Rapid SDH

Rapid SDH is an AI-powered software tool utilizing machine learning algorithms to analyze non-contrast head CT images for the expedited identification and notification of suspected hemispheric subdural hemorrhage, thereby aiming to optimize radiology workflow triage. While demonstrating high overall sensitivity and specificity in clinical studies, its performance varies with patient age and hemorrhage volume, and it operates as an adjunct to, not a replacement for, standard radiological review.

Rapid SDH, developed by iSchemaView, Inc., is a radiological computer-aided triage and notification software designed to assist radiologists in managing workflow for non-enhanced head CT (NCCT) scans. Its primary purpose is to identify and flag suspected cases of hemispheric Subdural Hemorrhage (SDH) for prioritized review, using an artificial intelligence/machine learning (AI/ML) algorithm to analyze images. Operating within the integrated Rapid Platform, the software processes DICOM images and generates priority notifications, including compressed preview images (for informational purposes only, not for diagnosis), to clinicians via PACS, workstation, email, or mobile application. Rapid SDH functions in parallel with standard image interpretation, does not alter original medical images, and is not a diagnostic device; clinicians must review full images on a diagnostic viewer for diagnostic decisions. The software's performance was assessed in a retrospective, blinded, multicenter, multinational study of 310 NCCT head image sets (147 SDH-positive), with ground truth established by three expert neuro-radiologists. Rapid SDH demonstrated high efficacy, achieving a sensitivity of 0.924 (95% CI: 0.871–0.956), a specificity of 0.987 (95% CI: 0.95477–0.996, interpreting the documented '95477' lower bound as 0.95477), and an AUC of 0.995 for identifying hemispheric SDH. The median notification processing time was 45 seconds. Indicated for patients over 21 years, it's validated for hemispheric SDH ≥ 1ml. However, performance showed variations, with lower sensitivity for patients ≤ 50 years (0.778) and SDH volumes < 10ml (0.724), and its validation excluded certain mimics like subdural hygroma; IV contrast or significant motion artifacts can also impact results.

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. How does the Rapid SDH algorithm's performance (sensitivity/specificity) change with varying levels of image quality or in the presence of subtle artifacts not classified as 'excessive patient motion'?; What specific imaging biomarkers or features does the AI/ML model primarily leverage to differentiate hemispheric SDH from other intracranial pathologies or normal variants, particularly those excluded from the validation set like subdural hygroma?; Considering the observed lower sensitivity for SDH volumes < 10ml (0.724) and in patients ≤ 50 years old (0.778), what are the clinical implications for these subgroups, and are there specific strategies or future model refinements planned to address these performance gaps?; Beyond 'time-to-exam-open,' how does the integration of Rapid SDH notifications (via PACS, workstation, email, mobile) quantitatively impact the overall diagnostic turnaround time, from scan acquisition to final radiological report, in diverse real-world clinical environments?; What is the practical false positive notification rate of Rapid SDH in routine clinical use on non-contrast CT scans, and what are the most common etiological factors for these false positives, especially concerning the cautionary note about intravenous contrast?