Hyper Insight - ICH

Hyper Insight - ICH is a software medical device designed to expedite the triage of potential intracranial hemorrhage cases by utilizing deep learning algorithms to analyze non-contrast head CT scans for signs of acute bleeding. The system operates by identifying increased radiodensity patterns characteristic of hemorrhage on CT images and rapidly alerting clinicians to suspected positive cases, thereby aiming to prioritize critical patient care workflows without making a definitive diagnosis.

Hyper Insight - ICH is a software as a medical device (SaMD) intended to aid in the rapid triage of patients with suspected intracranial hemorrhage (ICH). It utilizes deep learning algorithms to automatically analyze non-contrast computed tomography (CT) images of the brain. The software identifies patterns indicative of acute ICH, such as increased Hounsfield Unit (HU) density characteristic of blood compared to normal brain tissue. Upon detecting a suspected ICH, Hyper Insight - ICH generates a binary prediction and sends a notification to medical specialists. It is designed to integrate with existing third-party worklist applications, receiving DICOM images and returning findings to alert radiologists. Importantly, the device serves as a notification-only workflow tool, meaning it is not intended for diagnostic purposes; any images previewed via its mobile application are for informational use only, and clinicians must review original images on a diagnostic viewer for definitive assessment. The performance of Hyper Insight - ICH was evaluated in a retrospective study using 394 non-contrast head CT images (198 with ICH, 196 without) from 13 U.S. clinical sites, with ground truth established by neuro-radiologists. The study demonstrated a clinical sensitivity of 95.45% (95% CI: 91.55-97.90%) and a specificity of 98.47% (95% CI: 95.59-99.68%) for detecting ICH, meeting pre-specified performance goals. The area under the receiver operating characteristic curve (AUC) was 0.9864, indicating strong discriminative ability. The average time from image analysis to notification was 16.39 ± 5.46 seconds. Performance was also stratified and analyzed across various subgroups, including clinical site, patient demographics (age, gender, race), ICH subtype, CT slice thickness, CT scanner manufacturer, and ICH volume, generally showing robust performance across these variables.

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. Given the performance variations observed across different ICH subtypes in similar AI tools, how robust is Hyper Insight - ICH in detecting less common or subtle hemorrhage patterns, such as small subarachnoid hemorrhages or those with atypical presentations?; The document mentions integration with third-party worklist applications. What are the interoperability standards (e.g., DICOM, HL7, FHIR) and technical protocols supported by Hyper Insight - ICH to ensure seamless integration across diverse hospital IT environments and PACS/RIS systems?; Considering the average notification time of approximately 16 seconds, what are the computational resource requirements for the deep learning model, and how does its performance scale with increasing case volume in a high-throughput emergency department or radiology practice?; The retrospective study reported high overall sensitivity and specificity. Could you elaborate on the characteristics of the false positive and false negative cases, including common image artifacts, patient conditions, or hemorrhage locations that most frequently led to misclassification?; Beyond binary notification of suspected ICH, does Hyper Insight - ICH provide any quantitative data (e.g., estimated hemorrhage volume, location probability maps) or confidence scores associated with its findings that could further assist clinicians in prioritizing cases, even if such information is not for diagnostic use?