Radiomics Feature Extraction and Machine Learning for Enhanced MRI-Based Detection of Spinal Canal Stenosis

Purpose Proper diagnosis and grading of Lumbar Spine Degenerative conditions help guide treatment and potential surgery to help alleviate back pain and improve overall health and quality of life for patients. The aim of the present study is to diagnose and grade lumbar spinal stenosis in MRI images using radiomics features and machine learning models. Materials and Methods This investigation involved the collection of 1500 Sagittal T2 STIR (Short-TI Inversion Recovery)-weighted sequence encompassing individuals with varying degrees of spinal canal stenosis (Normal/Mild, Moderate, Severe). Two experienced radiologist performed the segmentation of Sagittal slices, and radiomics features were subsequently extracted from each region of interest. Initially, four different machine learning models were deployed: Random forest (RF), Adaptive Boosting (AdaBoost), XGBoost (XGB), Artificial Neural Network (ANN). The performance of these chosen models was subjected to a more comprehensive examination. Results The AdaBoost model exhibited notable performance in classification of spinal canal stenosis, achieving accuracy and precision rates of 80% and 81%, respectively. The Area Under the Curve (AUC) values for RF, AdaBoost, XGB, and ANN were calculated, yielding values of 92.60% and 90.70%, 92.80% and 93.60%, respectively. In terms of specificity, the ANN model performed better than other models with a value of 90%. Conclusion These machine learning models exhibited remarkable capability in automatically detecting and grading spinal canal stenosis on spinal MRI images. By systematically integrating radiologist expertise into the analytical workflow, this approach enables the efficient and reliable screening of large cohorts for spinal stenosis, thereby enhancing diagnostic throughput and supporting clinical decision-making.