SpinalDeepRS152: Classification and Prediction of Multi Class Spinal Diseases with Deep Residual Neural Network - ResNet152

In recent years, the prevalence of spinal diseases has increased due to sedentary lifestyles, aging populations, and increased diagnostic capabilities. The accurate and timely diagnosis of spinal conditions such as scoliosis, spondylosis, herniated discs, spinal stenosis, osteoarthritis, ankylosing spondylitis, degenerative disc disease, and vertebral fractures is crucial for good treatment. Traditional diagnostic methods rely heavily on the expertise of radiologists who manually inspect medical images, a process that is error pron. This study explores the application of deep residual neural networks, specifically ResNet152, for the automated classification and prediction of eight distinct spinal diseases using X-ray and magnetic resonance images. DL, particularly CNNs, has shown immense promise in the field of medical imaging. The ResNet152 model, A deep residual network effectively minimizes the issue of vanishing gradients, enabling the training of extremely deep networks. This study leverages the ResNet152 architecture to develop a robust and efficient model for classifying and predicting eight common spinal diseases. The dataset used in this study comprises X-ray collected from a diverse patient population. A total of 630 images were curated, each image annotated by experienced radiologists. The dataset includes images of patients diagnosed with scoliosis, spondylosis, herniated discs, spinal stenosis, osteoarthritis, ankylosing spondylitis, degenerative disc disease, and vertebral fractures. In addition, images of healthy individuals were included as controls. In this we have used the ResNet152 model which achieved an overall precision of 91. 29% in the test set, indicating high reliability in distinguishing between different spinal conditions. High precision and recall values were observed for all eight classes, with scoliosis and herniated discs achieving precision and recall rates above 98%.The confusion matrix revealed minimal misclassifications, with most errors occurring between closely related conditions such as spondylosis and degenerative disc disease.