An Intelligent Respiratory Tract Auscultation Audio Signal Classification Technique
Abdul Moaiz, Department of Mechatronics Engineering, University of Engineering & Technology, Peshawar, Pakistan.
Shahzad Anwar, Department of Mechatronics Engineering, University of Engineering & Technology, Peshawar, Pakistan.
Gulbadan Sikander, Department of Mechatronics Engineering, University of Engineering & Technology, Peshawar, Pakistan.
Zhang Dong, Institute of Automation, Qilu University of Technology, Jinan, PR China.
Corresponding Author:
Abdul Moaiz (amoaiz_mct@uetpeshawar.edu.pk)
Abstract:
Respiratory tract infections are one of the leading causes of death. There are many different types of respiratory diseases like pneumonia, LRTI, URTI, and COPD. The integration of AI with medical devices can greatly enhance the early diagnosis of diseases. This study proposes an LSTM-based RNN model. The model is trained on more than 1800 lung audio recordings. Data augmentation is employed to get a more balanced dataset. Feature extraction is performed through a combination of MFCC and genetic algorithm to select the best features. The performance of the proposed model was evaluated and the results show that MFCC-GA-LSTM produced the best results with an accuracy of 98.65% and an F1 score of 99.16. The developed method is compared to existing methods and shows considerable improvement in terms of classification accuracy.
Keywords:
RNN; LSTM; Data Augmentation; Respiratory Infections