The value of unsupervised machine learning algorithms based on CT and MRI for predicting sarcopenia.

Area of Science:

• Radiology and Medical Imaging
• Machine Learning in Healthcare
• Sarcopenia Research

Background:

• Sarcopenia, a progressive loss of skeletal muscle mass and strength, poses a significant health challenge.
• Accurate and early detection of sarcopenia is crucial for timely intervention and management.
• Unsupervised machine learning offers potential for automated analysis of medical imaging data.

Purpose of the Study:

• To evaluate the effectiveness of unsupervised machine learning algorithms (Gaussian Mixture Model, K-means, Otsu) for sarcopenia prediction.
• To compare the performance of these algorithms using computed tomography (CT) versus magnetic resonance imaging (MRI) data.
• To identify the most accurate and stable algorithm for sarcopenia detection.

Main Methods:

• Retrospective analysis of 191 sarcopenia and 327 control patients.
• Manual delineation of paravertebral muscles at L3/4 on CT and MRI images.
• Automatic segmentation using Gaussian Mixture Model (GMM), K-means, and Otsu algorithms.
• Logistic regression for predictive modeling and area under the curve (AUC) for performance evaluation.
• Five-fold cross-validation for model stability assessment.

Main Results:

• CT-based unsupervised algorithms outperformed MRI-based ones in sarcopenia prediction.
• The CT-based Otsu model achieved the highest predictive performance (AUC 0.986 training, 0.958 validation).
• CT-based GMM showed strong performance (AUC 0.990 training, 0.903 validation) and superior stability.
• CT-based K-means also demonstrated predictive capability (AUC 0.727 training, 0.772 validation).

Conclusions:

• Unsupervised machine learning models utilizing CT data are highly effective for sarcopenia prediction.
• CT-based Otsu and GMM models demonstrate exceptional efficacy, with AUCs exceeding 0.95.
• These findings support the use of CT-based machine learning for robust sarcopenia detection.