Multimodal GRU Framework for Precise and Early Diagnosis of COPD, Asthma, and Pneumonia

Due to the high severity and overlapped symptoms of Chronic Obstructive Pulmonary disease (COPD), asthma, and pneumonia, professionals and experts cannot provide clear observations for the early diagnosis of multirespiratory pulmonary disorders. The primary objectives of this work are to develop a multimodal framework based on multivariate respiratory sound analysis to discriminate early diagnosis of COPD, asthma, and pneumonia. The proposed framework extracts discriminative features from preprocessed lung sound signal samples using multivariate Fourier-Bessel Series Expansion (FBSE), Mel spectrogram and Mel-frequency cepstral coefficients (MFCC). These features are integrated using a mathematically weighted multimodal fusion method and then fed into a Gated Recurrent Unit (GRU) network for training. This enables the classification of respiratory conditions, distinguishing between healthy, COPD, pneumonia, and asthma. The framework achieved an accuracy of 97.8% for classifying healthy and pulmonary disorders and reported high precision rates for asthma, COPD, and pneumonia.