AI-Rad Companion Organs RT

AI-Rad Companion Organs RT is a software tool designed to automate the contouring of organs at risk (OAR) in DICOM CT images for adult radiation therapy planning. Its methodology employs deep-learning algorithms to process CT data and output RTSTRUCT files, with the science behind its performance validated through studies showing a median DICE score >0.80 and a median 95% Hausdorff distance of 2.0mm for OAR segmentation.

AI-Rad Companion Organs RT is a specialized post-processing software designed for adult patients, which leverages deep-learning algorithms to automatically delineate organs at risk (OAR) from DICOM CT imaging data. Its core methodology involves processing these CT scans to generate RTSTRUCT objects, which serve as initial contours for external beam radiation therapy treatment planning. The system operates with an automated workflow, supports configurable organ templates, and provides a web-based interface for clinicians to preview the AI-generated contours before they are used in subsequent clinical software for review, potential editing, and acceptance. The software's scientific basis in deep learning was validated through non-clinical testing, including a retrospective study on 113 multi-vendor CT datasets, demonstrating its efficacy in OAR segmentation. Performance results indicated a median DICE coefficient exceeding 80% across all contoured OARs and a median 95% Hausdorff Distance of 2.0 mm, metrics comparable to its predicate. AI-Rad Companion Organs RT is designed for compatibility with various DICOM-compliant CT scanners and treatment planning systems but is explicitly not for lesion detection or contouring, and all its outputs require trained medical professional oversight.

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 deep learning model's performance vary across different CT scanner manufacturers and imaging protocols not included in the 113 validation cases, particularly for organs with subtle boundary definitions?; What is the typical time saved in a clinical workflow by using AI-Rad Companion Organs RT for automated OAR contouring compared to fully manual contouring, and how does this impact overall treatment planning efficiency?; Given that the software only allows for preview and not editing of contours, what is the anticipated rate of contour rejection or need for significant downstream editing, and are there specific OARs more prone to requiring such modifications?; How does the algorithm handle anatomical variations or post-surgical changes in adult patients, and what is the robustness of the organ template configuration in adapting to such cases?; What are the specific failure modes of the deep learning algorithm, and how does the system alert users to potentially suboptimal or erroneous contours beyond the web-based preview requiring rejection?