Predicting Lumbar Spondylolisthesis: A Hybrid Deep Learning Approach

Spondylolisthesis is a chronic disease, and a timely diagnosis of it may help in avoiding surgery. Disease identification in x-ray radiographs is very challenging. Strengthening the feature extraction tool in VGG16 has improved the classification rate. But the fully connected layers of VGG16 are not efficient at capturing the positional structure of an object in images. Capsule network (CapsNet) works with capsules (neuron clusters) rather than a single neuron to grasp the properties of the provided image to match the pattern. In this study, an integrated model that is a combination of VGG16 and CapsNet (S-VCNet) is proposed. In the model, VGG16 is used as a feature extractor. After feature extraction, the output is fed to CapsNet for disease identification. A private dataset is used that contains 466 X-ray radiographs, including 186 images displaying a spine with spondylolisthesis and 280 images depicting a normal spine. The suggested model is the first step towards developing a web-based radiological diagnosis tool that can be utilized in outpatient clinics where there are not enough qualified medical professionals. Experimental results demonstrate that the developed model outperformed the other models that are used for lumbar spondylolisthesis diagnosis with 98% accuracy. After the performance check, the model has been successfully deployed on the Gradio web app platform to produce the outcome in less than 20 s.