Comparative Analysis of Selected Machine Learning Algorithms for Lumbar Spinal Stenosis Classification.

Abstract

Lumbar spinal stenosis (LSS) represents a prevalent etiology of low back discomfort among the adult population, attributable to a constriction that exerts pressure on the spinal cord or nerve roots. Accurate and timely diagnosis is crucial for effective management and treatment of LSS. This study presented a Machine learning and deep learning framework designed to classify lumbar spinal stenosis (LSS) severity by leveraging imaging data. The dataset of MRI images was obtained from the Kaggle online repository, the data came in a CSV file with a DCOM image folder. The dataset contained 48,692 images. The CSV and DCOM images were linked together. In doing this, missing, null, and invalid data were sought in the dataset and removed. Outliers too were checked and sorted out. 80% were used for training and 20% were used for testing. Using machine learning algorithms like Support Vector Machines (SVM) and Custom-made CNN. The models were trained to classify varying degrees of LSS severity, ranging from mild to severe. The models were evaluated using performance metrics such as accuracy, precision, recall, and F1-scores to quantify the model’s effectiveness. Due to the class imbalance in the dataset, the SVM and Custom-CNN models perform exceptionally well in the “mild” class but struggle with “severe” and “moderate” classes The results demonstrated that support vector machine (SVM), and Custom-made CNN models achieved accuracies of 89% and 90% respectively in classifying LSS severity. This study shows that the Custom- made CNN model performs better in classification of Lumbar spinal stenosis than the traditional Machine learning model using MRI imaging datasets. Cross-validation should be implemented to ensure the performance of the model is consistent across other subsets of data.