Hybrid DL and ML approach for MRI-based classification of bone marrow changes in lumbar vertebrae

Abstract

Alterations in the bone marrow changes lumbar vertebrae (BMCLVB) are considered important markers of chronic low back pain severity, particularly among patients with coexisting conditions like osteoporosis or cancer. Realizing these associations informs healthcare and insurance frameworks but also supports early intrusion planning for high-risk populations. This study aims to classification (BMCLVB) as normal or abnormal used magnetic resonance imaging (MRI) with machine learning (ML) model. A novel dataset comprising 1,018 BMCLVB MRI images was utilized to extract deep features via a pre-trained ConvNeXt-XLarge model. These features were then classified using different types in individual and ensemble ML algorithms. To ensure a comprehensive performance evaluation, all models were tested using accuracy, precision, recall, and F1-score. The combination of ConvNeXt-XLarge and logistic regression (LR) achieved a classification accuracy 93.14%, precision 93.22%, recall 94.83%, and F1-score 94.02%. These results highlight that the proposed model provides a fast and cost-efficient solution supporting the diagnosis of BMCLVB and potential to significantly improve clinical decision-making and patient care outcomes.