Quantitative Analysis of Spinal Canal Areas in the Lumbar Spine: An Imaging Informatics and Machine Learning Study

Area of Science:

• Radiology
• Medical Imaging Informatics
• Machine Learning

Background:

• Quantitative imaging biomarkers for spinal canal stenosis are currently lacking.
• Advancements in machine learning and medical imaging informatics offer potential solutions.

Purpose of the Study:

• To establish the foundation for quantitative imaging biomarkers for spinal canal stenosis diagnosis.
• To develop and validate machine learning algorithms for automated spinal canal measurement.

Main Methods:

• Machine learning models were trained to segment lumbar spinal canal and intervertebral disc areas on MRI.
• Spinal canal areas (L1-L5) were measured and compared against human delineations for validation.
• A normative cohort was analyzed to assess variations in spinal canal area with age, sex, and height.

Main Results:

• Machine-generated segmentations demonstrated comparability with human-generated ones.
• Measured spinal canal areas showed a statistically significant correlation with height (P < .05).
• No significant correlation was found between spinal canal area and age or sex.

Conclusions:

• Machine learning enables accurate, automated detection and quantification of spinal canal dimensions, comparable to human raters.
• This methodology can serve as a basis for future diagnostic tools to identify spinal stenosis.
• The established normative model provides a reference for detecting anatomical deviations indicative of spinal stenosis.