Stroke AI: Real-Time Detection. Predictive Intelligence.

Fast, accurate, consistent AI triage for acute stroke response and transfer decisions.

FDA 510(k)K240353Korea MFDSapprovedISO 27001ISO 13485HIPAA· BAA available

Where do you practice?

Every setting has a different bottleneck. Select yours.

Hyper Insight™ – ICH

AI triage for intracranial hemorrhage on non-contrast head CT

FDA 510(k) K240353ACUTE STROKE READY HOSPITALS

There is no radiologist down the hall.

At an acute stroke-ready hospital, Hyper Insight–ICH triages suspected brain hemorrhage on non-contrast CT within seconds and delivers a mobile alert with image previews to the off-site radiologist. The radiologist can quickly review the scan and alert the care team with the findings. As a result, treating physicians can deliver critical care within minutes of the scan, such as initiating blood-pressure control, reversing blood thinners, and setting a transfer in motion or avoiding an unnecessary one, while the formal read is underway.

0.9864AUC

FDA pivotal trial across 13 US sites

95.45%Sensitivity

Every ICH subtype above 90%

98.47%Specificity

Less than 2% are false alerts

16.39 ± 5.46Seconds

Average time from processing to alert

For the Radiologist

Faster decision, initial treatment response, and transfer decision.

The Alert Lands before the Report

Notification 16.39 ± 5.46 seconds after scan processing (FDA pivotal, 13 US sites), reaching the radiologist’s phone with a triage result and preview image while the study is still being read. The team acts; the formal read confirms.

Built for Radiologists

Reading with AI, accuracy rose to 97.03% from 94.71%. Reduced blind spots: all five subtypes above 90% sensitivity: EDH 100% · IVH 97.7% · IPH 97.3% · SAH 97.7% · SDH 94.7%, and every volume stratum, including under 1 mL (90.0%).

Largest External Validation

49,841 patients · 1,855,465 slices · AUC 0.992 · Sensitivity 94.4% · Specificity 98.2% (npj Digital Medicine, 2023).

For the Hospital

Avoid unnecessary transfers. Earn and save more.

Start at No Cost

A 6-month pilot access program (with extensions available): deployment, PACS integration and licensing covered. Regular pricing is up to 70% less than current in-market solutions.

Avoided Unnecessary Transfers

Each avoided transfer saves the transport bill and keeps the admission, and its revenue, at the local hospital. A DRG-paid hospital that keeps a patient, when clinically appropriate, bills the whole lump sum itself.

Fewer ICU Days, Shorter Stays

A bleed treated minutes earlier tends to grow less, meaning fewer ICU days and a shorter stay. Every avoided day is money the hospital saves.

The evidence behind the alert

Workflow-Actionable Alerts

Independent testing: AUPRC 0.97, precision 0.98, specificity 1.00; roughly 2% of alerts false, against about one in three for two of the four solutions evaluated.

When It Alerts, It Is Credible

PPV 98.44% and NPV 95.54% in the FDA pivotal trial, and a separate 9-reader study demonstrated reader-uplift. The alert carries enough weight to act on before the specialist read.

Independent Head-to-Head

436 real-world emergency NCCTs, read blind by neuroradiologists with no vendor involvement — first among four Korean-developed ICH solutions (two also FDA-cleared) on every confirmatory and calibration metric.

Hyper Insight™ Stroke AI

One platform across the stroke pathway — hemorrhage, infarction, aneurysm, perfusion and 3D vascular imaging, on CT and MR. One solution is FDA-cleared today; seven more are in the FDA process. Select any solution for its description, capabilities, and performance evidence.

Hyper Insight™ – ICH

Scan / Clicke-Label

NCCTLaunchedFDA 510(k) cleared

Triages suspected brain hemorrhage within seconds on non-contrast CT and alerts clinicians to potential findings.

Capabilities

Worklist prioritization
Mobile notifications
Instant specialist notification — auto alerts for suspected ICH
Seamless team collaboration — smart messaging & invitations
Mobile DICOM viewer — access anytime, anywhere

NCCTICH Binary TriageMobile Alert & DICOM Viewer

Click to zoom

Performance & evidence

US pivotal — patient-wise. 13 US clinical sites · AUC 0.9864 · Sensitivity 95.45% · Specificity 98.47% · PPV 98.44% · NPV 95.54% · Mean processing time 16.39 ± 5.46 seconds · Subtype performance 94.7%–100%. The trial included diverse racial demographics (Alaskan Native, Asian, Black, White and others).

What it means clinically

The only ICH AI with dual regulatory clearance (FDA + MFDS), a NECA innovative technology designation, peer-reviewed external validation across 49,841 patients published in npj Digital Medicine, a 16.39-second mean processing time, and proven reader performance improvement — across subtypes, volumes and scanner families.
Reduces the performance gap for non-specialist physicians in under-resourced settings by +3.43% sensitivity.
Independently validated best-in-class ICH triage: #1 among four Korean-developed (two also FDA-cleared) ICH solutions on every confirmatory and calibration metric (AUPRC 0.97, precision 0.98); roughly 2% of alerts are false versus about 1 in 3 for comparators.
Independently validated workflow improvement: hemorrhage read ~5× earlier (median reading order 7.25→1.5), early diagnosis 49%→76%, with reader accuracy, sensitivity and specificity statistically unchanged.
Sensitivity 98.1% for the 1 to <5 mL volume stratum and 90.0% for bleeds under 1 mL — every volume stratum above 90%.

Clinical Evidence

Regulatory trials, peer-reviewed publications, and independent third-party evaluation.

