Bimodal Framework for Cardiac Arrhythmia Analysis using Deep Learning

Medical image analysis plays a critical role in the early detection and treatment planning for various health conditions, including cardiac arrhythmias. Convolutional Neural Networks (CNNs) have shown immense potential in automating this process. This study introduces a novel approach to medical image analysis, specifically the classification of cardiac arrhythmias, through the creation of a hybrid model that combines the strengths of two CNN architectures: PPG 1D model and Spectrogram from ECG model. PPG 1D model, known for its computational efficiency and model size, is utilized as the primary feature extractor. This architecture excels at capturing essential image features at multiple scales while mitigating the risk of over fitting. In contrast, DenseNet, celebrated for its dense connections between layers, enhances feature reuse and gradient flow, ultimately improving feature representation. The hybrid model is developed by fine-tuning the pre-trained PPG 1D model with Spectrogram from ECG based classification. To further enhance the model's robustness and generalization capabilities, data augmentation techniques are implemented. This augmentation process ensures the model's effectiveness in handling variations in medical images. The results of this study demonstrates a substantial improvement in the accuracy of cardiac arrhythmia classification. This enhancement holds great importance as it equips healthcare professionals with a reliable tool for diagnosing cardiac arrhythmias more precisely. In summary, the potential of combining PPG 1D model and Spectrogram from ECG model significantly enhances the accuracy of cardiac arrhythmia classification. This innovative approach streamlines clinical workflows, paving the way for earlier and more accurate diagnoses of cardiac arrhythmias, leading to improved patient outcomes.