International Conference on Advanced Algorithms and Control Engineering · Mar 20, 2026
LiteMedNet: A Lightweight Direction–Frequency Enhanced Network for Cortical Bone Segmentation in X-ray Images
By Ji D, Liu H
Accurate cortical bone segmentation from X-ray images is essential for bone structure analysis and osteoporosis assessment. However, conventional U-Net-based models often involve large parameter sizes and high computational costs when processing high-resolution medical images. To address this issue, this study proposes a lightweight segmentation network, LiteMedNet, which employs MobileNet as the encoder backbone while retaining the skip-connection mechanism of U-Net. To enhance the segmentation of elongated cortical structures, two modules are introduced: a Spatial-Frequency Hybrid Feature Enhancement (SFHFE) module and a Vertical Strip Convolution (VSC) module. The proposed modules improve global structural representation and directional feature extraction. Experiments on a femoral X-ray dataset demonstrate that LiteMedNet achieves competitive segmentation performance with significantly reduced computational complexity. Furthermore, the Cortical Thickness Index (CTI) derived from the segmentation results shows a significant correlation with Bone Mineral Density (BMD) ($r \approx 0.46$,$p<0.001$), indicating its potential for automated bone health assessment.