Harnessing AI and Machine Learning Methods for Supporting Cough-Based Diagnostics in Severe Respiratory Diseases

Rishabh Chandaliya and Martin Serrano

Artificial Intelligence holds enormous promise in the field of health diagnostics, healthcare and wellbeing. Coronavirus Diseases (COVID), like other respiratory diseases, continue to be a focus research area due to their high index of mortality. The COVID-19 pandemic awakened society and particularly the research community to the need  for a better understanding about  symptoms, pre- and post- infections conditions and the importance of better identification of the multiple variations that may exist as result of mutations of COVID viruses.

In the Insight SFI Research Centre at the University of Galway researchers were able to  exploit large amount of data relating to the COVID-19 cough symptoms and conditions. The data was collected via web apps, mobile apps and edge devices. They used this data and combined it with international datasets such as the Coswara, Virufy, Coughvid and Cambridge University datasets, to train an AI-powered system to spot patterns in a COVID-19 cough, so that the disease could be diagnosed more readily, using an app, in various settings following spectral analysis of the data.

Audio and breathing samples were captured using web applications and the raw audio was converted into imagery including spectrograms and high resolution images and audio for features extraction and normalisation of the data. Large COVID datasets provided insights that were unavailable to researchers. By training the AI-powered engine to recognise features of the COVID cough and represent those features pictorially with comparisons across data points, the researchers had access to more information-rich images. The applications for health diagnostics are significant. A smart phone app could be used to capture a cough remotely (ie by a user at home or outside a clinical setting) and the app itself can provide diagnostic insights to the patient and the health care provider. This promises faster, safer and more cost effective diagnostics for communicable diseases or simply for an early detection based on symptoms and early stage health conditions.