Machine-Learning-Based Diagnostics of Cardiac Sarcoidosis Using Multi-Chamber Wall Motion Analyses

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning

Background:

• Cardiac sarcoidosis (CS) presents with non-specific clinical and phenotypical features, making diagnosis difficult.
• Accurate diagnosis of CS is crucial for effective patient management and treatment strategies.

Purpose of the Study:

• To develop and evaluate a machine learning-based diagnostic approach for CS using cardiac magnetic resonance imaging (CMR).
• To utilize multi-chamber volumetrics and strain feature tracking for improved CS detection.
• To assess the efficacy of various machine learning classifiers in differentiating CS patients from controls.

Main Methods:

• Cardiac magnetic resonance imaging (CMR) was performed on 45 CMR-negative sarcoidosis patients, 18 CMR-positive sarcoidosis patients, and 44 controls.
• Multi-chamber volumetric and strain parameters were analyzed using logistic regression, KNN, DT, RF, SVM, GBoost, XGBoost, and Voting classifiers.
• Feature selection techniques were employed to enhance classification accuracy, particularly for differentiating CMR-positive from CMR-negative patients.

Main Results:

• Random Forest (RF) and Voting classifiers achieved the highest prediction rates (81.82%) in a three-cluster analysis (controls vs. CMR-positive vs. CMR-negative).
• Logistic regression, RF, and SVM classifiers demonstrated high prediction rates (96.97%) in a two-cluster analysis differentiating controls from all sarcoidosis patients.
• Feature selection significantly improved the accuracy of logistic regression in discriminating between CMR-positive and CMR-negative patients (89.47%).

Conclusions:

• Supervised machine learning utilizing multi-chamber cardiac function and strain data offers a non-contrast method for accurate differentiation between healthy individuals and sarcoidosis patients.
• Feature selection effectively addresses the challenge of discriminating between CMR-positive and CMR-negative patients, leading to high prediction accuracy.
• The study suggests a potentially higher prevalence of cardiac involvement in sarcoidosis than previously recognized, necessitating a re-evaluation of clinical disease management strategies.