Seg Pro V3 (RT-300)

Seg Pro V3 is a software device that assists trained radiation oncology professionals in automatically generating initial contours of organs at risk (OARs) on DICOM medical images to support radiation therapy treatment planning. The device uses deep learning algorithms to process CT or MR images and outputs DICOM-RT structure sets that must be reviewed and edited by clinicians before use; it is not intended for lesion detection or autonomous clinical decision-making.

Seg Pro V3 uses deep learning algorithms to segment OARs from DICOM images and operates on a Linux-based server without a user interface, automatically outputting DICOM-RT structure sets to a configurable target node. It supports both CT and MR modalities with 166 CT OARs and 16 MR OARs across head-and-neck, chest, and abdomen-pelvis regions, whereas the predicate device (EFAI HCAPSeg) supports only CT with 80 OARs. Both devices are fully automated, require DICOM-compliant treatment planning systems, and share the same core architecture and operational workflow.

Validation activities followed IEC 62304:2006/A1:2016 (Medical device software – Software life cycle processes), FDA Guidance on Content of Premarket Submissions for Software (2005), FDA Guidance on Content of Premarket submissions for Devices Software Functions (2021), and FDA Guidance on Management of Cybersecurity in Medical Devices. Device design and risk management comply with 21 CFR Part 820.30 and ISO 14971:2019.

Seg Pro V3 is substantially equivalent to the predicate EFAI HCAPSeg because both share the same intended use (assisting radiation oncologists with initial OAR contours for therapy planning), use identical deep learning technology, output DICOM-RT structure sets automatically, operate without user interfaces, and must be used with DICOM-compliant treatment planning systems. The addition of MR support does not raise new safety or effectiveness questions, as both modalities are DICOM images processed through the same contouring architecture, and performance validation demonstrated mean DSC of 0.85 overall with consistent metrics across anatomical regions.