Improving Lumbar Disc Bulging Detection in MRI Spinal Imaging: A Deep Learning Approach

Abstract

Lower back pain is a common ailment that affects many people resulting from various spinal diseases such as disk bulge. Disc bulging, which refers to the narrowing of the intervertebral disc within the spinal canal, can cause lower back pain and lead to lumbar spine stenosis. Deep learning algorithms based on artificial intelligence are indispensable in the field of medicine, enhancing the precision of medical image diagnosis significantly and the ability to process massive amounts of data which contributes to reducing the workload on radiologists and doctors. Therefore, Deep learning techniques have become a more helpful tool to overcome this problem. For this purpose, this study employed the YOLO-v7, YOLOv8m object detection technique to build a model to detect lumbar spine disc discords using MRI data from 291 individuals suffering from lower back pain. The image data utilized for training the model were divided into three segments: 70% for training, 20% for validation, and 10% for testing, and validate the model results according to evaluation metrics. shows the optimal model validated externally utilizing new lumbar spine MRI images and assessment with the radiologist. YOLOv8m achieved a mean average in precision (93.7%), precision (90.5%), recall (79.2%), and F1-score (88.6%), accuracy (89.8%). A deep learning model demonstrated similar agreement to subspecialist radiologists in detecting and classifying lumbar spine stenosis on lumbar spine MRI. By selecting the most suitable YOLO model, doctors can significantly enhance their ability to detect lumbar spine stenosis at an early stage and effectively mitigate potential negative consequences for their patients.