Purple AI Stroke Diagnosis AI Clinical Study Published in International Journal

Results from Clinical Trials at Seoul National University Hospital and Ajou University Hospital

Detection of Small Lesions Using Only Non-Contrast Brain CT

Greatest Accuracy Improvement Among Non-Radiology Specialists

The clinical research paper on Purple AI's acute stroke diagnosis AI solution has been published in an international academic journal. Unlike previous stroke diagnosis AI systems, which mainly focused on diagnosing large vessel occlusions, Purple AI's solution is evaluated as highly promising for use in emergency medical settings because it can detect even small lesions.

Purple AI Acute Stroke Diagnosis AI Solution. Purple AI

Diagnostic Accuracy Improved by 3.6 Percentage Points with AI Assistance

On June 11, Purple AI announced, "A clinical research paper on our solution has been published in 'Scientific Reports,' an international academic journal from the Nature Publishing Group."

Acute stroke is a representative emergency disease in which rapid diagnosis and treatment during the golden hour are critical to patient outcomes. In emergency rooms, non-contrast brain CT scans, which do not use contrast agents, are often preferred due to their short scan times and high accessibility. However, early stroke lesions may not be clearly visible on imaging, making interpretation challenging even for experienced specialists.

Purple AI's solution detects suspected stroke lesions in non-contrast brain CT images and also provides the ASPECT score (Alberta Stroke Program Early CT Score), an indicator for assessing early changes in stroke. The ASPECT score is used by medical professionals to evaluate the extent and severity of brain damage and to determine the direction of treatment.

Detection of Small Vessel Occlusions and Posterior Circulation Lesions

This study was a multicenter clinical trial involving 917 cases from Seoul National University Hospital, Ajou University Hospital, and others. Nine medical professionals interpreted the same images without AI assistance and then re-interpreted them with AI support. As a result, diagnostic accuracy improved by 3.6 percentage points when AI assistance was used.

Notably, among the group of non-radiology specialists, the improvement in accuracy was the greatest, at 5.4 percentage points. This demonstrates that AI can serve as a diagnostic support tool in situations where specialized personnel are in short supply. The sensitivity of AI-only diagnoses also surpassed the sensitivity of medical professionals assisted by AI.

Purple AI's solution emphasizes its ability to detect even small acute stroke lesions of 5 mL or less using only non-contrast brain CT. Previous stroke diagnosis AI systems were often developed with a focus on large vessel occlusions, which involve relatively large lesions. In contrast, the Purple AI solution is designed to detect strokes caused by blockage of small vessels, cardioembolic strokes resulting from heart-originated clots blocking cerebral vessels, and lesions in the posterior cerebral circulation, which are especially difficult to interpret.

Consistent Performance Verified in 1,400 Cases from External Tertiary and General Hospitals

Separate external validation studies also confirmed the solution’s robust performance. Purple AI reported high diagnostic accuracy in 1,400 cases collected from four external tertiary and general hospitals. This demonstrates that the solution maintained consistent performance even in real-world clinical environments with different imaging equipment manufacturers and scanning protocols at each hospital.

Professor Yoon Taejin of the Radiology Department at Seoul National University Hospital, who led the clinical research, said, "Non-contrast brain CT is the first gateway for emergency stroke diagnosis, but even experienced specialists can face challenges in diagnosing stroke. AI assistance improved diagnostic accuracy for all interpreters, regardless of their expertise level, and the largest improvement was observed among non-radiology medical staff who are the first to see patients in emergency situations. This has significant clinical implications."

Meanwhile, Purple AI's stroke diagnosis AI solution has been implemented in more than 50 tertiary and general hospitals in Korea within about one year of obtaining approval from the Ministry of Food and Drug Safety, and an international clinical trial is underway for approval from the U.S. Food and Drug Administration (FDA).

This content was produced with the assistance of AI translation services.