This article presents a comprehensive systematic literature review (SLR) on the outcomes of using artificial intelligence (AI) for the recognition of diseases through voice analysis, with a focus on biomarkers. The SLR process involved the selection and analysis of relevant research studies, including their methodologies, datasets, and findings. This review highlights the effectiveness of AI-guided tools for screening COVID-19 infections using cough and sneezing sounds, discusses the state-of-the-art methods employed, and explores the potential of deep learning models in improving disease detection and prediction.
Eligibility of the Studies:
The SLR included a total of 28 studies that met the inclusion and exclusion criteria. The selected studies utilized various datasets, including cough sound data from laboratory-confirmed patients, crowd-sourced data, and data collected via mobile apps. While some studies focused on COVID-19 prediction, others provided essential information for the SLR without specific COVID-19 relevance. The analysis excluded studies with invalid measurements, unavailability of full text, and publication years before 2019. Eight studies were included based on references, resulting in a final set of 28 studies for synthesis.
Studied non-Laboratory Confirmed Dataset and State-of-the-Art Methods:
The review examined the use of deep learning methods, such as convolutional neural networks (CNNs), for the recognition of COVID-19 through cough and sneezing sounds. AI-guided cough screening tools employing CNNs showed promising results, with high sensitivity, specificity, and overall accuracy. The studies compared laboratory-confirmed datasets with non-laboratory-confirmed datasets, demonstrating the superior performance of AI-guided approaches when trained on reliable data. Ensemble learning, CNNs, and deep neural networks were utilized to identify COVID-19 in sneezing noises. The research also emphasized the potential of incorporating biomarkers and clinical data, such as respiration, speech, throat clearing, wheezing, and breathing, to enhance decision-making in disease screening.
Discussion and Conclusions:
Based on the selected studies, coughing was identified as one of the most prominent symptoms for the detection and prediction of COVID-19. AI-guided cough screening, along with clinical testing, can aid in the diagnosis of COVID-19, making it a valuable tool, particularly in resource-constrained areas. Deep features were found to extend the generalization of AI models without relying on past information, allowing for more precise predictions and insights. The use of AI-guided tools for COVID-19 screening showed superior performance when trained on laboratory-confirmed datasets. However, some studies lacked comprehensive method descriptions or employed biased models, highlighting the need for further improvements.
The review concluded that AI-guided solutions using biomarkers and clinical data, combined with advanced deep learning models like CNNs, hold promise for improving the analysis and prediction of COVID-19 and other diseases. Future research should explore the integration of ensemble deep neural networks and multimodal learning methods to enhance disease screening outcomes.
Overall, this systematic literature review provides valuable insights into the application of AI in biomarker recognition and disease detection from voice analysis. The findings demonstrate the potential of AI-guided tools for accurate and efficient screening, paving the way for advancements in healthcare and disease management.