Viz HDS, Viz Volume Plus, Viz ICH+

The Viz HDS system utilizes an artificial intelligence algorithm to automatically process non-contrast CT head scans, providing quantitative analysis of intracranial hyperdensities, lateral ventricles, and midline shift. Its methodology involves segmenting these structures from DICOM images and generating volumetric measurements and visual overlays to aid physicians in assessing single time-point brain scans.

The Viz HDS, Viz Volume Plus, Viz ICH+ system is a software-only medical image processing tool designed to assist clinicians by automating the analysis of non-contrast CT (NCCT) head scans. It employs a locked artificial intelligence machine learning (AI/ML) algorithm to automatically identify, label, visualize, and quantify key segmentable brain structures, specifically intracranial hyperdensities, lateral ventricles, and midline shift. The software processes DICOM-formatted NCCT scans acquired at a single time point, generating output in DICOM format that is sent to a PACS system. This output includes a summary series with calculated measurements and a segmentation series with RGB overlays on the original images to visually differentiate the segmented regions like hyperdensities and ventricles, and to display the midline shift. The system's output is intended for review by a physician alongside the patient's original images. The detailed output provides volumetric data for intracranial hyperdensities and lateral ventricles (including asymmetry ratio), and measures the distance of midline shift. Clinical performance data demonstrated the system's accuracy compared to ground truth established by radiologists. Mean absolute error for total hyperdensity volume was less than 7.5 mL, for lateral ventricle volumes less than 3 mL each, and for midline shift distance less than 2 mm (all at upper 95% CI). Furthermore, the DICE score, indicating agreement with manual measurements for hyperdensities and lateral ventricles, showed a lower 95% CI bound greater than 70%, meeting predefined performance goals across various stratification factors.

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/ML algorithm's performance vary across different CT scanner manufacturers and imaging protocols not included in the stratification data?; What is the processing time for a typical NCCT head scan, and how does this integrate into acute neurological emergency workflows?; Are there plans to expand the algorithm's capabilities to detect or quantify other critical brain pathologies or structures beyond hyperdensities, ventricles, and midline shift?; How robust is the algorithm in the presence of significant imaging artifacts or atypical patient anatomy?; What specific training datasets were used for the AI/ML algorithm, and what measures were taken to ensure diversity and minimize bias in its development and validation?