Automatic semantic segmentation and detection of vertebras and intervertebral discs by neural networks

Abstract

Objective: : To realize a tool for automatic segmentation and detection of lumbar disc degenerative disease and fractures in MRI images using convolutional neural networks. Material and methods: : We developed, on one hand, a Picture Archiving and Communication System (PACS) with a DICOM viewer (Digital Imaging and Communications in Medicine) to extract training data and implement 2 CNN networks dedicated to the segmentation task and on the other hand, the analysis of simple lumbar pathologies on the other hand.Two hundred forty-four MRI scans T2 weighted sagittal were selected at the university hospital Pasteur 2 of Nice - France.After accomplishing segmentation and classification task for all structures of interest, we trained two neural networks (U-Net++ and Yolov5x) to segment and detect discs and vertebras. Results: : Neural networks allowed semantic segmentations with a good accuracy of the order of 0.96 and 0.93 DICE index for intervertebral discs and vertebral bodies respectively.They are less effective in detecting of common pathologies (degenerative disc disease, disc herniation, vertebral fractures) with an area under the precision-recall curve of 0.88 for fractures and 0.76 for degenerative disc disease. Conclusion: : Our work has shown good efficiency for the segmentation of vertebral bodies and intervertebral discs, but progress remains to be made in the detection of degenerative discopathy and vertebral fractures.On this last point, we believe that the lower performance can be attributed to a lack of training images and that better performances will be achieved with the increase of the training data set on our home made PACS. We will also provide other regions of interest and conditions.

Keywords

• Artificial intelligence
• Deep neural network learning
• Magnetic resonance imaging
• Spinal pathologies
• Feasibility analysis