A novel method based on deep learning for herniated lumbar disc segmentation

Abstract

Lower Back pain (LBP) is a common disease. Therefore, a common cause of leg pain and lower back is a lumbar disc herniation. Herniated lumbar disc represents a displacement of disc material (annulus fibrosis or nucleus pulpous). In most cases, the pain goes away within days to weeks; however, it can last for three months or more. Segmentation and Detection are the two most important tasks in computer aided diagnosis system (CAD) [24]. Extraction of herniated lumbar disc from magnetic (MRI) resonance imaging is a difficult task for radiologist. Detection of herniated disc was achieved by different methods such as region growing, active contours, watershed technique and thresholding. In our case, to detect intervertebral disc from lumbar MRI we developed an approach using convolutional neural networks in order to find the type of herniated lumbar disc [24] such as median, foraminal or post lateral [24]. We proposed to explore the importance of axial view MRI to find the type of herniation. Previous works were concentrated only on the sagittal View. The main objective of this paper is to automatically detect the intervertebral disc in magnetic resonance images(MRI) with bounding boxes and their classes which can facilitate diagnoses based on axial view MRI [40]. Therefore, the aim of this study is to assist detection using lumbar axial view MRI. A novel method is proposed in this paper based on deep convolutional neural networks. This study introduces the application of the convolutional neural network model. A framework was developed that enables the application of shape priors in the healthy part of intervertebral disc anatomy, with user intervention when the priors were inapplicable.