VUNO, a medical artificial intelligence (AI) company, announced on Monday that its research demonstrating significant improvements in the algorithm performance of its AI-based heart failure screening device, VUNO Med®-DeepECG® LVSD (DeepECG LVSD), was recently published in the prestigious international journal <em>JMIR Medical Informatics </em>(Impact Factor 3.8).

The study focused on developing an advanced deep learning model that analyzes electrocardiogram (ECG) data to detect left ventricular systolic dysfunction (LVSD), a key precursor to heart failure, with greater precision. Researchers also evaluated the algorithm's performance and clinical applicability through rigorous internal and external validations.

The groundbreaking study utilized a massive dataset of approximately 260,000 patients to enhance diagnostic accuracy through an innovative recalibration technique.

Early detection of LVSD can significantly improve patient outcomes when treated promptly. However, the current standard diagnostic method, echocardiography, has limitations for large-scale screenings due to its high costs and limited equipment accessibility.

To address these challenges, various AI models have been developed. However, reports have consistently indicated that diagnostic performance tends to decline in patient groups with comorbid conditions.

The VUNO research team developed their algorithm using a sophisticated three-step learning strategy based on data from approximately 260,000 patients. They first pre-trained the model using a large dataset of ECGs, then fine-tuned it to detect LVSD using ECG data collected within two weeks of echocardiography (TTE) tests.

The team introduced a novel algorithm for LVSD screening through recalibration, which incorporated past test results and previous model predictions. They used data from Hallym University Sacred Heart Hospital for internal validation and conducted external validation with data from Yonsei University Wonju Severance Christian Hospital.

The model achieved an impressive AUROC of 0.956, demonstrating high performance regardless of comorbid conditions.

The study confirmed that the DeepECG LVSD, after pre-training and fine-tuning, achieved an AUROC of 0.953 in internal validation (0.947 externally). Additionally, the recalibrated model, which reflected patients' past test results, recorded an AUROC of 0.956 in internal validation (0.940 externally), showing improved performance compared to the pre-recalibration model (0.945 internal, 0.910 external).

This performance enhancement was consistent across all demographic groups, including those traditionally challenging for AI, such as patients with atrial fibrillation. Notably, the specificity for patients with atrial fibrillation improved dramatically from 0.518 to 0.900, potentially reducing the burden of unnecessary additional tests due to false-positive diagnoses.

Joo Seong-hoon, Chief Technology Officer (CTO) of VUNO, stated that the algorithm that incorporates patients' past test histories has significantly improved the screening performance of DeepECG LVSD, regardless of comorbid conditions. He emphasized that VUNO will continue to refine this technology to contribute to the early diagnosis and treatment of high-risk heart failure patients.

In other news, VUNO reported an annual revenue of 34.8 billion KRW (approximately 23.7 million USD) last year, marking a 35% increase from the previous year's 25.9 billion KRW (about 17.6 million USD).

The company's operating loss decreased to 4.9 billion KRW (roughly 3.3 million USD), down about 60% from the previous year's loss of 12.4 billion KRW (approximately 8.4 million USD). This achievement came without significant changes in operating expenses compared to last year. The operating costs for 2025 are projected to be 39.8 billion KRW (about 27.1 million), a modest increase of about 4% from 38.3 billion KRW (approximately 26 million USD) the previous year.

VUNO's flagship product, the AI-based cardiac arrest prediction device VUNO Med®-DeepCARS® (DeepCARS), generated sales of 25.7 billion KRW (about 17.5 million USD) last year, an increase of approximately 18% (4 billion KRW or 2.7 million USD) compared to the previous year. The AI-based ECG measurement device HATIV also continued its upward trend in sales, reaching 1.9 billion KRW (approximately 1.3 million USD) last year, following the launch of the kiosk-type HATIV K30.

VUNO, a medical artificial intelligence (AI) company, announced on Monday that its researc …

VUNO Study Shows Improved LVSD Screening Performance for AI-Based ECG Algorithm