Academics from various fields at Hitit University have developed artificial intelligence-powered software that can diagnose six diseases, as well as healthy coughs, from cough sounds.
Two biostatisticians from the university, a computer engineer, a pulmonologist, and a gastroenterologist, have begun work on software that will enable the diagnosis of respiratory diseases by converting human voices into mathematical data and images.
In over a year of research, cough sounds collected from 75 people were converted into mathematical data and visual graphics. The resulting data was taught to the AI using deep learning techniques, teaching the characteristics of cough sounds caused by conditions such as COPD, asthma, bronchitis, upper respiratory tract diseases, pneumonia, and reflux.
The software project, which can also detect control coughs that do not contain a disease and diagnose illnesses with a success rate of over 91 percent, has been awarded support by TÜBİTAK (The Scientific and Technological Research Council of Turkey).
The project will be developed with support from TÜBİTAK.
Assoc. Prof. Dr. Emre Demir stated that they obtained numerous parameters related to cough sounds through this study and made them suitable for analysis.
Demir stated that they had taught the obtained data to artificial intelligence, saying, "We can diagnose diseases such as COPD, asthma, bronchitis, upper respiratory tract diseases, pneumonia, and reflux from cough sounds."
Demir noted that they could increase the 91 percent success rate they achieved in the pilot program to over 95 percent with advanced equipment to be procured with the support of TÜBİTAK. He said, "We will obtain three sounds from each patient. We will obtain at least 100 sounds from each disease group. We plan to obtain over 2,000 sound data points. We will convert each sound into 10 different data points. We will have perhaps 5,000 parameters. This will enable us to obtain much more comprehensive data."
Patients will be able to diagnose themselves with their mobile phones.
Demir, emphasizing that they aim to develop mobile software in the third phase of the project, continued:
"Our ultimate goal with the project is to develop a model using deep learning and artificial intelligence from thousands of audio datasets. We believe this model will successfully diagnose respiratory diseases with a 95% success rate. In the third phase of the project, we want to develop a mobile application that will allow patients to directly diagnose their respiratory system using cough sounds recorded on their mobile phones."
Büşra Durak, Ph.D., from the Department of Chest Diseases at the Faculty of Medicine, also emphasized that cough is one of the most common reasons for visits to the chest diseases clinic. She said, "Cough provides us with very important information about diseases. There are different types of coughs, such as dry cough, productive cough, and wheezing cough. In patients presenting with a dry cough, we consider conditions such as asthma and reflux. In patients presenting with a productive cough, we consider conditions such as COPD, bronchitis, and pneumonia."
Durak stated that various laboratory and imaging tests are needed to diagnose these diseases, continuing:
"These tests can be exhausting for patients and create a significant burden on the healthcare system. With this project, we aim to diagnose respiratory diseases solely from cough sounds using sound analysis. We believe this will allow us to diagnose patients more quickly, facilitate faster treatment, and prevent significant burden on the healthcare system."
Gastroenterology specialist İbrahim Durak stated that reflux is the third cause of chronic cough, saying, "We normally have to perform serious procedures like upper gastrointestinal endoscopy and pH metering to diagnose reflux. Being able to diagnose cough sounds will be a great convenience for us. It will reduce both the significant financial burden on the healthcare system and patient volume. I've been excited about the project since I first heard about it."
