HealthPPT

HealthPPT is an AI-driven software designed to enhance clinical workflow by rapidly analyzing adult frontal chest X-rays for suspected pneumoperitoneum, thereby enabling prioritization of critical cases for radiologist review. The system utilizes a machine learning algorithm integrated with PACS/workstations to identify and flag potential findings, demonstrating high accuracy (AUC 96.75%) with adjustable sensitivity and specificity operating points to suit different clinical needs.

HealthPPT, developed by Nanox AI Ltd., is an artificial intelligence-powered software tool designed to assist in the clinical assessment of adult frontal chest X-rays by identifying features suggestive of pneumoperitoneum. Its primary function is to act as a workflow aid, analyzing PA or AP chest X-ray images using a proprietary algorithm and notifying a PACS/workstation or a standalone desktop application (ZebrAInsight) of suspected critical findings. This notification, which may include a compressed, non-diagnostic preview image, facilitates the prioritization and triage of radiological cases within a medical care environment, allowing clinicians to review potentially urgent studies sooner. The system operates in parallel with existing standard of care workflows, does not alter original images, and is not intended for standalone diagnosis, treatment recommendation, or to rule out pneumoperitoneum. The HealthPPT system comprises modules for data input/validation, the core pneumoperitoneum detection algorithm, PACS/workstation integration (IMA), and error reporting. Its performance was validated using a retrospective cohort of 216 anonymized chest X-ray cases (107 positive for pneumoperitoneum, 109 negative), with ground truth established by three US board-certified radiologists. The device demonstrated an Area Under the Curve (AUC) of 96.75%. It offers two operating points: a "balanced sensitivity and specificity" point achieving 92.52% sensitivity and 92.66% specificity, and a "high-specificity" point achieving 80.37% sensitivity and 97.25% specificity. The average analysis and notification time was reported as 4.78 seconds.

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 HealthPPT algorithm differentiate true pneumoperitoneum from common mimics or confounding factors on chest X-rays, such as Chilaiditi sign or bowel gas overlying the diaphragm?; What was the range and median size or subtlety of pneumoperitoneum cases in the validation set, and how does performance vary with the conspicuity of the finding?; Beyond AUC, sensitivity, and specificity, what is the standalone positive predictive value (PPV) and negative predictive value (NPV) of HealthPPT at its different operating points, considering a realistic prevalence of pneumoperitoneum in an acute care setting?; How does the integration with different PACS/workstation vendors impact the notification speed and user experience, and are there specific technical requirements for optimal integration?; Given the two operating points (balanced vs. high-specificity), what clinical scenarios or institutional preferences might guide the selection of one over the other for implementation?