AI Stroke Detector Gets FDA Nod
Harrison.ai has secured US FDA 510(k) clearance for its AI tool that triages acute infarcts on non-contrast CT brain scans. The approval is a major validation for ML-powered diagnostic imaging, as the platform is designed to accelerate stroke detection and help clinicians prioritize critical cases.
The key innovation lies in analyzing the initial, non-contrast CT scan—the very first image a potential stroke patient receives. This is a fundamental shift from most existing stroke AI, which requires a later, more complex CT angiography (CTA) scan that uses contrast dye. The system is designed to spot actual brain tissue injury, not just the large vessel blockages that competing tools focus on. In its FDA submission, the AI demonstrated superior performance, achieving up to 89.2% sensitivity in detecting confirmed infarcts on thin-slice CTs. This compares favorably to the closest FDA-cleared competitor on non-contrast CTs, which showed 63.5% sensitivity for identifying only vessel occlusions. Harrison.ai's tool reached over 80% sensitivity and specificity for identifying actual tissue damage. The platform's diagnostic breadth is a major technical advantage, covering infarcts across six major vascular territories of the brain, including the ACA, MCA, and PCA. This clearance is the company's ninth from the FDA, and its comprehensive brain scan solution can now identify 13 distinct radiological findings. Sydney-based Harrison.ai, co-founded by Dr. Aengus Tran, is well-capitalized for its US expansion, having raised over $240 million to date. The company's growth is also fueled by partnerships, including a joint venture with Sonic Healthcare to develop AI for pathology and a history with I-MED Radiology that provided vast datasets for training. This technology directly addresses the "time is brain" principle in stroke care, where millions of brain cells die every minute treatment is delayed. Studies of similar AI tools in UK hospitals found they helped patients receive critical clot-removal surgery more than an hour sooner, effectively doubling their chances of regaining independence. The AI model was trained on one of the largest and most diverse medical imaging datasets globally. Through its joint venture Annalise.ai (since rebranded as Harrison.ai Radiology), the company developed deep neural networks using millions of anonymized, radiologist-labeled images to achieve high accuracy.
