THE ART CENTER

Open to all, this is the art center that houses the world with performances and exhibitions.

AIAI-Powered Cardiac Arrest Prediction

AI

AI-Powered Cardiac Arrest Prediction

Product Brochure · 2026

DIGIRAY

Real-time ECG analysis that predicts cardiac arrest before it happens — protecting patients, empowering clinicians, and saving lives.

80%+

Accuracy

83%

Sensitivity

0.88

AUC Score

204K+

ECG Samples

Company

DIGIRAY Medical AI

Technology

CNN-LSTM Deep Learning

Clinical Partner

Seoul National University Hospital

Regulatory

MFDS SaMD Class II (In Prep)

Stage

Pre-commercial · Series A

The Problem

Cardiac arrest kills without warning

Existing hospital monitors react after events occur — leaving clinicians without actionable foresight.

17.9M

Cardiovascular deaths/year

World's #1 cause of death (WHO)

4–6 min

To irreversible brain damage

Every second of delay matters

~60%

Show prior ECG warning signs

Signals missed without AI

$70K+

Cost per in-hospital arrest

Economic burden on health systems

No widely deployed AI solution reliably predicts cardiac arrest before it happens.

Our Solution

Predict. Alert. Save lives.

DIGIRAY's AI monitors ECG in real time and alerts clinical staff before arrest occurs.

Acquire — ECG signal from standard bedside monitor (125Hz, no new hardware needed)

Preprocess — Automated noise removal via High-pass, Notch, and Wavelet filters

Extract — CNN identifies cardiac waveform features in 10-second segments

Classify — LSTM detects high-risk temporal patterns in sequential ECG data

Alert — Immediate early warning sent to clinical staff before arrest occurs

Non-invasive: Integrates with existing ECG infrastructure — no new hardware investment required for hospitals.

Core Technology

CNN-LSTM: A two-stage deep learning architecture

Purpose-built for clinical ECG analysis — and designed to scale to full multimodal vital sign monitoring.

Feature Extraction

Stage 1 · CNN

Conv1D layers + MaxPooling extract morphological features from raw ECG signal. Batch Normalization stabilizes learning across diverse patient profiles.

Sequence & Classification

Stage 2 · LSTM

Bidirectional LSTM (128 + 64 units) with Dropout (0.3) learns temporal risk patterns → Dense layer → Binary output: 0 = Normal, 1 = High-Risk.

Future expansion: Multimodal AI

Architecture already supports integration of blood pressure, SpO₂, respiratory rate, and additional vitals for a comprehensive early warning platform.

Data & Training Specifications

SNUH Clinical Data	27,499 samples · 51 patients
MIT-BIH Open Dataset	4,140 samples · 47 patients
AI Hub — Normal ECG	94,672 samples
AI Hub — High-Risk ECG	109,612 samples
Total Labeled Data	204,000+ ECG segments
Segment Length	10 seconds @ 125Hz
Train / Test Split	80% / 20%
GPU / Framework	NVIDIA A100 · TensorFlow 2.18
Optimizer / Loss	Adam · BinaryCrossEntropy
Training Config	Batch 128 · Epoch 200

Transfer Learning Strategy Pre-trained on large-scale AI Hub data → fine-tuned on SNUH clinical data via transfer learning → validated on held-out SNUH test set.

Clinical Performance

Exceeding target benchmarks — validated on real patient data

Four years of iterative development, with every model generation improving on validated clinical metrics.

0.80

Accuracy

0.83

Sensitivity

0.84

F1 Score

0.88

AUC Score

Development Stage	Accuracy	Sensitivity	Specificity	F1 Score	AUC
Year 2 — Initial dataset (SNUH)	—	0.61	0.43	—	0.56
Year 3 — CPR event window + open data	0.62	0.73	0.54	0.63	0.74
Year 3 — 15 min pre-arrest window	0.77	0.78	0.75	0.76	0.79
Year 4 — AI Hub internal validation	0.89	0.83	0.96	0.89	0.97
★ Year 4 — Transfer learning on SNUH (Final)	0.80	0.83	0.75	0.84	0.88

What Sensitivity (0.83) means

83% of true high-risk cardiac events are correctly identified — minimizing missed alerts that could cost patient lives.

What AUC (0.88) means

Strong discriminative power between high-risk and normal patients across all classification thresholds — clinically robust.

* Final model validated on held-out Seoul National University Hospital dataset via transfer learning.

Market & Applications

Where DIGIRAY deploys

From intensive care to remote monitoring — a platform built for the full spectrum of cardiac care settings.

ICU / CCU Monitoring

Continuous real-time cardiac risk assessment for critically ill patients in intensive and coronary care units.

Remote Patient Monitoring

Cloud-based ECG analysis for post-discharge patients and telemedicine platforms, reducing readmission risk.

OEM Device Integration

White-label AI engine embedded into patient monitor hardware by medical device manufacturers globally.

Emergency Medical Services

Pre-hospital triage support for paramedics — early identification of high-risk patients during transport.

Cardiac Rehab Centers

Ongoing monitoring for high-risk cardiac patients during rehabilitation and step-down care programs.

Clinical Research

De-identified cardiac risk analytics for pharmaceutical trials, epidemiological research, and risk modeling.

Global AI Healthcare

$45.2B

2026E · CAGR ~45%

Cardiac Monitoring AI

$8.1B

2025 · Fastest-growing

Korea Medical AI

$1.2B

2025 · Regulatory support

Why DIGIRAY

Clinically proven. Commercially ready. Mission-driven.

Four years of R&D, real clinical data, and a scalable platform built to protect human lives at scale.

Clinical validation complete

Trained and tested on real SNUH patient data — not just open datasets. Results are reproducible and auditable.

204,000+ labeled ECG samples

A proprietary dataset representing a significant and defensible competitive moat in the cardiac AI space.

Non-invasive & infrastructure-friendly

Works with standard ECG leads and existing hospital monitoring infrastructure — zero new hardware required.

Expandable multimodal platform

Architecture designed from the ground up to incorporate BP, SpO₂, and additional vitals — not an afterthought.

First-mover in Korean medical AI

Government digital health initiatives and regulatory fast-tracking create a favorable commercial environment.

Regulatory Pathway

MFDS (Korea) SaMD Class II · Filing in prep

FDA (USA) De Novo / 510(k) pathway

CE Mark (EU) MDR classification under review

Evidence base University hospital validation ✓

Partnership & Integration

API / SaaS EMR & monitoring system integration

OEM White-label for device manufacturers

Research De-identified data licensing