Nearest-Neighbor Dual-Path Contrastive Learning for Lumbar Disc Herniation MRI Image Classification

Lumbar disc herniation (LDH) is a common spinal condition that profoundly affects patients’ quality of life. Timely and precise diagnosis is essential for efficient therapy and enhancing patient outcomes. This paper introduces an innovative LDH classification framework utilizing nearest-neighbor dual-path contrastive learning, integrating global and local feature learning to improve lumbar MRI image classification efficacy. The global path delineates semantic linkages among samples by nearest-neighbor contrastive learning, enhancing global representations, whereas the local path employs clipped regions and data augmentation to highlight essential details, thus improving fine-grained feature modeling. The novel nearest-neighbor-based positive sample construction enhances feature consistency and classification accuracy by reducing the impact of irrelevant examples. Our method achieves state-of-the-art accuracy and robustness in complicated classification tasks, as shown by experimental results on the lumbar MRI dataset. This discovery enhances automated LDH diagnosis and offers a viable avenue for accurate and efficient automated diagnosis in intricate medical imaging contexts.