AI-Rad Companion Organs RT

AI-Rad Companion Organs RT is a medical image processing software that employs deep learning algorithms to automatically segment and contour predefined organs-at-risk from CT and MR scans for radiation therapy planning. Its purpose is to streamline the contouring process, providing DICOM-RTSTRUCT outputs that require review and confirmation by medical professionals before use in treatment planning systems.

AI-Rad Companion Organs RT is a post-processing software designed to automatically contour predefined anatomical structures, specifically Organs-at-Risk (OAR), from DICOM CT and MR medical images. It utilizes deep-learning-based algorithms to generate these contours, which are outputted in DICOM-RTSTRUCT format. The intended application for these contours is as input for clinical workflows, most notably for external beam radiation therapy treatment planning. It is crucial that trained medical professionals use this software in conjunction with Treatment Planning Systems (TPS) or other interactive contouring applications to review, edit, and ultimately accept the generated contours. The software is not designed for automatic lesion detection or contouring and is indicated for use in adult patients only. The system's capabilities have been updated to include MR contouring algorithms (a T1 model for 3 target organs and a T2 model for 6 target organs) and an enhanced CT contouring algorithm now covering an additional 38 organs, bringing the total to 166 OARs for CT. The MR contouring functionality has been validated specifically using data from Siemens Healthineers’ scanners. Performance validation for MR contouring on 66 cases showed an average Dice coefficient of 85.75% and an ASSD of 1.25mm. For CT contouring, validation on 414 cases yielded Dice coefficients such as 76.5% for Head & Neck, 82.1% for Thorax, 88.3% for Abdomen, and 84.0% for Pelvis. Ground truth annotations for validation were established through manual delineation by experienced annotators following RTOG guidelines and reviewed by board-certified radiation oncologists, with training and testing datasets kept independent.

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 MR contouring algorithms vary across different Siemens Healthineers' MR scanner models and field strengths, given its validation is specific to this manufacturer's data?; What is the quantifiable impact on contouring time and inter-observer variability when integrating AI-Rad Companion Organs RT into a standard radiation therapy planning workflow?; Are there specific anatomical sub-regions or organ types, particularly among the 38 new CT organs or 9 MR organs, where the deep learning algorithms consistently require more significant manual editing, and what are the underlying reasons?; Given the software is not intended to automatically detect or contour lesions, what mechanisms or warnings are in place if pathological tissue significantly alters organ morphology, potentially impacting contour accuracy?; How were the 'ground truth' annotations established for the diverse international datasets used in CT algorithm validation, and what was the inter-rater reliability among the annotators and reviewing radiation oncologists?