Students manufacturing devices diagnose respiratory diseases with AI

Equipment, which is researched and manufactured by the students of the University of Electronics and Electronics, Hanoi University of Science and Technology, to support the diagnosis of respiratory diseases based on breathing and processing analysis with FPGA combining artificial intelligence ( WHO). This device has the ability to record and analyze respiratory sound through electronic stethoscope with integrated sensor. The collected data is transferred to the artificial intelligence processing software (AI) to identify abnormal characteristics in sound, helping to detect diseases early such as pneumonia, asthma and chronic obstructive pulmonary disease Calculate (COPD).

The device consists of 2 main modules: the data collection using electronic stethoscope (Welch Allyn Meditron Electronic Stethoscope), used as the doctor ordered the patient’s examination tube. Can be placed behind or in front of the patient’s chest. Biomedical data after obtained is transferred to the processing module to compare with the respiratory sound of pneumonia, asthma or COPD.

The processing is the FPGA module (Board is the type developed based on AMD Zynq Ultrascale+™ MPSOC Development Board). After the interference filtering and data processing money, the AI ​​model integrated on this module conducts the classification, diagnosis and return of the results.

In addition, the device also has some other components such as display screen, keyboard, mouse that supports operation.

Le Trung Kien, the leader of the research team, said that in the current medical examination and treatment, the diagnosis depends on the experience of doctors and specialized medical equipment. While diagnosing visual images, respiratory audio analysis is subjective, depending on the headset and experience of the doctor. Therefore, “the group wants to develop a system that can support and analyze respiratory sounds more objectively,” Trung Kien said.

The project started in 2023. Initially, the group participated in a scientific topic on medical image processing. When this topic ends, the group shifts to the direction of respiratory sound and AI to detect and classify respiratory pathology.

Unlike medical image data that is available in large quantities, respiratory sound data is very limited. The group must collect data from international scientific storage, including the breathing of healthy people, people with COPD, asthma … After that, the application of signal processing algorithms and AI Analysis of breathing, programming disease identification models and comparing results with original data to evaluate accuracy. The data must be exploited from many sources, conducting processing money to ensure the accuracy and uniformity. “This process has many difficulties, especially the collection of data representing Vietnamese people due to the difference between nations,” said Nguyen Anh Phuong, a member of the research team.

After building AI model, students continue to develop hardware equipment to be able to apply practically. One of the big challenges is to search for the appropriate calculation platform. The group tested many solutions, from conventional microcontrollers to embedded computers like Raspberry Pi, but found that they were not strong enough to process audio data in real time. Finally, the group decided to use FPGA (Field-Programmable Gate Array), a micro-circuit capable of strong parallel processing, helping to increase the calculation speed. “FPGA allows to optimize AI algorithms, helping the device processing audio processing quickly without hardware too bulky. In addition, FPGA supports many open source programming languages, making it easy for the group Customized without copyright violations, “Trung Kien said.

Research products of the group won the first prize in the Council of Technology and Biomedical Equipment, the Contest for Scientific and Creative Student Studies of Hanoi University of Science and Technology in 2024.

Assoc.Prof.Dr. This is a new research direction, which can be deployed into compact, useful products for users to monitor their own health at home. If further research is continued, the product can be commercialized and applied in practice.

The goal of the group is to continue perfecting and developing equipment into respiratory health monitoring tools at home, similar to blood pressure monitors. In addition to breathing analysis, the group also intends to integrate the heart rate, blood pressure, spO2 to assess the overall respiratory health. Users will receive an early warning about the risk of disease, which can take the initiative to go to health facilities when necessary.