AI-Rad Companion Organs RT

AI-Rad Companion Organs RT is a medical image post-processing software that employs deep-learning algorithms to automatically contour Organs at Risk from adult DICOM CT scans, aiming to streamline radiation therapy treatment planning. Its methodology involves processing CT data to generate RTSTRUCT objects for review and integration into existing clinical workflows, with performance validated through metrics like DICE coefficient and ASSD against predicate and reference devices.

AI-Rad Companion Organs RT is a post-processing software designed to automatically contour anatomical structures from DICOM CT imaging data using deep-learning algorithms. Its primary application is to assist in clinical workflows, particularly for external beam radiation therapy treatment planning, by generating initial organ contours. The software processes CT input data and produces RTSTRUCT objects, which are intended for use by trained medical professionals. These professionals must use the output in conjunction with Treatment Planning Systems (TPS) and interactive contouring applications to review, edit, and ultimately accept the generated contours. The system supports configuration of an organ database and templates, and includes a web-based preview for accepting or rejecting contours, though editing is performed in separate applications. A significant update in the current version (VA40) is the addition of automated contouring for 29 new head and neck structures, expanding its capability to 108 Organs at Risk (OARs), compared to 79 in the previous version. The deep learning algorithm has been enhanced, and the software can now be deployed on edge devices as well as in the cloud. Performance validation was conducted using a retrospective study with 113 CT datasets, demonstrating that the software achieved a median DICE score of 0.85 and a median absolute symmetric surface distance (ASSD) of 0.93 mm for existing organs, comparable to its predicate. For new head and neck lymph node contouring, it achieved an average DICE score of 81.32%, which was found comparable to a reference device's performance on similar structures. The software is intended for adult patients only and does not automatically detect or contour lesions.

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 (DICE, ASSD) of AI-Rad Companion Organs RT vary across different CT scanner manufacturers and acquisition protocols not extensively covered in the validation cohorts?; What is the typical time-saving achieved in a clinical radiation therapy planning workflow when using AI-Rad Companion Organs RT for initial contouring compared to fully manual contouring, considering the necessary review and edit time?; Given that the models are locked, what is the planned lifecycle for model updates or retraining to incorporate new anatomical variations or improve accuracy on less common cases?; How robust is the algorithm in differentiating between Organs at Risk and nearby pathological tissues, especially when anatomical boundaries are distorted by tumors, since it is not intended to contour lesions?; What specific feedback mechanisms are in place for users to report contouring inaccuracies, and how is this feedback utilized for future algorithm enhancements or user training?