Multicenter Validation of Automated Segmentation and Composition Analysis of Lumbar Paraspinal Muscles Using...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Chronic low back pain is a significant global health concern with substantial socioeconomic impact.
• Changes in lumbar paraspinal muscles (LPM) are associated with chronic low back pain.
• Manual assessment of LPM on MRI is time-consuming and subject to variability.

Purpose of the Study:

• To develop and validate a deep learning (DL) method for automated segmentation and quantitative analysis of LPM.
• To assess muscle volume and fatty infiltration in LPM using multisequence, multicenter MRIs.
• To compare the performance of the automated DL method against manual segmentation and measurements.

Main Methods:

• A retrospective study utilizing 1,302 MRIs from 641 participants across five centers.
• A DL model was trained and externally validated for LPM segmentation, volume quantification, and fatty infiltration assessment.
• Performance was evaluated using Dice Similarity Coefficient (DSC) for segmentation and Intraclass Correlation Coefficients (ICCs) for measurements.

Main Results:

• The DL model achieved high segmentation performance with global DSC values of 0.98 (internal) and 0.93–0.97 (external).
• Statistical equivalence was confirmed between automated and manual measurements for muscle volume and fatty infiltration ratio (P < .05).
• High agreement was observed between automated and manual measurements (ICCs > 0.92).

Conclusions:

• The developed automated DL method accurately segments LPM on multisequence, multicenter MRIs.
• The automated method demonstrates statistical equivalence to manual measurements for muscle volume and fatty infiltration.
• This AI-driven approach offers a reliable and efficient tool for assessing LPM changes in chronic low back pain.