Qualitative versus quantitative lumbar spinal stenosis grading by machine learning supported texture...

Area of Science:

• Radiology
• Medical Imaging Analysis
• Machine Learning in Medicine

Background:

• Lumbar spinal stenosis (LSS) diagnosis relies on magnetic resonance imaging (MRI).
• Current qualitative assessment methods for LSS exhibit moderate reproducibility.
• Accurate grading of LSS is crucial for effective patient management.

Purpose of the Study:

• To compare the accuracy and reproducibility of qualitative MRI ratings versus quantitative texture analysis (TA) for LSS detection and grading.
• To evaluate the effectiveness of machine learning algorithms in analyzing TA for LSS assessment.
• To determine if TA can improve upon existing diagnostic methods for LSS.

Main Methods:

• Analysis of 343 T2-weighted lumbar spine MRI scans from 82 patients with severe LSS.
• Qualitative grading by an expert and two independent readers using standard and Schizas scales.
• Quantitative texture analysis (TA) using machine learning on defined regions of interest.
• Interreader agreement assessed using Cohen's Kappa and intraclass correlation.

Main Results:

• Qualitative LSS ratings demonstrated moderate reproducibility across both classification systems.
• Quantitative TA, analyzed with a decision tree classifier, showed superior performance for LSS grading compared to qualitative methods.
• Machine learning analysis of TA provided highly reproducible parameters for accurate severe LSS detection.

Conclusions:

• Qualitative LSS grading shows only moderate reproducibility, irrespective of the classification system used.
• Texture analysis combined with machine learning offers highly reproducible quantitative parameters.
• This quantitative approach enhances accuracy in detecting severe LSS, with minimal influence from grading score or cross-sectional area definition.