Annalise Enterprise CTB Triage Trauma

Annalise Enterprise CTB Triage Trauma is an AI-driven software designed to enhance radiology workflow efficiency by rapidly analyzing non-contrast brain CT scans to detect and flag potential acute hemorrhages for prioritized review. The system leverages a deep-learning convolutional neural network, trained on an extensive dataset of radiological images, to provide study-level notifications that assist clinicians in triaging critical cases without replacing their diagnostic judgment.

Annalise Enterprise CTB Triage Trauma is a software tool utilizing an artificial intelligence (AI) algorithm to analyze non-contrast brain CT studies for specific acute traumatic findings in adult patients aged 22 and older. Its core purpose is to aid clinicians in the triage and prioritization of studies by identifying features suggestive of acute subdural/epidural hematoma, acute subarachnoid hemorrhage, intra-axial hemorrhage, and intraventricular hemorrhage. The AI employs a convolutional neural network, developed through deep-learning techniques and trained on a large dataset of over 200,000 CT brain imaging studies reviewed by radiologists. The device integrates with existing image and order management systems, such as PACS/RIS, to process studies and issue notifications for prioritizing cases with suspected findings in the worklist, without ever downgrading a study's existing priority. The system is intended for use by trained clinicians qualified to interpret brain CTs and is not designed for standalone clinical decision-making, ruling out findings, or directing attention to specific image regions; it solely provides study-level notifications. Its performance was validated through standalone evaluations where device outputs were compared against a reference standard established by US board-certified neuroradiologists. These studies demonstrated the device's effectiveness across various operating points and slice thicknesses, showing high sensitivity and specificity for detecting the targeted hemorrhages. For example, for acute subdural/epidural hematoma at ≤1.5mm slice thickness, one operating point achieved 91.4% sensitivity and 86.7% specificity. Triage effectiveness was also confirmed, with a reported mean turn-around time of approximately 81.6 seconds for positive 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 algorithm's performance vary with different CT scanner manufacturers and imaging protocols beyond those explicitly listed in the training and validation datasets (Toshiba, GE, Siemens, Philips, Canon)?; What is the impact of co-existing, non-hemorrhagic brain pathologies (e.g., tumors, old infarcts, edema) on the algorithm's false positive and false negative rates for the targeted acute hemorrhage types?; What are the specific image pre-processing steps applied to the DICOM data before the AI algorithm analyzes it, and how robust is the algorithm to variations in image acquisition parameters like noise levels or reconstruction kernels?; Given the training dataset size of over 200,000 studies, how were potential biases related to patient demographics, geographical origin of data, or prevalence of rare hemorrhage presentations addressed and mitigated during model development and validation?; How does the device handle images with significant artifacts (e.g., motion artifacts, beam hardening, streak artifacts from metallic implants), and what is the threshold or mechanism for determining if image quality is insufficient for reliable analysis?