A Deep Learning Approach to Classify Fabry Cardiomyopathy from Hypertrophic Cardiomyopathy Using Cine Imaging on...

Area of Science:

• Cardiovascular Imaging and Artificial Intelligence
• Cardiac MRI Analysis for Differentiating Cardiomyopathies

Background:

• Accurate classification of left ventricular hypertrophy (LVH) is challenging, particularly differentiating it from hypertrophic cardiomyopathy (HCM) and Fabry disease.
• Current diagnostic methods often require multidisciplinary specialist input, leading to potential variability and diagnostic inconsistencies.
• While T1 mapping on cardiac MRI aids differentiation, distinguishing HCM from Fabry disease using standard echocardiography or MRI cine images remains difficult.

Purpose of the Study:

• To develop and validate an AI-powered classification model for distinguishing between HCM and Fabry disease using cardiac MRI.
• To address the diagnostic challenges faced by cardiologists in differentiating these specific LVH conditions.

Main Methods:

• Development of the MRI short-axis view left ventricular hypertrophy classifier (MSLVHC), an AI model trained on MRI short-axis (SAX) view cine images.
• Performance evaluation using metrics including F1-score, accuracy, and AUC on internal (Taipei Veterans General Hospital - TVGH) and external (Taichung Veterans General Hospital - TCVGH) datasets.
• Validation through a single-blinding study and external testing to assess real-world reliability and effectiveness.

Main Results:

• The MSLVHC model achieved high performance on the TVGH dataset, with an F1-score of 0.846, accuracy of 0.909, and AUC of 0.914.
• External validation on the TCVGH dataset demonstrated sustained effectiveness, yielding an F1-score of 0.727, accuracy of 0.806, and AUC of 0.918.
• The AI model proved reliable and useful in differentiating HCM from Fabry disease in an external testing scenario.

Conclusions:

• The AI-driven MSLVHC model offers a standardized and accurate approach to classifying LVH, specifically differentiating HCM from Fabry disease.
• This tool shows significant promise in assisting specialists, improving diagnostic consistency, and potentially streamlining the diagnostic pathway for complex cardiomyopathies.
• The model's validated performance across different datasets highlights its potential as a valuable addition to cardiovascular diagnostic imaging.