Overcoming Imbalanced Data Challenges in Spinal Disorder Detection: An Integrated YOLOv5 and CNNs Approach with SMOTE

This paper outlines an innovative methodology for the identification and categorization of spinal disorders, a significant health concern that dramatically influences individuals' well-being. The advanced procedure utilizes the potent capabilities of the YOLOv5 object detection framework and Convolutional Neural Networks (CNN) coupled with Synthetic Minority Over-sampling Technique (SMOTE). This combination adeptly addresses the challenges associated with class imbalance in diagnosing spinal disorders. Employing these techniques on the Burapha Spine Dataset, our model has achieved success in pinpointing and classifying four distinct spinal disorders, irrespective of their varied occurrence rates. The approach focuses on detecting each vertebra separately and then taking two neighboring vertebrae for spinal disorder classification via a CNN model incorporating SMOTE oversampling. As a result, this procedure considerably boosts diagnostic precision in spinal disorders.