Aorta-CAD

Aorta-CAD is a software device employing a deep learning algorithm to analyze adult chest radiographs for findings indicative of chronic aortic conditions, specifically aortic calcification and dilated aorta. It functions as a concurrent reading aid for physicians, generating labeled bounding boxes around suspicious regions of interest to assist in the detection of Aortic Atherosclerosis and Aortic Ectasia.

Aorta-CAD is a computer-assisted detection (CADe) software designed to assist physicians in identifying suspicious regions of interest (ROIs) on adult chest radiographs that may indicate chronic aortic conditions. Employing a deep learning algorithm and computer vision techniques, the software analyzes these images and, if ROIs are detected, generates bounding boxes around them. These boxes are labeled with specific radiographic findings: "Aortic calcification," suggestive of Aortic Atherosclerosis, or "Dilated aorta," suggestive of Aortic Ectasia. Aorta-CAD produces a DICOM Presentation State file, an overlay that physicians can toggle on or off within their Picture Archiving and Communication System (PACS) viewer, functioning as a concurrent reading aid. The software is intended to support, not replace, the physician's judgment or other standard diagnostic procedures. The performance of Aorta-CAD was evaluated through standalone testing and a clinical reader study. In standalone tests involving 5,000 chest radiograph cases, the software demonstrated high overall sensitivity (0.910), specificity (0.896), and an Area Under the Curve (AUC) of 0.974. For specific findings, the AUC was 0.972 for Aortic calcification (sensitivity 0.922, specificity 0.894) and 0.948 for Dilated aorta (sensitivity 0.830, specificity 0.897). A multiple reader, multiple case (MRMC) retrospective clinical study with 244 cases showed that reader accuracy, measured by case-level AUC, was significantly superior when aided by Aorta-CAD compared to unaided readings (p < 0.001 for both categories). Specifically, reader AUC improved by 0.0984 for Aortic calcification and by 0.0885 for Dilated aorta when using the software.

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 Aorta-CAD algorithm's performance vary across different patient demographics (e.g., age, sex, BMI, presence of comorbidities) not explicitly detailed in the summary of the 5,000 case standalone assessment?; What is the typical processing time for Aorta-CAD per chest radiograph study in a real-world clinical setting, and how might this impact workflow efficiency in high-volume radiology departments?; The FROC curves terminate at the device’s box-level sensitivity due to the "cascaded nature of the Aorta-CAD predictions"; could you elaborate on this methodological aspect and its implications for detecting multiple or very subtle ROIs?; Given the Positive Predictive Value (PPV) for "Dilated aorta suggestive of Aortic Ectasia" was 0.296 in the standalone testing, what clinical guidance or training is provided to users to effectively manage the relatively high rate of false positives for this specific finding?; How robust is Aorta-CAD to variations in chest radiograph image quality, acquisition parameters, or X-ray equipment from different manufacturers, and were these variables assessed during the performance testing?