A Comprehensive Review of Artificial Intelligence for Lumbar Spine MRI Analysis and Clinical Assessment

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Lower back pain is one of the major diseases to disability worldwide, with lumbar intervertebral disc degeneration (IVDD), disc herniation (DH), and lumbar spinal stenosis (LSS) being the most common underlying causes. Magnetic resonance imaging (MRI) is the imaging modality of choice for assessing these conditions; however, conventional MRI interpretation continues to be time-consuming, subjective, and prone to considerable interobserver variability. Recent advances in artificial intelligence (AI), particularly deep learning and advanced AI frameworks, have shown strong potential for optimizing the accuracy, consistency, and efficiency of lumbar spine MRI analysis. This review offers an in-depth overview of AI applications for lumbar spine MRIanalysis of IVDD, DH and LSS. It includes lumbar spine anatomy, pathophysiology of diseases, MRI grading systems and publicly available datasets including SPIDER, LumbarDISC and LSpineSMRI. AI methodologies are comprehensively reviewed across deep learning methods and advanced AI frameworks, covering feature–classifier hybrids, multi-stage functional cascades, cross-modal semantic fusion, and consensus ensemble systems, alongside widely adopted architectures such as CNNs, U-Net, YOLO, and transformer-based models for localization, segmentation, classification, and severity grading. Reported performance in the reviewed studies showed Dice similarity coefficients exceeding 0.90 for segmentation tasks and AUC values up to approximately 0.98 for lumbar spinal stenosis assessment. The review further discusses the key challenges to clinical translation, encompassing dataset heterogeneity, class imbalance, domain shift, limited external validation, and insufficient interpretability, while outlining future research priorities such as multimodal learning, self-supervisedlearning, foundation models, and explainable AI (XAI) to enhance model robustness, transparency, and clinical applicability. This review presents a clinically driven and technically structured perspective on the current landscape of AI for lumbar spine MRI analysis and identifies directions for developing more robust, interpretable, and clinically deployable diagnostic systems.