Innovative acoustic cough analysis system – new quality in the diagnosis and monitoring of respiratory diseases
A team of Polish scientists has developed a groundbreaking real-time cough detection system that combines the precision of artificial intelligence with the capabilities of mobile acoustic analysis. Thanks to the use of modern machine learning algorithms and deep neural networks, the new solution opens up completely new perspectives in the diagnosis of respiratory diseases – both in the clinical and home environment.
Effectiveness of models in cough detection – accuracy, precision, sensitivity and F1-score.
The system enables automatic detection and classification of cough sounds with high accuracy, while maintaining low computational requirements, which makes it exceptionally practical in portable applications. The key element of the developed technology is the use of three complementary deep learning models. The CNN (Convolutional Neural Network) architecture achieved an accuracy of 83%, effectively functioning in typical acoustic conditions. The ResNet-50 network, designed to work in environments with high levels of sound interference, provided an even higher level of accuracy – 85%. In turn, the MobileNet model, while maintaining the highest efficiency – 89%, is distinguished by its minimal demand for computing power, which allows for implementation in mobile devices and embedded systems.
Optimizing the time of acoustic signal analysis turned out to be one of the key research challenges. In the experiments, the team analyzed the effectiveness of cough detection depending on the length of spectrogram segments. The best results were achieved for time intervals lasting 1.5 seconds, which coincides with the average length of a single cough reflex. 1-second segments also gave satisfactory results, while shorter (0.5 s) and longer (3 and 5 s) turned out to be significantly less effective. These results were used to precisely calibrate the model and increase the efficiency of real-time classification.
The application of the developed system covers a wide range of environments and situations. In healthcare facilities, it can be used for early detection of respiratory infections, including COVID-19, which is crucial for infection control. In nursing homes, this technology can be used to continuously monitor the health of seniors, reacting immediately to the first symptoms of the disease. Potential use in public places such as schools, offices or offices allows for the rapid identification of infection foci and prevention of their spread.
An integral element of the solution is the CoughApp application, which allows for visualization and analysis of data collected by the system. The intuitive interface available both in a web browser and on mobile devices allows for convenient and quick access to results – regardless of place and time.
The developed technology is a significant step towards next-generation acoustic diagnostics. Its advantages include high classification efficiency, low hardware requirements and implementation flexibility. The system has the potential to play a key role in transforming the way we monitor and detect respiratory diseases – offering a fast, non-contact and automatic solution.
Technical details and research results are presented in the original scientific publication: