SHOP

News

South Korean researchers develop AI that reads lung diseases with breathing sounds

Everyone’s Lab Succeeds in AI Study to Read Abnormalities with Pulmonary Aesthetic

Stethoscope examination with different interpretations for different medical personnel, AI is expected to improve accuracy

There is a possibility that artificial intelligence(AI) will be created to listen to patients’ breathing sounds and read whether they have lung diseases.

At the world’s most prestigious voice AI society “INTERSPEECH 2023,” which will be held in Ireland in August, a research paper that analyzes the breathing sounds of patients to read lung diseases was listed. This paper is about a study that classifies major signs of lung disease such as Crackle and Wheeze to find the possibility of disease. Unlike AI models that analyze lung diseases by reading CT or X-ray images, there are not many technologies to find diseases through sound yet, so it is evaluated as a study that will bring a new approach to the medical reading field.

In general, lung sound diagnosis is conducted by a doctor using a stethoscope to directly listen to the patient’s breathing sound and find out if there is an abnormality. However, this method had the disadvantage of being subjective to interpretation, no matter how much expertise a medical person had, and its accuracy was reduced depending on the situation. There was also a risk that doctors could be infected during the treatment process.

This study was conducted to compensate for these shortcomings. The researchers succeeded in predicting abnormal lung auscultation sounds such as blisters and thousand tones using transformer models mainly used in commercialized AI models such as ChatGPT and AI technology using hierarchical labels of lung auscultation. Since the patient’s sound data is analyzed with AI, the possibility of infection can be reduced as doctors do not have to contact the patient directly with a stethoscope, and objective results can be produced with AI that has learned a lot of data. As a result of this achievement, the paper was published in INTERSPEECH, the world’s most prestigious academic society in the field of voice AI.

What is more meaningful about this achievement is that domestic office workers working for different companies and college students who are devoted to their studies worked together. The research was conducted at the AI community company ‘Everyone’s Research Institute’. Everyone’s Research Institute is a community company where several people gather to freely study common topics of interest regardless of their workplace, affiliation, or age. If you post a study on a subject, interested researchers are free to participate. Participants are diverse, including office workers, professors, and students. This study also produced results by freely gathering people interested in lung auscultation analysis and prediction AI models.

Nine researchers participated in the study, including Bae Sang-min and Kim Sung-nyeon, KAIST’s Kim Jae-chul, Kyungpook National University’s Kim Joon-woo, Dongguk University’s graduate students, Smartsound’s Cho Won-yang and Baek Hye-rim, and ‘Everyone’s Research Institute’s Son So-yeon, Ha Chang-wan and Tae Kyung-pil.

Lee Jung-ho, CEO of Smartsound, said, “The sound of the multiple and thousand sounds is the representative sound of major lung diseases. Through this study, we proved the possibility that AI can play an auxiliary role in early prevention and diagnosis of lung diseases.” He also expected, “This study will contribute to the early detection of lung diseases in the future.”

“This study is the first case of applying a pre-learned model for large visual and audio datasets to respiratory sound classification tasks,” said a researcher who participated in the study. “Currently, the accuracy of lung sound classification still exceeds low performance, but we expect the study to improve lung sound classification performance.”

[Source] : THE AI(https://www.newstheai.com)

[Keyword] : Breathing sound, lung disease, AI reading, AI stethoscope, interspeech 2023, everyone’s laboratory, Crackle, Wheeze