Application of a Deep Learning Approach to Analyze Large-Scale MRI Data of the Spine

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Biomedical engineering

Background:

• The German National Cohort (GNC) possesses extensive standardized MRI datasets of the entire spine.
• Handling large-scale spinal MRI data requires advanced computational tools like artificial intelligence (AI).
• Establishing normative biometric reference values for spinal structures is crucial for clinical applications.

Purpose of the Study:

• To present an AI software tool designed for analyzing spine MRI datasets.
• To demonstrate the capability of AI in generating standardized biometric reference values for spinal components.
• To validate the AI tool's performance against manual segmentation methods.

Main Methods:

• Development and training of a 3D U-Net AI algorithm on 330 representative GNC MRI datasets.
• Validation and testing of the AI algorithm using a subset of the GNC data.
• Application of the trained AI algorithm to segment and analyze vertebral bodies (VB), intervertebral discs (VD), and spinal canals (SC) across the full dataset (n=10,215).

Main Results:

• Successful and reliable segmentation of VB, VD, and SC was achieved using the AI-based algorithm.
• The AI tool demonstrated excellent agreement with manually segmented spinal MRI datasets.
• A software tool was successfully developed to provide age, sex, and height-matched comparative biometric data.

Conclusions:

• AI-based algorithms are effective for the reliable segmentation of large-scale spinal MRI datasets.
• The developed AI software tool can generate standardized biometric data for spinal structures.
• Future analysis of the complete GNC MRI dataset (nearly 30,000 subjects) will enable the generation of comprehensive normative standard values.