A diagnostic method for cardiomyopathy based on multimodal data.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Machine learning models for hypertrophic cardiomyopathy (HCM) and dilated cardiomyopathy (DCM) diagnosis often rely on digital electrocardiogram (ECG) data.
• Practical application is limited as many ECGs exist in paper form, reducing the accuracy of current diagnostic models.
• There is a need for improved machine learning approaches to accurately diagnose cardiomyopathies from diverse data sources.

Purpose of the Study:

• To develop and evaluate a multimodal machine learning model for enhanced diagnosis of HCM and DCM.
• To integrate features from echocardiogram reports, biochemical data, and ECGs for improved diagnostic accuracy.
• To overcome limitations of existing models that primarily use digital ECG data.

Main Methods:

• Utilized an artificial neural network (ANN) for feature extraction from echocardiogram reports and biochemical data.
• Employed a convolutional neural network (CNN) for feature extraction from electrocardiogram (ECG) data.
• Integrated extracted features and input them into a multilayer perceptron (MLP) for final diagnostic classification.

Main Results:

• The multimodal fusion model demonstrated strong performance metrics.
• Achieved a precision of 89.87% and a recall of 91.20%.
• Reported an F1 score of 89.13% and a precision of 89.72%.

Conclusions:

• The proposed multimodal fusion model significantly outperforms existing machine learning models for cardiomyopathy diagnosis.
• The model shows effectiveness in improving diagnostic accuracy by integrating multiple data types.
• This approach offers a promising solution for more accurate and practical diagnosis of HCM and DCM.