HALO

HALO is an artificial intelligence-powered, cloud-based software designed to expedite the triage of potential stroke patients by analyzing CT angiograms of the brain for signs of Large Vessel Occlusion. The system operates by automatically processing images, identifying suggestive patterns, and rapidly notifying specialists to facilitate prompt review, without replacing the clinician's final diagnostic responsibility.

HALO is a cloud-based radiological computer-aided triage and notification software designed by Nico.Lab B.V. to assist medical specialists in the rapid evaluation of stroke patients. The system processes CT angiograms (CTA) of the brain using artificial intelligence algorithms to detect image patterns suggestive of intracranial Large Vessel Occlusion (LVO) in the anterior circulation. Original CTA images are transmitted to HALO's cloud servers for automated analysis, operating in parallel with standard care workflows. If a suspected LVO is identified, HALO sends a notification to a designated specialist, who can then remotely review the results via a DICOM web viewer, facilitating timely attention to potential stroke cases. The intended users of HALO are medical specialists such as neurologists, radiologists, and emergency department physicians involved in the diagnosis and care of stroke patients in acute settings. It is crucial to note that HALO's output is for notification and triage purposes only and should not be used for primary diagnosis or to make clinical decisions, as the final diagnosis remains the responsibility of the specialist. In a multi-center retrospective clinical study involving 287 patient scans, the HALO algorithm demonstrated a sensitivity of 90.1% and a specificity of 90.7% for LVO detection, with a median notification time of 1 minute and 4 seconds for positive cases. The system is compatible with CT scanners from GE Healthcare and Philips.

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 of the HALO AI algorithm vary with different CT scanner models from GE Healthcare and Philips, particularly regarding image quality and acquisition protocols?; What is the false positive notification rate in real-world clinical settings outside of a structured study, and what are the common characteristics of these false positives?; How does the integration of HALO into existing hospital IT infrastructures and PACS systems affect the overall workflow efficiency beyond just the notification time?; What specific AI architectural components or deep learning models are utilized by HALO, and how were they trained and validated to ensure generalizability across diverse patient populations?; Are there plans to expand HALO's capabilities to detect other critical neurological conditions or LVO in other circulations (e.g., posterior) visible on CTA scans?