A diagnostic method for cardiomyopathy based on multimodal data.

Objectives: Currently, a multitude of machine learning techniques are available for the diagnosis of hypertrophic cardiomyopathy (HCM) and dilated cardiomyopathy (DCM) by utilizing electrocardiography (ECG) data. However, these methods rely on digital versions of ECG data, while in practice, numerous ECG data still exist in paper form. As a result, the accuracy of the existing machine learning diagnostic models is suboptimal in practical scenarios. In order to enhance the accuracy of machine learning models for diagnosing cardiomyopathy, we propose a multimodal machine learning model capable of diagnosing both HCM and DCM.

Methods: Our study employed an artificial neural network (ANN) for feature extraction from both the echocardiogram report form and biochemical examination data. Furthermore, a convolutional neural network (CNN) was utilized for feature extraction from the electrocardiogram (ECG). The resulting extracted features were subsequently integrated and inputted into a multilayer perceptron (MLP) for diagnostic classification.

Results: Our multimodal fusion model achieved a precision of 89.87%, recall of 91.20%, F1 score of 89.13%, and precision of 89.72%.

Conclusions: Compared to existing machine learning models, our proposed multimodal fusion model has achieved superior results in various performance metrics. We believe that our method is effective.