Machine learning-based methods for detecting respiratory abnormalities using audio and visual analysis: A review

Abstract

Respiratory abnormalities pose a significant health burden, often demanding timely and accurate diagnosis. Traditional methods have limitations, prompting exploration of novel approaches. This review explores the potential of machine learning, leveraging both audio and visual analysis, for detecting respiratory abnormalities. We explore various methods employed, analyzing audio features like MFCCs and spectral energy, and exploring diverse visual features like chest wall motion and depth maps. Different machine learning techniques, including CNNs and RNNs, are discussed, highlighting their applications in detecting specific conditions like asthma and COPD. We systematically evaluate their performance, analyzing strengths and limitations of both audio-based and visual-based approaches. Further, we explore the potential of multimodal analysis, combining audio and visual information for enhanced performance. Reviewing existing studies, we assess the advantages gained over unimodal methods. Finally, we explore the clinical implications and future directions, discussing potential applications like remote monitoring and early diagnosis, while acknowledging remaining challenges and ethical considerations. By comprehensively examining the landscape of machine learning for audio and visual analysis in respiratory abnormality detection, this review offers valuable insights for researchers and clinicians, ultimately accelerating the translation of these promising methods into real-world practice.