Accipiolx

AccipioIx is an artificial intelligence-powered software designed to expedite the identification of potential acute intracranial hemorrhage in adult non-contrast head CT scans, thereby enabling prioritization for radiological review. It utilizes deep learning algorithms for image analysis, including CNN-based segmentation and feature detection, to provide case-level triage notifications within the acute care setting.

AccipioIx, from MaxQ AI Ltd., is a radiological computer-aided triage and notification software designed for the acute care environment. Its primary function is to aid clinicians by prioritizing the review of adult non-contrast head CT (NCCT) scans that show signs suggestive of acute intracranial hemorrhage (aICH). The system integrates into radiology workflows by analyzing NCCT scans using an artificial intelligence algorithm. This algorithm identifies suspected aICH and provides a case-level output to PACS/workstation systems, facilitating worklist prioritization or triage. AccipioIx operates alongside standard care protocols and serves as an assistive tool, not intended for stand-alone diagnostic decisions or to definitively rule out hemorrhage. The core technology of AccipioIx leverages deep learning methodologies. Its AI algorithm, trained on a diverse set of CT cases, executes several computational steps, including skull stripping, image registration, tissue segmentation via a Convolutional Neural Network (CNN)-based method, and feature identification using hyperdensity compared to pre-set thresholds. This iteration improves upon its predecessor with an enhanced segmentation technique and new algorithmic features for better aICH detection, including an assessment of the entire case in addition to individual findings. Clinical performance testing on 360 cases yielded a sensitivity of 97% and specificity of 93% for aICH detection, with an average processing time of 1.17 minutes per case. It demonstrated 100% sensitivity for intra-axial hemorrhages, 92% for extra-axial hemorrhages, and a high Negative Predictive Value (NPV) of 99.8%.

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 CNN-based segmentation method specifically compare to the previous machine-vision based segmentation in terms of accuracy and computational efficiency for identifying different types and locations of intracranial hemorrhage?; What were the characteristics of the 7% of cases where AccipioIx produced false positives (100% - 93% specificity) and the 3% of cases with false negatives (100% - 97% sensitivity), and are there common features that could inform further algorithm refinement?; Given the Positive Predictive Value of 43.3%, how might this impact workflow in terms of the number of non-hemorrhage cases being flagged for urgent review, and what strategies are recommended for managing these notifications?; How was the training dataset for the deep learning algorithm curated to ensure diversity in patient demographics, scanner types, and hemorrhage characteristics, and what was the size of this training set?; Could the 'additional algorithmic features for enhanced detection of aICH' that assess the 'full case' be detailed further, particularly how they complement the detection of individual aICH findings?