Researchers from the Warsaw University of Technology dealing with artificial intelligence have developed a tool that helps in analysing the acoustic activity of the gastrointestinal tract. The solution allows for non-invasive and inexpensive diagnosis of some diseases of the digestive system.

Bowel sounds at night

A team of researchers led by Robert Nowak, PhD, DSc, Eng., Associate Professor from the Faculty of Electronics and Information Technology - based their research on recordings made by doctors from Poznan University of Medical Sciences. With special, sensitive contact microphones attached to the abdomen, bowel sounds of 20 patients were recorded. Recordings were made at night — at the time when there is minimal additional interference. As a result, high-quality digital sound materials were obtained.

Selection

In order to obtain appropriate audio samples, filtration needs to be performed to remove interference and sounds of heartbeat, breathing and blood noise in the veins. Following that, doctors mark the fragments in the recordings where intestinal sounds are heard. It calls for a considerable amount of work, but a set of examples confirmed by a doctor is indispensible for the algorithm to have a high quality foundation, and consequently — very high efficiency.

- ‘The audio signal, represented as a sequence of samples taken at an appropriate frequency and of adequate quality, is cleared of noise and converted into a spectrogram. The spectrogram is a convenient way of recording the signal from the point of view of analysis. Then the analysis program divides the signal into frames with a length of 20 milliseconds, which corresponds to 441 samples’ — says Robert Nowak, PhD, DSc, Eng., Associate Professor, head of Artificial Intelligence Division at Faculty of Electronics and Information Technology — Then the algorithm makes a binary decision and determines the true/false logical value based on the frame descriptions — indicating whether the frame has intestinal sound. Such analysed and statistically organised material about all the frames from the recording goes to the doctor.

Effectively and stress-free

The effect of a trained algorithm is an excellent tool supporting diagnostics for physicians. The method is non-invasive — the recording analysed by artificial intelligence and the knowledge of a doctor is all that is required. For patients, this means a convenient alternative to troublesome examinations, such as colonoscopy. For the health service — savings. The tests indicate a very high quality (accuracy, specificity and sensitivity) of the solution prepared by AI specialists from the University of Technology.

More about the analysis of intestinal sounds using deep neural networks can be found in the article published in a monthly - Sensors.

To all those interested in the subject of AI and looking for a study organising knowledge regarding, among others, machine learning, artificial neural networks and ethical aspects, we recommend the latest study by Mirosław Muraszkiewicz, PhD, DSc, Professor, and Robert Nowak, PhD, DSc, Eng., Associate Professor — "Artificial intelligence for engineers. General methods”.