DASI Dimensions (V1.0)

DASI Dimensions (V1.0) is a clinical decision support software that processes patient CTA scans using an AI-enhanced workflow, incorporating proprietary algorithms and human quality assurance, to deliver precise anatomical measurements of cardiovascular structures for pre-procedural TAVR planning in aortic stenosis patients. The system's methodology relies on automated detection of aortic root control points and calculation of key dimensions, with its accuracy, reliability, and usability validated through comprehensive non-clinical testing.

DASI Dimensions (V1.0) is a standalone, non-invasive clinical decision support software designed for cardiologists and radiologists to aid in pre-procedural planning for Transcatheter Aortic Valve Replacement (TAVR) in patients with aortic stenosis. The system operates off-site, receiving patient-specific multiphase DICOM computed tomography angiography (CTA) chest image files uploaded by clinicians via a web portal. It then processes these images using proprietary algorithms, which include a static deep learning artificial intelligence (AI) model for detecting key aortic root control points, and calculates various anatomical measurements. DASI Simulations engineers perform quality checks on these AI-assisted detections and calculations before a report containing pre-defined images and measurements (diameters, lengths, angles, areas, perimeters) of cardiovascular structures like the aorta, aortic root, and aortic valve is released to the end-user for viewing and downloading. The core technology involves automated segmentation and centerline detection of cardiac structures, specifically targeting the aortic root, aortic valve, sinus of valsalva, sinotubular junction, and ascending aorta relevant to TAVR. Non-clinical validation studies demonstrated the system's performance. For instance, in a performance validation study, AI-generated control points achieved an 85.3% success rate (points within a 3mm error tolerance), exceeding the study's acceptance threshold. Automatic measurements, such as primary outputs like annulus area, showed mean percentage errors around 1% compared to clinician manual measurements, meeting acceptance criteria of ≤10% error in ≥95% of cases. Operator variability was low (ICC 0.96 for annulus area), and usability testing confirmed the interface was user-friendly. Sensitivity analysis indicated that key measurements like annulus area remained within a 10% error margin even with 3mm perturbations in control points. The software provides measurements but is not diagnostic, and clinicians retain full responsibility for all treatment decisions.

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 AI model in detecting aortic root control points vary across different patient anatomies or image qualities not extensively covered in the validation cohort?; What is the specific definition and clinical relevance of 'primary,' 'secondary,' and 'tertiary' measurements, and how were these categories determined for setting differential error acceptance criteria?; Given that DASI Simulations engineers perform quality checks, what is the typical intervention rate, and what are the common reasons for manual adjustments to the AI-generated control points or measurements?; How does the 'static' nature of the deep learning AI model accommodate evolving anatomical understanding or new imaging protocols without requiring frequent re-validations?; What are the performance metrics of DASI Dimensions when compared directly against the predicate device (K153736, 3mensio Workstation) using the same dataset, particularly for the overlapping functionalities?