FEops HEARTguideTM, FEops HEARTguideTM ALPACA

FEops HEARTguide™ ALPACA is a software system that employs artificial intelligence, specifically nonadaptive machine learning with human oversight, within a service-based model to provide clinicians with detailed anatomical visualization and quantitative measurements from CT scans for preprocedural planning of structural heart interventions like LAAO and TAVI. Its methodology relies on AI-driven segmentation and measurement tools, which have shown high accuracy in non-clinical validation studies for key anatomical parameters essential for these cardiac procedures.

FEops HEARTguide™ ALPACA is a software system designed to assist clinicians in preprocedural planning and sizing for structural heart interventions, specifically Left Atrial Appendage Occlusion (LAAO) and Transcatheter Aortic Valve Implantation (TAVI) procedures. It operates on a service-based model where clinicians submit patient CT data, and FEops prepares an anatomical analysis which is delivered via a web application. This analysis includes interactive 3D visualizations, DICOM Multiplanar Reconstructions (MPR), segmentation of cardiovascular structures, measurement and annotation tools, and reporting capabilities, allowing clinicians to review results and plan interventions. The system is intended for use by qualified clinicians as an adjunct to their professional judgment and other evaluations, and not as a replacement for implant device instructions for use. The core science of FEops HEARTguide™ ALPACA involves artificial intelligence, specifically nonadaptive machine learning algorithms, for the segmentation of cardiovascular structures from CT data. This automated segmentation is followed by human supervision and a quality check by a FEops Case analyst, with clinicians also able to review and modify the results. Non-clinical performance testing on retrospective cohorts demonstrated the system's accuracy. For LAAO planning, the semi-automatically identified landing zone diameter showed a mean difference of 1.4% ± 4.6% compared to ground truth, meeting a ±18% performance goal. For TAVI, the semi-automatically identified aortic annulus perimeter-based diameter showed a mean difference of 0.5% ± 1.9% against ground truth, meeting a ±10% goal. Segmentation accuracy, measured by Dice scores, was also high for both LAAO (mean 0.98 for semi-automatic) and TAVI (mean 0.97 for semi-automatic) analyses.

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 nonadaptive machine learning algorithms in FEops HEARTguide™ ALPACA vary across different CT scanner manufacturers and imaging protocols not included in the validation cohorts?; What is the typical turnaround time for the service-based model, from clinician data submission to FEops delivering the anatomical analysis, and how might this impact urgent cases?; The document mentions 'human supervision and a quality check by a FEops Case analyst' for AI segmentation. What are the specific qualifications and training protocols for these analysts, and what is the process if a discrepancy is found by the reviewing clinician?; Given that the fully automatic LAAO landing zone diameter had a superior Limit of Agreement of 18.3% and its confidence interval extended to 22.6%, potentially exceeding the ±18% performance goal, what guidance is provided to clinicians when interpreting these fully automatic outputs?; How does FEops HEARTguide™ ALPACA integrate or compare with the FEops HEARTguide™ Simulation Application (K214066) in terms of workflow and combined utility for preprocedural planning?