JLK-SDH

JLK-SDH is an AI-powered radiological software that analyzes non-contrast CT images of the head to detect suspected subdural hemorrhage (SDH) and sends notifications to radiologists. It operates as a parallel workflow tool alongside standard clinical care, enabling radiologists to identify and communicate patient images to specialists without modifying the standard diagnostic workflow.

JLK-SDH uses an artificial intelligence/machine learning algorithm hosted on JLK servers that receives DICOM-formatted CT images, performs analysis, and sends push notifications via a mobile application. Like the predicate device, it provides notification-only functionality with no image alteration or marking. The key difference is that JLK-SDH notifies radiologists who then contact appropriate specialists, whereas the predicate (Viz SDH) directly notifies neurovascular/neurosurgical specialists, but both achieve equivalent clinical utility within U.S. standard-of-care workflows.

Not stated in this summary. The document references FDA's Guidance for Industry on 'Content of Premarket Submissions for Device Software Functions' (June 14, 2023) and compliance with the DICOM standard, but does not cite specific ISO, IEC, or ASTM consensus standards.

JLK-SDH is substantially equivalent because it shares the same intended use (notification-only triage for SDH detection), identical indications, and comparable technological characteristics (AI/ML algorithm, DICOM compatibility, parallel workflow, non-diagnostic preview images, mobile notification interface) with the predicate Viz SDH. Performance validation on 560 independent NCCT scans demonstrated sensitivity of 97.1% and specificity of 97.4%, meeting or exceeding the predicate's established performance. The notification recipient difference (radiologist vs. specialist) does not raise new safety or effectiveness questions because U.S. hospital workflows include duty radiologists who perform the same triage and specialist consultation function, providing equivalent clinical benefit.