Lumbar Spine Discs Classification Based on Deep Convolutional Neural Networks using Axial View MRI

Abstract Axial Lumbar disc herniation recognition is a difficult task to achieve, due to many challenges such as complex background, noise, blurry image. Lumbar discs are small joints that lie between each two vertebrae (L1-L2, L2-L3, L3-L4, L4-L5 and L5-S1). The segmentation and localization of the different discs are the most important tasks in Computer aided diagnosing of herniation. During the last five years, deep learning based methods have set new standards for many computer vision and pattern recognition research. In this work, our objective is to develop an automatic system based on deep convolutional neural network. This Network processes the input MRI (Magnetic Resonance Imaging) in multiple scales of context and then merges the high-level features to enhance the capability of the network to detect discs from lumbar spine. In this study, we are particularly interested in convolutional neural networks (CNN); it was characterized by a topology similar to a visual cortex of mammals. In fact, these kind of techniques has been applied successfully in many classification problems. In order to recognize herniated lumbar disc in Magnetic resonance imaging (MRI), we have chosen to use Convolutional neural networks based on VGG16 architecture. Experiments were carried on our own dataset from Sahloul University Hospital of Sousse.