Chronic low back pain (cLBP) profoundly impacts quality of life, yet its underlying mechanisms remain poorly understood. Three-dimensional (3D) ultrasound imaging offers valuable insights into cLBP but poses challenges of manual annotation due to large data volume and poor image quality. To address these issues, a Cross-Scan Fusion Network (CSFN) is proposed for anatomical tissue layer segmentation in 3D ultrasound images. CSFN leverages a VoxelMorph-based registration model with Projected Hausdorff Loss (PHDL) to generate synthetic image-mask pairs, enabling both sample-efficient learning (CSFN-SEL) and semi-supervised learning (CSFN-SSL). Our results demonstrated that CSFN-SEL achieved superior segmentation performance compared to fully supervised nnU-Net (baseline) across varying labeled samples, with mean Dice coefficients of 74.62% ± 2.45% (5 samples) and 80.93% ± 1.87% (19 samples). CSFN-SSL further improved performance, achieving 79.33% ± 1.96% and 81.14% ± 1.43% for 5 and 19 samples, respectively, significantly outperforming baseline methods (p < 0.05). These findings highlight the effectiveness of CSFN in enhancing segmentation accuracy and robustness under limited labeled data conditions, offering a feasible solution for advancing cLBP imaging and analysis.