Viz ICH

Viz ICH is a radiological computer-aided triage and notification software designed to accelerate the detection and communication of suspected intracranial hemorrhage (ICH) to specialists. It employs an artificial intelligence algorithm to analyze non-contrast CT brain scans, providing rapid alerts to potentially improve response times in critical care settings, with demonstrated high sensitivity and specificity in clinical studies.

Viz ICH is a software-based, notification-only tool designed to assist hospital networks and trained clinicians by rapidly identifying and communicating potential cases of intracranial hemorrhage (ICH). The system functions as a parallel workflow tool, independent of the standard of care image interpretation pathway. It utilizes an artificial intelligence (AI) machine learning algorithm to analyze non-contrast CT (NCCT) images of the brain, specifically those acquired in acute settings like stroke protocols. Upon detecting image features suggestive of ICH, the system automatically sends a notification to a designated specialist, such as a neurovascular specialist or neurosurgeon, alerting them to the case. This notification includes the ability to preview images via a mobile application, though these compressed images are for informational purposes only and not intended for diagnostic use beyond the initial alert; clinicians must review non-compressed images on a diagnostic viewer for care-related decisions. The core science involves an AI/ML algorithm hosted on Viz.ai's servers, which processes NCCT images of the head forwarded from DICOM compliant CT scanners from various manufacturers. The algorithm is trained to identify patterns indicative of ICH without altering the original images or marking them. The system's performance was evaluated using 387 NCCT scans, demonstrating a sensitivity of 95% (95% CI: 91%-98%) and specificity of 96% (95% CI: 92%-98%) against ground truth established by neuro-radiologists, meeting pre-specified goals. The area under the ROC curve was 0.97, and the average time to alert a specialist was significantly reduced to 0.49 minutes. Performance was further stratified by factors including clinical site, patient demographics, ICH subtype and volume (with lower sensitivity for ICH volumes ≤1mL at 69%), slice thickness, and various CT scanner parameters, generally maintaining high performance.

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's performance vary with different ICH subtypes (e.g., subarachnoid vs. intraparenchymal hemorrhage) and ICH volumes, particularly for subtle or very small hemorrhages below 1mL, and what are the implications for missed notifications?; Given the system's role as a 'notification-only, parallel workflow tool,' what studies have been or could be conducted to assess its real-world impact on patient outcomes, beyond notification speed, such as time-to-treatment or clinical decision-making?; The performance data includes various CT scanner manufacturers and models; are there specific scanner parameters or image reconstruction methods that significantly impact the AI's detection accuracy, and how is the algorithm generalized or calibrated for new, unrepresented scanner types?; Since the mobile application provides compressed, non-diagnostic images for preview, what are the human factors considerations and training protocols implemented to ensure clinicians consistently use the primary diagnostic viewer for decision-making and avoid potential misinterpretation from preview images?; The summary mentions the AI/ML algorithm is hosted on Viz.ai's servers. What are the data security, privacy (HIPAA compliance), and system reliability (uptime, latency) measures in place for transferring and processing sensitive patient imaging data externally?