ART-Plan (v.2.2.0)

ART-Plan (v.2.2.0) is a software platform designed to assist medical professionals in radiotherapy treatment planning by providing tools for image contouring, multi-modal image registration, and data-driven support for treatment adaptation decisions. It employs deep learning algorithms for automated anatomical segmentation and the generation of synthetic CT images from MRI and CBCT, alongside a validated dose calculation engine, to enhance the precision and efficiency of radiotherapy for cancer patients.

ART-Plan (v.2.2.0) is a software platform designed for radiotherapy treatment planning for cancer patients (excluding those under 18 years of age) and is intended for use by trained medical professionals such as radiotherapists, oncologists, and physicists. The system facilitates improved workflow through modules for patient data management (Home), region of interest contouring and pseudo-CT generation from MRI (Annotate), multi-modal image registration (SmartFuse), and decision support for treatment replanning (AdaptBox). Core functionalities include manual, semi-automatic, and deep-learning driven automatic segmentation for various anatomical sites including head/neck, thorax/breast, abdomen, pelvis, and brain, across CT, MR, and synthetic CT images. A key feature of ART-Plan (v.2.2.0) is its use of deep learning for generating synthetic CT (sCT) images from MRI (pelvis male, brain) and, notably with this version, from CBCT for pelvis male, head/neck, and thorax/breast, aiding adaptive radiotherapy. The AdaptBox module leverages these sCTs from daily CBCTs, performs auto-delineation, and calculates dose on both planning CT and sCT to compare metrics for replanning decisions. Extensive validation confirmed its clinical acceptability: auto-segmentation met Dice Similarity Coefficient (DSC ≥0.8 or inter-expert criteria) and Hausdorff Distance (HD95 ≤5.75mm) targets; synthetic CTs demonstrated geometric accuracy (Jacobian Determinant ≈1 ±5%) and dosimetric non-inferiority (e.g., DVH parameter differences <2%); and the dose engine showed comparable accuracy to existing solutions, with all performance tests successfully meeting predefined acceptance criteria.

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 deep-learning based synthetic CT generation from CBCT vary across the newly added anatomies (Thorax, Breast, Head & Neck) compared to previously supported ones like pelvis male?; The document mentions DSC (mean)≥ mean (DSC inter-expert) + 5% as an acceptance criterion; what specific inter-expert DSC values were used for different organs, and how were these reference values established?; Given that the AdaptBox module relies on user-defined metrics for replanning decisions, what measures are in place or recommended to ensure consistency and minimize inter-user variability in defining these critical metrics?; What are the typical processing times for the deep learning-based auto-segmentation and synthetic CT generation for various anatomies, and how do these times scale with image resolution or complexity?; Beyond comparison to commercial solutions for the dose engine validation, what were the specific clinical dosimetric criteria used and how does its performance compare to gold-standard phantom measurements or Monte Carlo simulations, especially for synthetic CT inputs?