cvi42 Auto Imaging Software Application

cvi42 Auto is a software application designed to enhance cardiovascular diagnostics by utilizing machine learning for semi-automated quantitative analysis of cardiac MR and CT images, focusing on cardiac function, calcium assessment, and coronary artery evaluation. The system's methodology involves processing DICOM images with algorithms trained on diverse datasets, aiming to provide clinicians with accurate and efficient decision support tools, validated through performance testing that showed high classification and measurement accuracy.

cvi42 Auto is an advanced software as a medical device (SaMD) for viewing, post-processing, and evaluating cardiovascular magnetic resonance (MR) and computed tomography (CT) images. It aims to support physicians in qualitative assessment, quantitative measurements of cardiac structures and vessels, calcium scoring, and confirming physician-identified lesions. The core technology leverages machine learning for semi-automatic contouring of heart chambers in MR images for functional analysis, identifying calcified plaque in coronary arteries from non-contrast CT images, and placing centerlines in coronary vessels from CT images for stenosis assessment. These ML algorithms were trained on diverse, multi-site datasets to promote generalization, with distinct separation between training and validation data. The software includes a graphical user interface for comprehensive qualitative and quantitative analysis, including volume, mass, function, and signal intensity changes, with integrated reporting. It can also generate qualitative 3D views of the heart from CT data. cvi42 Auto processes DICOM files independently of the imaging equipment vendor. Performance validation on 235 patient images demonstrated strong results: CMR function analysis yielded 97-100% series classification accuracy and 5-10% volumetric Mean Absolute Error; calcium analysis achieved 86-99% classification performance; and coronary analysis showed 82-94% centerline performance. The machine learning functionalities are intended for adult patients, while manual tools are unrestricted.

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 machine learning algorithms in cvi42 Auto for cardiac function (e.g., chamber segmentation) compare to expert manual segmentation in terms of inter-observer and intra-observer variability?; What specific types of machine learning models are employed for cardiac function, calcium assessment, and coronary analysis, and how were they optimized for diverse scanner and image parameters?; Given the validation dataset included images from 'major vendors,' what is the range of performance variation observed across different scanner manufacturers or imaging protocols for the semi-automated analyses?; How does the software address potential biases in the training data related to patient demographics or specific pathologies, and what steps are taken to ensure robustness and fairness across different patient populations not heavily represented in the training set?; What are the typical processing times for the semi-automated machine learning workflows (cardiac function, calcium assessment, coronary analysis) on standard clinical workstations, and how does this integrate into existing clinical reporting workflows?