| Summary: | Lung disease, ranking third globally for causes of death with over 3 million annual fatalities according to the 2015 World Health Organization (WHO), underscores the significance of lung sound features in respiratory illness diagnosis. Auscultation, a subjective early detection method, led to the development of a Computer-Aided Diagnosis (CAD) system employing the VGG16 model for medical imaging complexity. Data limitations necessitated augmentation, enhancing VGG16's performance compared to non-augment results. Respiratory Disease Detection (RDD) task was introduced. Short-Time Fourier Transform (STFT) facilitated audio feature extraction, while VGG16, using transfer learning and fine-tuning, proved effective on a Kaggle-sourced dataset. Augmentation techniques, including pitch shifting, time stretching, and horizontal flipping, addressed class imbalance. The study introduces innovative data augmentation techniques to overcome the challenge of limited training data, demonstrating the effectiveness of augmentation in enhancing the VGG16 model's performance. © 2024 IEEE.
|
|---|