HALO

HALO is a cloud-based artificial intelligence software designed to rapidly analyze CT angiograms of the brain to identify potential Large Vessel Occlusions (LVOs) in suspected stroke patients. The underlying science involves AI-driven image analysis, which achieved a sensitivity of 91.1% and specificity of 87.0% for LVO detection in clinical evaluations, supporting its function in timely case identification.

HALO is a cloud-based image processing software application that employs artificial intelligence algorithms to analyze patient imaging data, specifically contrast-enhanced CT angiograms of the brain. Its primary purpose is to assist in the evaluation of brain vasculature in patients suspected of stroke by identifying imaging patterns suggestive of an intracranial Large Vessel Occlusion (LVO) in the anterior circulation (ICA, M1, or M2). The system operates in parallel with standard care, sending notifications directly to appropriate medical specialists, such as neurologists, radiologists, or emergency department physicians, when a potential LVO is detected. It is crucial to note that HALO's output is intended as a notification tool to facilitate timely review and is not meant for primary diagnosis or to replace the clinical judgment of medical specialists; it is indicated for use with CT scanners from GE Healthcare. The performance of HALO was evaluated in a multi-center clinical study involving 364 patient CTA scans. The algorithm demonstrated a sensitivity of 91.1% and a specificity of 87.0% for LVO detection, with an Area Under the Curve (AUC) of 0.97, using a ground truth established by an expert panel of neuroradiologists. The median time from CT acquisition to notification for detected LVO cases was 4 minutes and 31 seconds, with a range from 3 minutes 47 seconds to 7 minutes 12 seconds. These performance characteristics support its role as a tool to flag scans potentially requiring urgent specialist review in LVO 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 HALO's AI algorithm differentiate between true LVOs and mimics, and what is its performance on specific LVO locations (e.g., ICA, M1 vs. M2 beyond the general anterior circulation mention)?; What was the range of image quality and specific CT scanner models from GE Healthcare included in the validation study, and how does performance vary with these factors?; What is the observed real-world impact of HALO's notification system on door-to-needle or door-to-reperfusion times in stroke treatment?; Beyond providing a DICOM web viewer link, what are the detailed integration capabilities of HALO with existing hospital PACS and EHR systems for seamless workflow adoption?; What are the specifics of the non-clinical software testing performed, particularly concerning robustness against varied image inputs and the cybersecurity measures for patient data on cloud servers?