FDA

Hyper Insight–ICH (SKAIICH-02) — FDA 510(k) validation

US pivotal trial across 13 US clinical sites, N=394. AUC 0.9864 · Sensitivity 95.45% · Specificity 98.47% · PPV 98.44% · NPV 95.54% · Mean processing time 16.39 ± 5.46 s. Subtype performance 94.7%–100%.

Basis of FDA 510(k) K240353 · racially diverse cohort

Peer-reviewed

External validation — Yun et al., npj Digital Medicine 2023

Six institutions, 49,841 patients · 1,855,465 slices. Accuracy 0.977 · Sensitivity 0.944 · Specificity 0.982 · AUC 0.992. Validated across GE, Philips, Siemens and Toshiba scanner families.

The largest external validation disclosed in the ICH AI category

Independent

Independent 4-way ICH AI comparison — Kim et al. 2026 (Diagn Interv Radiol)

436 real-world emergency NCCTs read blind by neuroradiologists with no vendor involvement. Hyper Insight–ICH ranked first on every confirmatory and calibration metric — AUPRC 0.97, precision 0.98, specificity 1.00. Roughly 2% of alerts false versus about 1 in 3 for two comparators.

Single-center retrospective; detection-sensitivity differences not statistically significant

Independent

Real-world workflow & reading-order impact — Kim et al. 2025 (Diagn Interv Radiol)

Before/after emergency-department study. Hemorrhage cases read ~5× earlier (median reading order 7.25→1.50); early-diagnosis rate 49%→76%, with reader accuracy, sensitivity and specificity statistically unchanged.

Workflow evidence, not a detection-accuracy claim

MFDS

SKAIICH-01 — MFDS clinical validation & multi-reader study

Blinded multi-reader crossover study, 296 cases, 9 readers across three expertise tiers. Accuracy 94.71%→97.03% (p<0.0001); non-specialist physicians gained the most (+3.43% sensitivity, p=0.0274).

Basis of the reader-uplift figures cited across this site

Peer-reviewed

Brain infarction reader study — Scientific Reports 2026 (Yun et al.)

Multi-center randomized crossover superiority study, N=917 (367 infarction, 550 normal), 9 readers. AI assistance improved accuracy +3.60%, sensitivity +4.26%, specificity +3.15%; non-radiologist physicians gained most (+5.38% accuracy).

Nature-group publication · NCCT-based acute ischemic stroke detection

MFDS

SKAICA-01 — brain aneurysm, MFDS pivotal

Prospective multicenter randomized crossover superiority design, N=872. Standalone AUC 0.8575; size-stratified sensitivity from <3 mm (74.57%) to ≥10 mm (100%); non-radiology physicians gained +22.28pp sensitivity with AI.

Two academic centers · primary and secondary endpoints met

Government HTA

NECA innovative medical technology designation (2024)

Korean national health technology assessment designation for innovative medical technology — a regulatory quality signal distinct from device clearance.

Ministry of Health & Welfare designation

What Clinicians Say

Trusted by Medical Professionals

It will be very helpful for physicians who lack expertise but must make initial judgments in emergency cases.

Non-Radiology Physician

Korean Academic University Hospital

Successfully detected small findings and correctly identified negatives as negative. The AI software helps clinicians who have difficulty distinguishing beam hardening artifacts.

Neuroradiologist

Korean Academic University Hospital

When detecting small SAH (subarachnoid hemorrhage), the AI assistance helps accurately review many images in a short amount of time.

Radiologist

Korean Academic University Hospital

What impressed me most was the software's ability to distinguish real hemorrhage from look-alikes. Partial volume artifacts — where bone and brain tissue overlap — can easily mimic a small subarachnoid hemorrhage on CT, and normal intracranial calcifications appear hyperdense enough to fool most deep learning algorithms into flagging them as bleeds. These are areas where even experienced neuroradiologists have to slow down and look twice. PurpleAI's system correctly identified both as non-hemorrhage, avoiding false positives that would otherwise erode a clinician's trust in the tool.

Dr. Yoo, MD, PhD

Department of Radiology, Korea University Anam Hospital

The system consistently detected small-volume hemorrhages that are among the hardest findings to catch on non-contrast CT — thin subdural collections along the anterior falx cerebri, trace subarachnoid blood in a single sulcus. These are the cases that get missed on overnight reads or by physicians who don't specialize in neuroimaging. The AI flagged them accurately and localized them with color mapping, giving the reader a clear starting point instead of having to hunt through every slice.

Dr. Kim, MD

Department of Radiology, Eunpyeong St. Mary's Hospital, College of Medicine, The Catholic University of Korea

Testimonials from medical end users at Korean academic university hospitals

See It In Action

Product Demo

ICH DetectedNow

Patient ID: 12847 · NCCT

Urgent Alert2m

Brain aneurysm suspected

Analysis Complete5m

No abnormalities detected

Cross-Platform

Optimized for all devices

📱

iOS

✓

🤖

Android

✓

Dr. Kim

ICH case needs immediate review

2:34 PM

On it. Checking now

2:35 PM

Dr. Park

I'll join the consultation

2:36 PM

Type a message...

Mobile DICOM Viewer

DICOM Access Anytime, Anywhere

Your stroke care team stays connected with real-time mobile notifications and a full-featured DICOM viewer.

Instant Specialist Notification

Auto Alerts for Suspected ICH

Seamless Team Collaboration

Smart Messaging and Invitations for Instant Case Review

Mobile DICOM Viewing

Full Diagnostic-Quality Images Review

Predictive Intelligence

StrokeShieldAI™

"Advance from rapid stroke detection toward predictive intelligence that anticipates patient trajectories through multimodal data, transforming acute stroke care from reactive response to proactive decision-making."