Foramina Localization - Semantic Segmentation of Neural Foramina on Lumbar Spine MRIs with Convolutional Neural Networks

Abstract

Lumbar foraminal stenosis is the diseases of compressed nerve roots in the lumbar foramina and can cause great pain with the patient. Traditionally, this disease is diagnosed by a radiologist on lumbar MRI scans. In the past decade computer aided diagnosis (CAD) has made its rise due to the recent successes of deep neural networks. Especially in the domain of automatic medial image analyses CAD has seen a tremendous growth of interest among researchers. This work presents a deep neural network that automatically localizes the lumbar foramina from MRI scans. The pipeline consists of two stages, first a semantic segmentation on a MRI volume is performed, and secondly the coordinates of the foramina are determined. Contrary to past research, our network makes no use of data driven postprocessing techniques or hand-crafted features.