Computer Analysis of Lumbar Spine Magnetic Resonance Imaging Scans via Neural Network Algorithms

Abstract

The paper contains a comparative analysis of the computer methods of the lumbar spine investigation. The authors suggest a method of computer analysis of the lumbar spine magnetic resonance imaging (MRI) scans for the determining of the pathology and describe the developed software. Key feature of the developed software system is the ability to perform automatic segmentation and analysis of T2 weighted MRI scans of the lumbar spine. The paper is relevant due to the lack of researches and software to determine the quantitative parameters of segmented anatomical bodies, as well as due to the inability of existing software systems to identify the pathologies in the segmented objects of the lumbar spine. Developed software solution is a complex of software subsystems based on the neural network that allow to fulfill: the segmentation of intervertebral discs on T2 weighted MRI scans of the lumbar spine via convolutional neural network; identification of the quantitative parameters of the selected anatomical bodies for further analysis by a radiologist; identification the pathologies of Schmorl hernias, protrusion and extrusion on segmented intervertebral discs. The suggested method and developed software are implemented in the radiological department of the research medical institute.