Multi-Modal COVID-19 Detection Using Cough Sounds and Medical Information with Attention-Enhanced Deep Learning

The COVID-19 pandemic has highlighted the need for fast, non-invasive, and cost-effective diagnostic tools. Cough sounds, as a prominent symptom of respiratory diseases, present a promising modality for automated COVID-19 detection. In this study, we propose a novel multi-modal deep learning framework for COVID-19 detection that leverages cough sounds and patient-specific medical information. Our approach extracts two types of acoustic features—Mel-Frequency Cepstral Coefficients (MFCCs) and Mel spectrograms—and integrates them with clinical metadata to improve diagnostic ac-curacy. The MFCC branch employs 1D convolutional layers followed by Efficient Channel Attention mechanism. The Mel spectrogram branch utilizes ResNet-50 combined with ECA to capture spatial and frequency-level patterns. Medical information is processed through fully connected layers. Outputs from all three branches are concatenated and passed through fully connected layers for final classification. The framework was evaluated on publicly available datasets, including COUGHVID and Coswara. Experimental results show that our method achieves an average Area Under the Receiver Operating Characteristic Curve (AUC) of 90.1%, outperforming existing approaches in the literature, which report average AUCs of 88.5% and 77.1%. These findings highlight the potential of the proposed approach as a reliable and accessible tool for COVID-19 detection.