AISAP Cardio V1.0

AISAP Cardio V1.0 is a machine learning-driven software designed to augment cardiac Point-of-Care Ultrasound by automatically analyzing images to detect valvular pathologies and quantify key cardiac parameters in adult patients. Its methodology relies on neural networks trained on extensive clinical data, demonstrating improved diagnostic accuracy and providing adjunctive decision support for physicians.

AISAP Cardio V1.0 is a machine learning-based software platform designed to automatically process and analyze cardiac Point-of-Care Ultrasound (POCUS) images for adult patients. It generates a report detailing diagnostic assessments and measurements of critical cardiac structural and functional parameters, including the presence and severity (categorized as none, mild, moderate, or severe based on American Society of Echocardiography recommendations) of valvular pathologies such as mitral, tricuspid, and aortic regurgitations, and aortic stenosis. The software also measures Left Ventricular Ejection Fraction (LVEF), right and left ventricular dimensions, Right Ventricular Fractional Area Change (RV FAC), atrial areas, ascending aorta diameter, and Inferior Vena Cava (IVC) diameter. AISAP Cardio V1.0 is intended as an adjunctive tool to support qualified physicians in interpreting cardiac POCUS images, aiding in the assessment of major valvular findings and associated cardiac measurements, with the understanding that patient management decisions should not rely solely on its outputs and comprehensive echocardiography should be considered for suspected significant pathology. The device utilizes neural network models (classification, regression, and instance segmentation) trained on a vast dataset of over 140,000 exams, comprising more than 1 billion frames, from two academic institutions adhering to ASE guidelines; this training data encompassed adults with a mean age of 67 years and a BMI range of 15-49, alongside various comorbidities. Clinical performance testing demonstrated high accuracy: standalone structural/functional measurement studies on 200 cases showed low Root Mean Square Error (RMSE) for LVEF (6.20%) and IVC maximal diameter (0.25 cm), among other parameters. Standalone valvular pathology assessment on 329 cases yielded high Area Under the Curve (AUC) values (e.g., 0.975 for mitral regurgitation, 0.993 for aortic regurgitation) and strong sensitivity/specificity for detecting greater than mild pathologies. A clinical reader study involving 260 cases confirmed that physician interpretations aided by the device showed significantly improved AUC, accuracy, and Kappa agreement compared to unaided interpretations across all four assessed valvular conditions, while the system's automated cardiac view classification achieved over 98% accuracy.

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 AISAP Cardio V1.0 vary across POCUS images acquired by operators with different levels of experience (e.g., novice vs. expert POCUS users)?; What is the impact of specific patient comorbidities, beyond those listed in the training data (e.g., severe arrhythmias, prior complex cardiac surgeries), on the accuracy of the software's valvular assessments and measurements?; How does AISAP Cardio V1.0's diagnostic accuracy for valvular pathologies and cardiac measurements compare directly to comprehensive transthoracic echocardiography (TTE) performed in a standard echo lab setting, particularly in borderline or complex cases?; What are the specific characteristics of the "more than 1 billion frames" and "over 140,000 individual exams" used for training, particularly regarding the diversity of ultrasound machine vendors, image quality, and prevalence of specific pathologies within the training set?; How does the integration of AISAP Cardio V1.0 into clinical workflows affect the overall time to diagnosis and physician decision-making efficiency in real-world point-of-care settings?