AI-Rad Companion (Cardiovascular)

AI-Rad Companion (Cardiovascular) is an advanced software tool utilizing deep learning algorithms to automate the quantitative and qualitative analysis of cardiovascular structures from CT DICOM images. Its purpose is to enhance diagnostic support for clinicians by providing automated heart and aorta segmentation, coronary calcium volume quantification, and standardized aortic diameter measurements with threshold-based classifications, improving upon previous model-based techniques.

AI-Rad Companion (Cardiovascular) is an image processing software designed to assist radiologists and physicians in evaluating cardiovascular diseases using previously acquired Computed Tomography (CT) DICOM images. It provides quantitative and qualitative analysis by post-processing non-cardiac chest CT data from adult patients, aiming to support healthcare professionals in emergency medicine, specialty care, urgent care, and general practice by automating and enhancing the assessment of cardiac structures and vasculature. The core methodology of AI-Rad Companion (Cardiovascular) employs deep learning algorithms, an advancement from previous model-based approaches, for several key image post-processing functions. These include the segmentation and volume measurement of the heart, quantification of total coronary calcium volume, segmentation of the aorta, and the measurement of maximum aortic diameters at nine American Heart Association (AHA) specified landmarks. The software further provides threshold-based highlighting of enlarged aortic diameters and classifies these diameters into categories based on AHA standards. Performance validation on non-cardiac chest CT data demonstrated a high logarithmic correlation coefficient (0.96, N=381) for total coronary calcium volume against a predicate device, and an average absolute error of 1.6 mm (N=193) for aorta diameter measurements, indicating consistent performance across different CT vendors and slice thicknesses.

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 deep learning algorithms for heart and aorta segmentation compare when applied to CT scans with contrast media, given its current validation primarily on non-cardiac chest CTs which may often be non-contrast?; What is the specific range of Hounsfield Units targeted for calcium quantification and the maximum slice thickness (stated as <=3mm) where performance is validated, and are there known confounders (e.g., image artifacts, adjacent high-density structures) that significantly impact its performance?; Beyond the 9 AHA landmarks, can the AI-Rad Companion (Cardiovascular) software be adapted or trained to measure aortic diameters at other clinically relevant locations, or assess other aortic pathologies like dissection or intramural hematoma?; How does the software's threshold-based classification of aortic diameters account for patient-specific factors such as body surface area or age, which can influence normal aortic dimensions, beyond the general AHA guidelines?; What was the demographic and pathological diversity of the training and validation cohorts for the deep learning models, and how might this affect the generalizability of AI-Rad Companion (Cardiovascular) to different global patient populations or rare conditions?