A research group including Professor Kenichi Tsujita and Lecturer Masanobu Ishii from the Department of Cardiovascular Medicine, Faculty of Life Sciences, Kumamoto University, along with AMI Inc. (CEO Shimpei Ogawa, Kagoshima City) announced the development of a new technology that uses AI to estimate cardiac status from heart sounds and electrocardiograms. By combining the company's portable device, capable of simultaneously measuring heart sounds and an electrocardiogram, with deep learning, B-type natriuretic peptide (BNP) levels can be estimated in just 8 seconds of measurement. This method is non-invasive and rapid, significantly reducing the physical and temporal burden on patients. Applications are expected for early detection of heart failure and home-based monitoring. The results were published in Circulation Journal on June 17.
Heart failure is a condition in which the heart cannot work sufficiently to pump the blood needed by the body; risk increases with age and is characterized by high rehospitalization and mortality rates. Early detection and appropriate treatment are necessary, but methods of measuring substances such as BNP and NT-proBNP through blood tests are time-consuming and pose a significant burden to patients, presenting challenges.
As a result of this research, a new model called the "eBNP model" was developed to predict the BNP concentration in the blood, an important biomarker, and its performance was evaluated.
The model was trained using data obtained from 1,035 patients who underwent cardiac ultrasound examination. Dialysis patients were excluded from the study. For external validation, data (validation set) from 140 patients selected from 818 patients at different hospitals was used. The study measured how accurately the model could identify patients with high BNP levels (sensitivity) and how accurately it could exclude patients without high BNP levels (specificity).
As a result, the eBNP model demonstrated excellent performance even in the external validation dataset. Particularly high accuracy was achieved in the ability to accurately identify patients with BNP levels of 100 pg/mL or higher. Additionally, in patients with high BMI (obesity), sensitivity was somewhat lower and the ability to correctly predict BNP levels was slightly reduced, but very good results were shown in patients with normal BMI.
Background noise at conversational levels had little impact on model performance, but accuracy decreased slightly when sound became very loud.
These results revealed that the eBNP model has the ability to accurately predict BNP levels and is useful for heart failure diagnosis.
Ishii commented: "This research demonstrated that BNP values can be estimated using a super stethoscope. Even in well-equipped hospitals, BNP test results take about an hour, but using a super stethoscope can avoid the trouble of blood collection and patient discomfort, and I believe it can be utilized in situations such as home medical care where blood tests cannot be performed immediately. In the future, with social implementation in mind, it will be necessary to advance validation assuming practical application scenarios in actual medical settings."
Journal Information
Publication: Circulation Journal
Title: Deep Learning for Cardiac Overload Estimation — Predicting B-Type Natriuretic Peptide (BNP) Levels From Heart Sounds and Electrocardiogram
DOI: 10.1253/circj.CJ-25-0098
This article has been translated by JST with permission from The Science News Ltd. (https://sci-news.co.jp/). Unauthorized reproduction of the article and photographs is prohibited.