TruPlan Computed Tomography (CT) Imaging Software

TruPlan CT Imaging Software assists in pre-procedural planning and post-procedural evaluation of Left Atrial Appendage Closure (LAAC) by providing advanced visualization, segmentation, and measurement tools from CT scans. The device leverages machine learning algorithms, trained on diverse datasets, for automated left heart segmentation and initial landing zone detection, which are then subject to user verification and refinement.

TruPlan Computed Tomography (CT) Imaging Software is a medical device designed to assist qualified users with image-based pre-procedural planning, sizing, and post-procedural evaluation for Left Atrial Appendage Closure (LAAC) procedures in adult patients. It utilizes pre-existing CT data, manually uploaded by the user, to enable visualization and measurement of heart and vessel structures. The software provides functionalities such as segmentation of cardiovascular structures, various image reconstruction and visualization techniques (2D review, Volume Rendering, MPR), simulation of TEE, ICE, and fluoroscopic views, along with measurement and annotation tools (distance, perimeter, area, angle, signal intensity, coordinates), and reporting capabilities. TruPlan is intended as an aid to the existing standard of care, not a replacement for current planning or follow-up applications. The software incorporates machine learning for two key functions: Left Heart Segmentation, which generates a 3D rendering of the left heart (left ventricle, left atrium, LAA) for visualization, and Landing Zone Detection, which initializes the landing zone for LAAC device sizing, requiring user review and correction. The Left Heart Segmentation algorithm was trained on 113 cases and validated on 533 anonymized patient images, achieving 99.81% bone removal and 97.37% LAA visualization. The Landing Zone Detection algorithm was trained on 273 cases and validated on 100 samples, with landing zone plane distance within 10 mm in 97% of cases and contour center distance within 12 mm in 99% of cases. Training data for these models was sourced from multiple clinical sites and CT scanner vendors to ensure model generalization.

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 TruPlan's machine learning algorithms for Left Heart Segmentation and Landing Zone Detection vary across different CT scanner manufacturers and imaging protocols not extensively represented in the training/validation datasets?; What is the typical learning curve for clinicians to become proficient in using TruPlan, particularly the manual correction of machine learning-generated initializations?; Beyond the reported accuracy metrics for segmentation and detection, what is the quantifiable impact of TruPlan on procedural planning time and inter-observer variability compared to traditional methods?; Are there any known anatomical variations of the LAA or patient characteristics where the machine learning algorithms exhibit reduced performance or require more significant manual user adjustment?; Could the machine learning models in TruPlan be adapted or retrained for pre-procedural planning of other cardiovascular interventions beyond LAAC?