Lumbar and pelvic CT image segmentation based on cross-scale feature fusion and linear self-attention mechanism

• Published in: Scientific reports Vol. 15; no. 1; pp. 28131 - 15
• Main Authors: Li, Chaofan, Chen, Liping, Liu, Qiong, Teng, Jinnan
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 01.08.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 631/1647/245/1887; 639/166/985; 639/766/930/2735; Accuracy; Algorithms; Computed tomography; Convolutional neural network; Cross-scale feature fusion; Datasets; Deep learning; Fractures; Hip joint; Humanities and Social Sciences; Humans; Image processing; Image Processing, Computer-Assisted - methods; Linear attention mechanism; Lumbar Vertebrae - diagnostic imaging; Machine learning; Magnetic resonance imaging; Methods; multidisciplinary; Neural networks; Pelvic Bones - diagnostic imaging; Pelvis; Pelvis - diagnostic imaging; Science; Science (multidisciplinary); Signal processing; Spine; Spine (lumbar); Tomography, X-Ray Computed - methods; Universal segmentation; Vertebrae; Universal segmentation; Linear attention mechanism; Cross-scale feature fusion; Convolutional neural network
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-13569-0

The lumbar spine and pelvis are critical stress-bearing structures of the human body, and their rapid and accurate segmentation plays a vital role in clinical diagnosis and intervention. However, conventional CT imaging poses significant challenges due to the low contrast of sacral and bilateral hip tissues and the complex and highly similar intervertebral space structures within the lumbar spine. To address these challenges, we propose a general-purpose segmentation network that integrates a cross-scale feature fusion strategy with a linear self-attention mechanism. The proposed network effectively extracts multi-scale features and fuses them along the channel dimension, enabling both structural and boundary information of lumbar and pelvic regions to be captured within the encoder-decoder architecture.Furthermore, we introduce a linear mapping strategy to approximate the traditional attention matrix with a low-rank representation, allowing the linear attention mechanism to significantly reduce computational complexity while maintaining segmentation accuracy for vertebrae and pelvic bones. Comparative and ablation experiments conducted on the CTSpine1K and CTPelvic1K datasets demonstrate that our method achieves improvements of 1.5% in Dice Similarity Coefficient (DSC) and 2.6% in Hausdorff Distance (HD) over state-of-the-art models, validating the effectiveness of our approach in enhancing boundary segmentation quality and segmentation accuracy in homogeneous anatomical regions.