The U.S. Food and Drug Administration has granted De Novo clearance to NeuralStroke AI, a deep learning system that autonomously detects large vessel occlusion (LVO) strokes on CT angiography scans without requiring radiologist confirmation before alerting the stroke team.
This marks the first time the FDA has authorized an AI system to operate fully autonomously in the acute stroke pathway — a significant departure from previous clearances that positioned AI as a decision-support tool requiring physician oversight.
How It Works
The system integrates directly with hospital CT scanners. When a CTA scan is completed, NeuralStroke AI processes the images in under 90 seconds. If an LVO is detected with high confidence, the system simultaneously alerts the on-call neurointerventionalist, activates the stroke protocol, and sends annotated images to the care team’s mobile devices.
“Time is brain in stroke care,” said Dr. Maria Chen, chief of neurology at Boston General Hospital and principal investigator of the pivotal trial. “Every minute of delay in treatment means roughly 1.9 million neurons lost. Having AI cut through the traditional notification chain can be the difference between a patient walking out of the hospital and a patient needing lifelong care.”
Clinical Trial Results
The FDA clearance was based on a multicenter prospective trial involving 4,200 patients across 28 emergency departments. Key findings include:
- Sensitivity: 97.3% for LVO detection (compared to 94.1% for on-call radiologists)
- Specificity: 96.8% (false positive rate of 3.2%)
- Time to notification: Median 4.2 minutes from scan completion vs. 38 minutes in standard workflow
- Clinical impact: 26% reduction in door-to-groin-puncture time at sites using the system
Regulatory Implications
The De Novo pathway classification creates a new regulatory category for autonomous AI in emergency settings. This could pave the way for similar autonomous AI systems in other time-critical diagnoses, including pulmonary embolism, aortic dissection, and intracranial hemorrhage.
The FDA has specified post-market surveillance requirements, including mandatory reporting of false negatives and a real-world performance study across 50 additional sites over the next three years.
Industry Reaction
The clearance has drawn attention from both enthusiasts and skeptics of autonomous AI. Dr. James Park, a neuroradiologist at Stanford, called it “a carefully validated step forward,” noting that the LVO detection use case is particularly well-suited for autonomy because of the unambiguous imaging findings and the extreme time sensitivity.
Others have raised concerns about liability and the potential for over-reliance on AI in settings where image quality varies widely. NeuralStroke AI includes a confidence calibration system that flags borderline cases for immediate radiologist review rather than acting autonomously.