Machine Learning to Automatically Differentiate Hypertrophic Cardiomyopathy, Cardiac Light Chain, and Cardiac...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Cardiac amyloidosis, caused by protein deposition, leads to poor outcomes and is often diagnosed late.
• Distinguishing cardiac amyloidosis from hypertrophic cardiomyopathy and its subtypes (ATTR, AL) using imaging is challenging.
• Current diagnostic methods lack reliable image-based classification for amyloidosis subtypes.

Purpose of the Study:

• To investigate a machine learning (ML) algorithm for identifying cardiac amyloidosis using cardiovascular magnetic resonance (CMR) data.
• To assess the ML algorithm's ability to differentiate between cardiac amyloidosis subtypes (AL and ATTR).
• To evaluate the performance of ML models with varying data inputs, including demographics and CMR imaging features.

Main Methods:

• A retrospective, multicenter study included patients with hypertrophic cardiomyopathy, AL/ATTR amyloidosis, and healthy volunteers.
• A 3-stage ML algorithm was trained using clinical information, semiautomated CMR imaging data, and qualitative CMR features.
• The algorithm sequentially differentiated healthy controls, hypertrophic cardiomyopathy, and amyloidosis subtypes (AL vs. ATTR).

Main Results:

• The ML algorithm demonstrated high accuracy in differentiating participants at each stage (AUCs 1.0, 0.99, 0.92).
• Excellent performance was maintained even when using only demographics and imaging data (AUCs 0.99, 0.98, 0.88).
• The model retained strong accuracy after excluding late gadolinium enhancement data (AUCs 1.0, 0.95, 0.86).

Conclusions:

• A trained ML model utilizing CMR imaging and patient demographics can accurately detect cardiac amyloidosis.
• The ML approach effectively differentiates between AL and ATTR cardiac amyloidosis subtypes.
• This AI-driven method shows promise for earlier and more precise diagnosis of cardiac amyloidosis.