One liner: An AI smartphone algorithm that can screen for respiratory diseases via forced cough recordings. Working on deploying at low-resource clinics in Ghana, with the OKB Hope Foundation and KATH hospital!
In rural Ghana, over 80% of village level clinics don’t have access to diagnostic lab equipment, with the doctor to patient ratio sitting at 1:15,000. Consequently, over 50% of asthma and pneumonia cases, 70% of COPD cases, and 41% of tuberculosis cases are either under-diagnosed or misdiagnosed.
Since September 2021, I have been exploring the use of deep learning in AI to classify voluntary cough sounds from the three most common respiratory diseases in rural Ghana: asthma, pneumonia (including COVID-19 pneumonia), and pulmonary tuberculosis. This presents a low-cost, rapid, and non-contact mobile decision support tool which healthcare workers of limited experience can use to screen patients in high-volume clinics.
Over the last four months I built a successful (~75% accuracy) prototype employing less than 350 recordings in total to train a model that could distinguish between COPD, asthma, COVID-19, and healthy subjects. This work was done in collaboration with Dr. Richard Fletcher who leads MIT’s Mobile Tech Lab.
We found asthma, healthy, and COVID-19 cough data publicly available online, and our COPD cough data was obtained via reaching out to researchers from the University of Southern Florida.
The objective of this prototype was to determine the possibility of using cough sounds to distinguish between more than one disease simultaneously. Previous research into cough-based diagnosis has only involved two categories: disease vs. healthy or disease vs. disease. We realized these existing algorithms are not practical in a clinical setting where there are several lung diseases that need to be considered at once when performing screening.
Below: A prototype algorithm demonstration where we imported 4 different cough recordings, and the model gave a prediction.
https://www.youtube.com/watch?v=QdLET69214A
The asthma, COVID-19, and healthy cough data we used to train our prototype was poor-in quality (being crowdsourced). Although we were able to perform some audio preprocessing, we know there is potential for much higher levels of accuracy with more + better quality data.
In June 2022, we formed a partnership with the OKB Hope Foundation, a nonprofit based in Ghana working to make healthcare more accessible. We have been collaborating with them to collect higher quality asthma and COVID-19 cough recording data as well as new tuberculosis and pneumonia data (this includes non-COVID-19 pneumonia). This data will be collected at **Komfo Anokye Teaching Hospital (KATH) in spring + summer of 2023.** Ultimately, this data will be used to train an improved prototype, one which we will aim to implement at OKB’s clinics to increase prescription of proper treatment.