Data Construction Project for Artificial Intelligence Learning: Chest Auscultation Sound Data

The incidence of cardiovascular diseases worldwide is steadily increasing. According to the report of the American Heart Association, there were 271 million cardiovascular diseases in 1990, and 523 million cases in 2019, about doubling in 30 years. The number of deaths due to cardiovascular disease is also steadily increasing from 12.1 million in 1990 to 18.6 million in 2019.

Physical examination, which is the most basic skill in patient care, consists of inspection, auscultation, percussion, and palpation. Among them, auscultation is the most widely used test in all areas where a stethoscope is used, and it is a basic examination that is essential from primary medical institutions to tertiary medical institutions for non-invasive initial diagnosis in patients complaining of chest symptoms.

However, if a specialist in the field with a lot of experience does not interpret it carefully, it is difficult to make a decision, and the deviation of the test results is large, so a significant number of patients depend on expensive follow-up tests (ultrasound, CT, MRI, etc.) This leads to a vicious cycle of incurring costs and unnecessary treatment.

Recently, with the development of machine learning techniques, computing technologies, and artificial intelligence (AI) based on a lot of data, various learning technologies are applied as tools for disease diagnosis and prognosis prediction in medicine.

Through machine learning-based chest auscultation sound analysis, there is an expectation that disease diagnosis and prognosis prediction will be able to overcome differences and interpretations by examiners. It can be very helpful in preventing overuse of tests and reducing medical costs.

Inclusion criteria

• Adults who are 20 years and older

Exclusion criteria

• Patient refusal
• Uncertain radiographs
• Uncertain tests results