InferRead CT Stroke.AI

InferRead CT Stroke.AI is a software application utilizing a deep learning algorithm to analyze non-enhanced head CT images for the suspected presence of intracranial hemorrhage. Its purpose is to assist radiologists in workflow triage by visually marking potentially positive cases on a worklist, thereby facilitating earlier review and prioritization within the standard care pathway.

InferRead CT Stroke.AI is a radiological computer-assisted triage and notification software designed for the analysis of non-enhanced head CT images. The system operates on an Ubuntu OS and can be deployed as an onsite server, utilizing a deep learning algorithm to process DICOM series. It consists of four main modules: NeoViewer, Docking Toolbox, RePACS, and DLServer. The Docking Toolbox receives and inspects DICOM series; those meeting requirements are processed for intracranial hemorrhage (ICH) detection on a first-in-first-out basis. When a potential hemorrhage is detected, the system marks the case on a worklist, prompting users for preemptive triage and prioritization, without altering the original medical image or removing cases from the standard queue. The device is intended to assist hospital networks and trained radiologists by flagging suspected positive ICH findings to improve workflow triage. It operates in parallel with the standard of care, providing a worklist with marked cases. A retrospective clinical study involving 369 non-contrast brain CT scans demonstrated the software's performance: it achieved a sensitivity of 0.916 and a specificity of 0.922 for ICH detection against a ground truth established by neuroradiologists. The area under the receiver operating characteristic curve (AUC) was 0.962. Furthermore, the InferRead time-to-notification was significantly shorter (mean 1.07 minutes) compared to the standard of care time-to-open-exam (mean 75.4 minutes), indicating its potential to expedite the review of critical cases.

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 performance (sensitivity and specificity) of the InferRead CT Stroke.AI deep learning algorithm vary with different severities or types of intracranial hemorrhage (e.g., subarachnoid, intraparenchymal)?; What is the computational time required for the algorithm to process a typical non-enhanced head CT series, and how does this scale with increasing patient volume?; What were the characteristics of the false positive and false negative cases in the performance study, and what steps are being taken to address these?; How does the 'marking cases in the worklist' notification method impact radiologist workflow and cognitive load compared to other notification systems, such as pop-ups used by predicate devices?; Are there plans to expand the algorithm's capabilities to detect other acute neurological conditions visible on non-enhanced head CT scans beyond intracranial hemorrhage?