ART-Plan

ART-Plan is a software tool developed to enhance radiotherapy planning by providing AI-powered automatic and semi-automatic contouring of anatomical structures and organs at risk on 3D medical images, along with multi-modal image registration capabilities. Its methodology integrates deep learning algorithms within the Annotate module for segmentation (primarily on CT images) and advanced registration techniques in the SmartFuse module to process and display DICOM data, emphasizing that all system outputs necessitate clinician review and validation to ensure accuracy for subsequent treatment planning.

ART-Plan is a software medical device designed to assist clinicians in the radiotherapy treatment planning process for cancer patients. Its core functionalities center on the contouring of target anatomical regions and organs at risk (OARs) on 3D medical images, primarily utilizing two modules: SmartFuse for image registration and Annotate for contouring. The SmartFuse module supports the registration of multi-modal images (CT, PET, MR) using both rigid and deformable techniques, displaying them in fused or non-fused views to aid image comparison and delineation. The Annotate module allows for manual, semi-automatic, and AI-driven automatic contouring, specifically employing deep learning algorithms to delineate OARs and healthy lymph nodes on CT images across various anatomical sites like the head and neck, thorax, abdomen, male pelvis, and brain. The system operates as a web-based platform compatible with DICOM 3.0 datasets, providing tools for image visualization, manipulation (such as windowing, zoom, MIP, MPR), and Regions of Interest (ROI) definition and measurement (including intensity, Hounsfield units, SUV, and 2D/3D size). Although ART-Plan aims to streamline workflows and enhance standardization through automation, it mandates that all contours and registrations generated automatically or semi-automatically undergo rigorous verification, potential modification, and final validation by trained medical professionals. Non-clinical testing confirmed the device's usability, the performance of its autosegmentation algorithms (including tests on European data, against AAPM requirements, MIM, on US patient data, and for external contours), and the quality of its image fusion and registration capabilities, with all tests reportedly passed.

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. The document mentions autosegmentation performance testing on 'US data'; however, the primary supported modality for segmentation is listed as CT. Can ART-Plan's Annotate module reliably perform automatic segmentation directly on ultrasound images for radiotherapy planning, or was this US data used in a different context (e.g., for fusion with CT)?; What specific deep learning architectures are employed in the Annotate module for automatic segmentation of organs at risk and lymph nodes, and what were the key characteristics (size, diversity, cancer types) of the datasets used for training and validation these models?; The 'Technical Information Comparison' table indicates 'None' for 'Region/Volume Quantification' for ART-Plan, yet it lists 'Region/volume of interest measurements' including SUV and volume. Could you clarify the extent of quantitative analysis ART-Plan supports beyond basic measurements, particularly in comparison to features like RECIST/PERCIST analysis found in predicate devices?; How does ART-Plan's performance in AI-driven automatic contouring and deformable image registration, particularly in terms of geometric accuracy (e.g., Dice score, Hausdorff distance, target registration error) and processing time, quantitatively compare to the primary predicate AccuContour and other listed predicates?; The document states that ART-Plan's automatic segmentation includes 'healthy lymph nodes.' Given their variability and typically small size, what is the validated accuracy and robustness of this feature across different anatomical regions (e.g., head & neck, pelvis) and varying patient anatomies?