AI-Rad Companion Organs RT

AI-Rad Companion Organs RT is post-processing software that automatically contours pre-defined anatomical structures (organs-at-risk) from CT and MR medical images using deep-learning algorithms. The generated contours are intended for use in external beam radiation therapy treatment planning workflows and must be reviewed, edited, and accepted by trained medical professionals using treatment planning systems or interactive contouring applications before clinical use.

The subject device uses the same deep-learning-based algorithmic approach and software architecture as the predicate. Key modifications include an enhanced CT contouring algorithm covering 37 new organs and structures (203 total OARs vs. 166 in predicate) and multi-guideline support with corresponding UI updates. The MR algorithm remains unchanged from the predicate. Both devices process DICOM CT and MR data, support any DICOM-compliant treatment planning system, and provide edge and cloud deployment options.

Compliance verified against ISO 62366-1 (usability), ISO 14971:2019 (risk management), IEC 62304 (software lifecycle), NEMA DICOM PS 3.1–3.20 (digital imaging), ISO 15223-1 (symbols), IEC 82304-1 (health software safety), IEC 81001-5-1 (health IT system safety), and ISO 20417 (manufacturer information). Software validation per FDA guidance on device software functions (June 2023) and cybersecurity recommendations (September 2023).

The intended use is unchanged from the predicate. For existing organs in the predicate, the average Dice score difference is <3%, demonstrating equivalent segmentation performance. For the 37 new organs added, validation against 579 retrospective CT cases from multiple geographic sites shows performance metrics (DICE, ASSD, fail rates) comparable to or exceeding reference literature and cleared predicate devices using clinically appropriate baseline thresholds. The MR algorithm is identical and unchanged. Risk analysis supports equivalent safety and effectiveness, with no new safety concerns raised by the algorithmic enhancements or UI modifications.