Журнал позвоночника

Analysis of Large-Scale MRI Spine Data Using a Deep Learning Approach

Abstract

Alcantara Ortigoza

The German National Cohort (GNC) has the potential to provide standardized biometric reference values for intervertebral discs (VD), vertebral bodies (VB) and the spinal canal (SC) thanks to its uniform MRI datasets covering the entire spine. Artificial intelligence (AI) tools are required to manage such massive amounts of big data. An AI software tool for analyzing spine MRI and generating normative standard values will be presented in this manuscript. Age, sex and height parameters were evenly distributed among the 330 representative GNC MRI datasets that were chosen at random. A 3D U-Net was used to train, validate and test an AI algorithm. In the end, the entire dataset (n = 10,215) was looked at by the machine learning algorithm. An AI-based algorithm was used to successfully segment and analyze VB, VD and SC. For the purpose of analyzing spine MRI data and providing age, sex and height-matched comparative biometric data, a software tool was developed. The reliable segmentation of MRI datasets of the entire spine from the GNC using an artificial intelligence algorithm was possible and achieved excellent agreement with manually segmented datasets. In the not-too-distant future, it will be possible to generate genuine normative standard values by analyzing the entire GNC MRI dataset, which includes nearly 30,000 subjects.