AI-Assisted Digital Auscultation
Record heart and lung sounds.
Let AI classify the findings.
JABES analyzes heart and lung sounds captured with a digital electronic stethoscope using artificial intelligence, assisting clinicians in auscultation review. Built on stethoscope technology first developed in South Korea in 2003, it automatically classifies three heart-sound and five lung-sound categories.
About JABES
A Pioneer in Digital Auscultation
JABES was the first digital electronic stethoscope developed and mass-produced in South Korea, in August 2003. It was later deployed in telemedicine (U-health) programs in Gangwon Province and Sinan County, South Jeolla Province, accumulating real-world clinical use.
A 7-level volume control amplifies auscultated sound up to 18 times, paired with on-the-spot AI analysis of the recording — reducing the performance gap that depends on an examiner's experience.
A Screening Aid, Not a Diagnosis
JABES is a screening aid that supports a clinician's auscultation review; it does not diagnose or confirm disease. Results are reference findings only — the treating clinician always makes the final diagnostic and treatment decisions.
It classifies three heart-sound categories (normal, systolic murmur, diastolic murmur) and five lung-sound categories (normal, crackle, rhonchi, stridor, wheeze).
How It Works
Choose a Position
Select from four heart positions (aortic, pulmonic, tricuspid, mitral) or four anterior/posterior lung positions.
Record
Capture a 9-second recording with the digital stethoscope. A live spectrogram and waveform confirm signal quality, and silence or noise is flagged automatically.
Review the AI Result
The classification appears immediately after recording. Results are saved per patient and can be exported as a PDF report.
Technology Validation
The classification model behind JABES is grounded in a cross-source, cross-device generalization study conducted jointly by the Department of Industrial and Systems Engineering at Dongguk University and Chung-Ang University Hospital. The research compared a lightweight convolutional neural network against large AudioSet-pretrained audio transformers (AST, BEATs), and evaluated both label-noise cleaning and generalization across different clinical sites and recording devices. JABES continues to be refined based on these findings.
Key Features
Live Signal Visualization
Spectrogram, amplitude, power-analysis, and level-meter views let you confirm signal quality while recording.
Automatic Quality Gating
Silence, noise, and low-confidence predictions are detected automatically, so non-auscultation recordings are not saved as results.
File Upload Analysis
Beyond live recording, existing WAV or MP3 files can be uploaded and run through the same analysis pipeline.
PDF Report Generation
Generates a report combining patient and examiner information, classification results, and spectrograms.
Lightweight On-Device AI
A 0.24M-parameter model runs on CPU, delivering results quickly without a dedicated GPU server.
Account-Based Access Control
Only authenticated users can access patient data and analysis results.
Frequently Asked Questions
Does JABES diagnose disease?
No. JABES is a screening aid that supports a clinician’s auscultation review; it does not diagnose or confirm disease. Results are reference findings only, and the treating clinician always makes the final diagnostic and treatment decisions.
What sounds can it classify?
Three heart-sound categories (normal, systolic murmur, diastolic murmur) and five lung-sound categories (normal, crackle, rhonchi, stridor, wheeze).
How accurate is it?
Accuracy continues to improve through ongoing research and validation. As JABES is a screening aid rather than a diagnostic tool, results should always be used as a reference alongside a clinician’s own clinical judgment.
Can it be used without the stethoscope hardware?
Yes. You can record directly with the JABES digital stethoscope, or upload an existing WAV or MP3 audio file for the same analysis.
Is recorded data stored securely?
Recordings and analysis results are processed and stored behind account-based authentication; only logged-in, authorized users can access them.