[Paper Review] IATos: AI-powered pre-screening tool for COVID-19 from cough audio samples

D. Trejo Pizzo, S. Esteban|arXiv (Cornell University)|Apr 27, 2021

COVID-19 diagnosis using AI18 citations

TL;DR

IATos is an AI-powered pre-screening tool that uses deep learning to classify cough audio samples for COVID-19 detection, achieving 86% accuracy in distinguishing positive from negative cases using RT-PCR as the gold standard. The system leverages large-scale audio data from Buenos Aires to enable scalable, low-cost, and non-invasive population-level screening as a complement to clinical testing.

ABSTRACT

OBJECTIVE: Our objective is to evaluate the possibility of using cough audio recordings (spontaneous or simulated) to detect sound patterns in people who are diagnosed with COVID-19. The research question that led our work was: what is the sensitivity and specificity of a machine learning based COVID-19 cough classifier, using RT-PCR tests as gold standard? SETTING: The audio samples that were collected for this study belong to individuals who were swabbed in the City of Buenos Aires in 20 public and 1 private facilities where RT-PCR studies were carried out on patients suspected of COVID, and 14 out-of-hospital isolation units for patients with confirmed COVID mild cases. The audios were collected through the Buenos Aires city government WhatsApp chatbot that was specifically designed to address citizen inquiries related to the coronavirus pandemic (COVID-19). PARTICIPANTS: The data collected corresponds to 2821 individuals who were swabbed in the City of Buenos Aires, between August 11 and December 2, 2020. Individuals were divided into 1409 that tested positive for COVID-19 and 1412 that tested negative. From this sample group, 52.6% of the individuals were female and 47.4% were male. 2.5% were between the age of 0 and 20 , 61.1% between the age of 21 and 40 , 30.3% between the age of 41 and 60 and 6.1% were over 61 years of age. RESULTS: Using the dataset of 2821 individuals our results showed that the neural network classifier was able to discriminate between the COVID-19 positive and the healthy coughs with an accuracy of 86%. This accuracy obtained during the training process was later tested and confirmed with a second dataset corresponding to 492 individuals.

Motivation & Objective

• To evaluate the sensitivity and specificity of a machine learning-based cough classifier for detecting COVID-19 using cough audio recordings.
• To develop a scalable, low-cost digital pre-screening tool to reduce strain on healthcare systems and minimize exposure risks.
• To validate the performance of the AI model using real-world RT-PCR-confirmed cases from public testing facilities in Buenos Aires.
• To explore the feasibility of deploying such a tool for population-wide screening in urban settings.
• To contribute open-source methods and data to support global innovation in pandemic response technologies.

Proposed method

• The study employs a deep neural network classifier trained on cough audio samples collected via a citywide WhatsApp chatbot during the pandemic.
• Audio recordings were categorized as positive or negative based on RT-PCR test results, forming the ground truth for model training and evaluation.
• The model was first trained on a dataset of 2,821 individuals (1,409 positive, 1,412 negative), then retrained on a larger dataset of 143,351 individuals (18,271 positive, 125,080 negative) to improve robustness.
• The system uses audio signal processing techniques to extract features from spontaneous or forced coughs, focusing on respiratory sound patterns linked to infection.
• Model performance was validated using a held-out test set of 492 individuals, ensuring generalization beyond the initial training data.
• The final model was optimized for high specificity in detecting negative cases, despite class imbalance favoring negative samples (90% of data).

Experimental results

Research questions

• RQ1 What is the sensitivity and specificity of a machine learning-based cough classifier for detecting COVID-19 using RT-PCR as the gold standard?
• RQ2 Can cough audio recordings from a diverse urban population reliably distinguish between individuals with and without SARS-CoV-2 infection?
• RQ3 How does model performance scale when trained on large, real-world datasets collected during a pandemic outbreak?
• RQ4 To what extent can such a tool reduce the burden on conventional testing infrastructure through pre-screening?
• RQ5 Can the system be effectively deployed as a scalable, low-cost, and safe digital triage tool in public health settings?

Key findings

• The neural network classifier achieved 88% accuracy in distinguishing between COVID-19 positive and negative coughs on the initial dataset of 2,821 individuals.
• When tested on an independent validation set of 492 individuals, the model maintained strong performance, confirming its reliability.
• After retraining on a larger dataset of 143,351 individuals, the model achieved 86% accuracy, demonstrating scalability and robustness.
• The model showed improved performance in identifying negative cases, likely due to the high proportion of negative samples (90%) in the second dataset.
• The system demonstrated potential for large-scale deployment as a pre-screening tool, capable of processing thousands of cough samples daily with minimal infrastructure.
• The study confirms that cough audio analysis can serve as a viable, non-invasive, and cost-effective complement to RT-PCR testing in pandemic management.