Design of a Hybrid Bioinspired Deep Learning Model for Identification of Heart Diseases Using Clinical Parameters

There has been an exponential increase in heart diseases due to an imbalanced diet, improper lifestyle, and abnormal clinical parameters. The majority of the models that researchers have developed to diagnose improper heart conditions are either extremely difficult to understand or lack flexibility in terms of the parameters that are incorporated in the classification process. Researchers have developed a wide variety of models to diagnose improper heart conditions. In addition to this, the performance of some of these models is significantly diminished when compared to an increase in the number of heart disease classes. This text proposes the design of a novel hybrid bioinspired deep learning model for the identification of heart diseases utilizing clinical parameters to circumvent the problems that have been outlined in this text. The model initially collects user-specific clinical parameters from multiple sources, that includes blood pressure levels, glucose levels, age, heart rate, smoking conditions, and eating habits, and converts them into multimodal parameter sets. These sets are evaluated via the estimation of Fourier, Wavelet, Cosine, and Convolutional feature sets. Due to the extraction of these features, the model can represent clinical parameters in multimodal forms. This allows for better pattern recognition and improved classification performance for different disease types. The extracted features are post-processed via a Grey Wolf Optimization (GWO)-based feature selection model, which aims in maximizing inter-class variance levels. The post-processed feature sets are classified via a customized 1D CNN-based model, that assists in the identification of different heart diseases, including myopathy, arrhythmias, congenital heart defects, heart valve disease, and coronary artery diseases with high accuracy levels. The proposed model was compared with various standard deep learning techniques, and it was observed that the model was able to showcase 2.3% higher accuracy, 1.9% higher precision, and 2.5% better recall on large-scale datasets, while the model was also able to reduce computational delay due to feature selection, which makes it highly useful for a wide variety of clinical use cases. © 2023, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.