An Enhanced YOLOv11 Framework for Automatic Lumbar Spine Level Detection
Students & Supervisors
Student Authors
Supervisors
Abstract
Detection of lumbar spine levels is crucial for clinical diagnostics, surgical planning, and treatment of spinal disorders. Manual annotation is time-consuming, error-prone, and subject to variability. We propose an enhanced deep learning framework based on the YOLOv11 object detection model to automatically detect and classify five lumbar intervertebral levels: L1–L2, L2–L3, L3–L4, L4–L5, and L5–S1. To optimize performance, Ghost Convolution (GhostConv) layers replaced standard convolutional (Conv) layers in the early stages to reduce computation by generating more feature maps through inexpensive operations. Additionally, a C2f (Concatenate-to-Fusion) module was used in place of CSPSA (Cross Stage Partial Spatial Attention) to enable efficient feature reuse with fewer parameters and lower memory usage. The model was trained on the JM-LS dataset. Data augmentation was applied, including horizontal/vertical flips, cropping (5–20%), grayscale conversion (15% of images), brightness variation (±18%), Gaussian blur (up to 0.8 pixels), and random noise (up to 0.94%). The proposed YOLOv11 model sets a new benchmark, achieving 96.6% precision, 96.3% recall, a 96% F1-score, and 99% mAP@50.
Keywords
Publication Details
- Type of Publication:
- Conference Name: 2025 28th International Conference on Computer and Information Technology (ICCIT)
- Date of Conference: 19/12/2025 - 19/12/2025
- Venue: Cox’s Bazar, Bangladesh
- Organizer: IEEE Bangladesh Section