Heuron ICH

Heuron ICH is an artificial intelligence-powered software designed to expedite the triage of potential intracranial hemorrhage cases by analyzing non-contrast head CT scans and rapidly notifying clinicians of suspected positive findings. Utilizing a deep learning convolutional neural network, it aims to improve workflow efficiency in radiology by prioritizing urgent cases, with its performance in detection and notification speed validated through a multi-center study.

Heuron ICH is a radiological computer-aided triage and notification software designed to analyze non-contrast head CT (NCCT) images for adults and transitional adolescents aged 18 and older. Its primary purpose is to assist medical specialists and hospital networks in streamlining workflows by rapidly identifying and communicating suspected positive findings of Intracranial Hemorrhage (ICH), thereby aiding in the prioritization and triage of medical images. The system integrates with a hospital's Picture Archiving and Communication System (PACS), automatically receiving NCCT images in DICOM format. It employs an artificial intelligence algorithm, specifically a deep learning convolutional neural network (CNN) with a distinct learning method and training data, including a dataset from the RSNA Brain CT Hemorrhage Challenge 2019, to analyze these images. Upon detecting a suspected ICH, Heuron ICH sends notifications via email, SMS, or mobile app, which include compressed preview images for informational purposes only and not for diagnosis beyond notification. The software does not alter original medical images, provide segmentations, or offer intermediate analytical outputs, and is not a standalone diagnostic tool; clinicians must review full NCCT images per standard care. Performance was evaluated in a retrospective, multi-center study using 600 NCCT images, demonstrating a sensitivity of 86.3% and specificity of 87.6% for ICH detection, an AUC of 0.945, and an average notification time of approximately 60.3 seconds, noted as significantly faster than standard care timelines.

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 Heuron ICH algorithm (sensitivity, specificity, AUC) vary across different subtypes and severities of intracranial hemorrhage?; What is the impact of excluding Siemens scanner CT images outside the 2.5-5mm slice thickness range on the overall clinical utility and potential for missed detections in a real-world setting?; Given that the training dataset was from the RSNA 2019 challenge, what measures are in place to ensure the algorithm's robustness and generalizability to diverse patient populations and CT scanner technologies not extensively represented in the initial training or the 600-image validation set?; How does the reported time-to-notification of approximately 60.3 seconds translate into actual improvements in patient outcomes or overall emergency department/radiology department efficiency metrics in prospective clinical studies?; What are the specific imaging characteristics of false positive and false negative cases identified by Heuron ICH, and what ongoing efforts are there to refine the algorithm to minimize these?