Rapid NCCT Stroke

Rapid NCCT Stroke is an AI-powered software designed to expedite the triage of suspected Intracranial Hemorrhage and Large Vessel Occlusion in adult stroke patients by analyzing non-enhanced head CT scans. It utilizes specialized algorithms for ICH detection and a combination of ASPECTS and Hyperdense Vessel Sign analysis for LVO suspicion, aiming to provide rapid notifications to clinicians and improve workflow efficiency in time-sensitive stroke care.

Rapid NCCT Stroke is a radiological computer-aided triage and notification software designed to analyze nonenhanced head CT (NCCT) images for adult patients. Its primary purpose is to assist hospital networks and trained clinicians in prioritizing workflow by flagging and communicating suspected findings of Intracranial Hemorrhage (ICH) and Large Vessel Occlusion (LVO) in the ICA and MCA-M1 regions directly from NCCT scans. The system employs an artificial intelligence algorithm, operating within the integrated Rapid Platform, to analyze these images. For ICH detection, it utilizes a dedicated ICH algorithm, while LVO suspicion is determined through a combined analysis of ASPECTS and Hyperdense Vessel Sign (HVS) algorithms. Notifications of suspected positive cases are delivered via PACS, email, or mobile devices, including compressed preview images intended for informational and triage purposes only, not for primary diagnosis. Performance of Rapid NCCT Stroke was evaluated in a retrospective, multicenter study involving 254 cases. The device demonstrated high sensitivity (0.962) and specificity (0.974) for ICH detection. For LVO detection on NCCT, the software achieved a sensitivity of 0.635 and specificity of 0.951, and also showed non-inferiority to expert readers and superiority to non-expert readers in LVO triage. A significant advantage highlighted is the rapid notification time, averaging 2.5 minutes, compared to the standard care's mean image review initiation time of 58.72 minutes. Although LVO detection sensitivity on NCCT is inherently lower than on CTA, the device aims to expedite initial LVO suspicion, with cautions that CTA or MRA remains crucial for definitive stroke workup and that the device is not for ruling out LVO. Subgroup analyses across different demographics and scanner types were conducted, with the device consistently aiming to improve time-to-notification for critical stroke findings.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. Considering the LVO sensitivity of 0.635 on NCCT, what specific training data and algorithmic features allow the software to differentiate subtle early ischemic changes (ASPECTS) and hyperdense vessel signs from common NCCT artifacts or mimics that could lead to false positives or negatives?; The document reports lower LVO detection performance on Philips and Siemens scanners. What were the root causes identified for this variability, and what ongoing or planned model adjustments or site-specific calibrations are recommended to ensure equitable performance across different imaging hardware?; How does the system handle cases with 'Hemorrhagic Transformation' or 'Hematoma' listed under 'Contraindications/Exclusions,' given these are often critical findings the software is designed to detect (ICH) or might co-exist with LVO? Does this imply it won't process these scans or flags them differently?; The limitation for Rapid ICH is ≥ 0.4ml. What is the clinical significance of missing hemorrhages smaller than 0.4ml in the context of initial triage, and how does this threshold compare to the detection capabilities of human readers in emergent settings?; Beyond time-to-notification, has any data been collected on the software's impact on downstream clinical decisions, such as time-to-CTA/MRA, time-to-treatment, or reduction in workflow bottlenecks for the LVO pathway, especially considering the caution about not using it as a rule-out device?