Radify Triage

Radify® Triage is an artificial intelligence software tool that analyzes adult chest X-ray images to detect suspected pleural effusion and pneumothorax, facilitating worklist prioritization for radiologists. By employing deep convolutional neural networks to identify features indicative of these critical conditions from DICOM images, it provides passive, case-level notifications within the PACS environment to streamline review without offering standalone diagnostic interpretations.

Radify® Triage is a radiological computer-assisted triage and notification software designed for analyzing adult chest X-ray (AP and PA view) images. Its primary purpose is to identify pre-specified suspected critical findings, specifically pleural effusion and/or pneumothorax, by using an artificial intelligence algorithm. The software processes DICOM format images received via PACS and provides a case-level output back to the PACS, flagging images with suspicious findings to enable worklist prioritization by appropriately trained medical specialists. Importantly, Radify® Triage functions as a passive notification tool within the standard care workflow; it does not send proactive alerts, direct attention to specific image portions, remove cases from the queue, or indicate the absence of a condition, and its results are not intended for standalone clinical decision-making. The core technology of Radify® Triage is a deep convolutional neural network (DCNN) trained on a dataset primarily from South Africa (93.7%) and the USA (6.3%). Its performance was validated using an independent test dataset of 1229 chest X-rays from US sites, with ground truth established by three board-certified radiologists. For pneumothorax detection, the software achieved an AUC of 97.43%, sensitivity of 94.81%, and specificity of 97.91%. For pleural effusion, it demonstrated an AUC of 97.61%, sensitivity of 94.39%, and specificity of 96.42%. The system showed consistent performance across different genders, age groups, image acquisition devices, and in cases with co-existing conditions, with an average notification time of 3 seconds to the worklist.

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 reported performance (AUC, sensitivity, specificity) of Radify® Triage translate to its utility in diverse clinical settings with varying prevalence rates of pneumothorax and pleural effusion?; Given that the training dataset was predominantly from South Africa (93.7%), what is the robustness and generalizability of the AI algorithm to patient populations and imaging practices unique to other geographical regions not extensively represented in training or the US-based validation?; What specific types or severities of pleural effusion and pneumothorax (e.g., small, trace, tension) was the algorithm most and least effective at identifying, and what are the implications for its use in prioritizing subtle but critical cases?; Beyond the listed clinical confounders (cardiomegaly, consolidation, infiltrates, opacities, nodules), what other co-existing pathologies or image artifacts were observed to impact the algorithm's performance during validation?; How does the 3-second average notification time for Radify® Triage compare to the total image processing and radiologist review time, and what is the net impact on overall workflow efficiency and patient care timelines?