Testing Failed to Overtake COVID-19 Spread
The COVID-19 pandemic showed that humanity lacked the preparedness and means to test the population reliably and quickly enough to contain the spread of the virus. As it spread throughout the world, from nation to nation, from city to city, the virus always seemed several steps ahead of our capacity to test the population, identify and contain outbreaks.
The standard means of testing adopted worldwide during most of the pandemic has been real-time reverse transcription-polymerase chain reaction (RT-PCR). Although providing reliable test results, RT-PCR has several shortcomings that made it non-viable to contain the pandemic:
• Costly – Requires expensive chemical reagents and devices and expert personnel.
• Time-consuming – The results typically take 2 to 48 hours.
• Unsafe sample collection – The sample collection process implies a violation of social distancing, which increases the chance of infection spread.
AI-Powered Point-of-Care Testing for COVID-19
The payoff of implementing rapid point-of-care tests that prevent the need for unnecessary contact with healthcare professionals could be immense: humanity would possess practical tools for outbreak detection and containment, which could ultimately prevent the loss of life.
Recently, due to the growing availability of adequate datasets, reliable machine learning algorithms, and the widespread use of smartphones, it is now possible to supplement diagnostic and disease stratification efforts with point-of-care testing (POCT).
While physicians routinely use audio signals – such as breath sounds during auscultation – as a tool to diagnose diseases or evaluate their progression since the dawn of the medical field, the prominent symptoms of COVID-19 include cough and breathing difficulties. Combining both facts mentioned above implies there is potential to use the power of information technology to give healthcare professionals some help.
The DiCOVA Challenge And The WHO’s Stringent Requirements
Several research teams and industry groups worldwide are working to achieve an accurate COVID-19 diagnostic tool. This tool would analyze the sound of a forced cough with a specialized algorithm developed by machine learning, even despite the complete absence of symptoms. An exceptionally high moment of this effort happened during early 2021 in the DiCOVA (Diagnostics of COVID-19 using Acoustics) challenge.
The DiCOVA challenge launched on 4 February 2021 and ended on 23 March 2021. The main goals of this challenge were twofold:
- To provide a benchmark for monitoring the progress in acoustic-based diagnostics of COVID-19.
- To provide speech and audio researchers a problem of immediate societal relevance they can solve with their particular set of skills.
An additional, more ambitious goal would be to achieve the requirements defined by the World Health Organization (WHO), which stated that a sensitivity of 70% (at a specificity of 95%) is necessary for a proper POCT tool.
Teams from many different countries, academia or industry, competed to develop the best possible classification system to diagnose COVID-19 by analyzing acoustic biomarkers. The biomarkers consisted mainly of cough recording features. The challenge used data from a public dataset named Coswara. The team’s tracked their results on a digital leaderboard that operated in near-real-time.
Although not even the highest-scoring team reached the criteria at the end of the challenge, a fusion of the four top-scoring teams achieved the WHO’s benchmark by a relatively small margin, with a sensitivity of 70.73%.
This AI vs COVID-19 Story Is Not Over Yet
The DiCOVA challenge demonstrated the concept of using machine learning algorithms to diagnose acoustic biomarkers to result in a practical point-of-care diagnostic tool. If validated as a diagnostic tool, such technology would indeed be a game-changer in humanity’s fight against the pandemic and any other respiratory disease for that matter.
The researchers who participated in the DiCOVA challenge have presented their works at the DiCOVA Special Session at the Interspeech 2021 conference, which took place in Brno, Czech Republic, from 30 August to 3 September 2021. Here is a page that summarizes their results.
The possibility that any person with a smartphone can quickly and inexpensively tell if their cough sounds like they have COVID-19 or not is possibly one of the noblest uses of the smartphone ever devised. With this technology, the citizen will be finally able to answer the question: Is my cough COVID-19? Until then, you could download a free smartphone application for cough tracking and see the dynamics of your cough, share them with your physician.
- Meister JA, Nguyen KA, Luo Z. Audio feature ranking for sound-based COVID-19 patient detection. arXiv:210407128 [cs, eess]. Published online April 14, 2021. Accessed August 3, 2021. http://arxiv.org/abs/2104.07128
- Pahar M, Klopper M, Warren R, Niesler T. COVID-19 cough classification using machine learning and global smartphone recordings. Computers in Biology and Medicine. 2021;135:104572. doi:10.1016/j.compbiomed.2021.104572
- Sadiq F, Masood K, Muhammad A, Riaz Khan A. Screening of COVID-19 using Cough Audio Frequencies. IJATCSE. 2021;10(3):1851-1861. doi:10.30534/ijatcse/2021/511032021
- Sharma N, Krishnan P, Kumar R, et al. Coswara — A Database of Breathing, Cough, and Voice Sounds for COVID-19 Diagnosis. In: Interspeech 2020. ISCA; 2020:4811-4815. doi:10.21437/Interspeech.2020-2768
- Sharma NK, Muguli A, Krishnan P, Kumar R, Chetupalli SR, Ganapathy S. Towards sound based testing of COVID-19 — Summary of the first Diagnostics of COVID-19 using Acoustics (DiCOVA) Challenge. arXiv:210610997 [cs, eess]. Published online June 21, 2021. Accessed August 3, 2021. http://arxiv.org/abs/2106.10997
- Topol EJ. Is my cough COVID-19? Lancet. 2020;396(10266):1874. doi:10.1016/S0140-6736(20)32589-7
- WHO Target Product Profiles for COVID-19 Diagnostics. World Health Organization; 2020