Rapid ICH

RAPID ICH is an AI-powered software designed to expedite the triage of suspected intracranial hemorrhage by analyzing non-contrast head CT scans and notifying clinicians of potential positive cases in parallel with standard care. Its methodology relies on a deep learning algorithm to achieve high sensitivity and specificity in ICH detection, aiming to improve workflow efficiency in acute care settings without replacing standard diagnostic procedures.

RAPID ICH, developed by iSchemaView, Inc., is a radiological computer-assisted triage and notification software designed to aid in the workflow prioritization of suspected intracranial hemorrhage (ICH). It operates as a module within the integrated RAPID Platform, analyzing non-enhanced head CT (NCCT) images using an artificial intelligence algorithm. The system processes DICOM-compliant image data, and upon identifying a suspected ICH, it notifies the clinical team via PACS/Workstation, email, and a mobile application. These notifications include compressed preview images intended for informational purposes only, prompting clinicians to review the full, original images on a diagnostic workstation as part of the standard care pathway; the software does not alter the original medical image and is not intended as a diagnostic device. The core science behind RAPID ICH involves a deep learning algorithm trained on medical images to detect potential ICH findings. The system functions in parallel with existing standard of care workflows, aiming to expedite the review of critical cases without removing any cases from the standard worklist. Performance data from a retrospective, blinded, multicenter study involving 336 cases demonstrated a sensitivity of 89.9% and a specificity of 94.3% in identifying ICH, both exceeding an 80% performance goal. The average time to notification, encompassing DICOM exam acquisition, de-identification, analysis, and alert dissemination, was reported as 2.28 minutes. The software is intended to be used in conjunction with other patient information and professional judgment.

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 AI algorithm differentiate between true ICH and potential mimics, such as calcifications or other hyperdense lesions, especially given the caution about IV contrast leading to false positives?; What was the range of ICH volumes and locations included in the training and validation datasets, and how does the software's sensitivity vary with the size, subtlety, and location of the hemorrhage?; Beyond sensitivity and specificity, what are the reported positive predictive value (PPV) and negative predictive value (NPV) of RAPID ICH in diverse real-world clinical populations with varying ICH prevalence?; How does the system handle images with significant motion artifacts, beam hardening, or other common CT quality issues, and what is the documented failure rate or rate of non-interpretable analyses due to poor image quality?; What specific features of the deep learning architecture contribute to its performance, and how was the algorithm validated against diverse patient demographics, co-morbidities, and scanner types/manufacturers beyond the 336 cases mentioned?