DrAid for Radiology v1

DrAid for Radiology v1 is an AI-driven software designed to identify suspected pneumothorax in adult frontal chest X-rays, facilitating prioritized radiologist review by integrating with PACS and providing passive notifications. Its methodology leverages a deep learning algorithm to analyze DICOM images, with stand-alone performance testing on diverse datasets demonstrating high sensitivity and specificity for pneumothorax detection.

DrAid for Radiology v1 is a radiological computer-assisted triage and notification software product. Its purpose is to automatically identify suspected pneumothorax on frontal chest X-ray images from adult patients (over 22 years old) and notify the Picture Archiving and Communication System (PACS) to enable prioritized review by appropriately trained medical specialists. The software does not alter the order or remove cases from the reading queue but aims to aid in the prioritization and triage of radiological medical images. Chest radiographs in DICOM format are automatically received from the user's image storage system (e.g., PACS) or other radiological imaging equipment (e.g., X-ray systems). The software device de-identifies a copy of each chest radiograph and its AI algorithm analyzes each image to identify features suggestive of pneumothorax. Based on the analysis, it notifies PACS/workstation of the presence of pneumothorax as indicating either a 'flag' or '(blank)', allowing specialists to group suspicious exams. The device operates in parallel with standard care workflow, providing analysis results as DICOM files and an on-device notification in PACS about prioritized cases. Key modules include data input and validation, the AI algorithm, an API Cognitive service for study analysis and results transmission, and an error codes feature. The performance of DrAid for Radiology v1 was validated in two pivotal stand-alone studies. One study utilized a dataset from the National Institute of Health (NIH) to demonstrate generalizability to the US population demographics, consisting of 565 radiographs (386 negative, 179 positive pneumothorax cases), truthed by three US board-certified radiologists. For this NIH dataset, the device achieved a sensitivity of 0.9387, specificity of 0.9947, and an AUC of 0.9667. A second study used a dataset from four Vietnamese hospitals to demonstrate generalizability to different scanner types, consisting of 285 radiographs (110 negative, 175 positive pneumothorax cases), also truthed by three US board-certified radiologists. For this Vietnamese dataset, performance was: sensitivity 0.9535, specificity 0.9464, and AUC 0.9500. The aggregate results from both datasets showed a sensitivity of 0.9461, specificity of 0.9758, and an AUC of 0.9610. The average time from image availability to notification on the PACS worklist was 3.83 minutes.

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. What are the performance characteristics of DrAid for Radiology v1 when analyzing chest X-rays from patients with pre-existing lung conditions or a history of thoracic surgery, and how might these conditions affect its pneumothorax detection accuracy?; How does the AI algorithm differentiate between a true pneumothorax and image artifacts or other pathologies that might mimic its appearance on a frontal chest X-ray?; The documentation mentions an average time-to-notification of 3.83 minutes. What is the variability or range of this notification time, and how does it perform under high-load scenarios in a busy clinical environment?; What specific de-identification methods are employed by the software for DICOM images before analysis, and what measures are in place to ensure patient data privacy and security, especially with cloud-based components like the API Cognitive Service and AI Algorithm?; Given the device is indicated for adults (>22 years), what are the key anatomical or pathological differences that preclude its use in pediatric patients, and are there plans to validate or adapt it for younger populations?