by embedding learnable offset parameters into axial convolutions, adaptively adjusting convolution pathways to accommodate diverse local structures, thereby enhancing the capability to capture axial anatomical features.•We employ a hierarchical multi-scale architecture that enables parallel feature extraction through multi-scale convolution kernels, systematically identifying morphological differences and variation patterns across anatomical structures of varying dimensions.•We propose the Double Cross-Attention Module (DCAM) that generates axial attention maps through CMAM-derived morphological features, enhancing cross-axial anatomical representation while improving overall model performance.

External IDs:dblp:journals/ijon/HuangFZLT25