IRISeg

IRISeg is a standalone software application that processes DICOM-formatted contrast-enhanced CT images of kidneys to create 3D segmentation models for surgical planning and intraoperative display. It provides both manual editing tools and machine learning-based auto-segmentation for four kidney structures (parenchyma, artery, vein, collecting system), with output intended for visual, non-diagnostic use by qualified professionals.

IRISeg is substantially equivalent to its predicate (IRIS 1.0 System, K182643) in classification, regulation, host hardware compatibility, user interface design, supported input/output formats, and visualization features. The key difference is that IRISeg's ML auto-segmentation targets only four structures (excluding masses), whereas the predicate supported five; IRISeg also uses a standalone rather than integrated system architecture.

Software development followed IEC 62304 Edition 1.1 (2015-06). Risk management complied with ISO 14971 and AAMI CR34971 for AI/ML applications. Cybersecurity evaluation used FDA's September 2023 guidance. ML algorithm performance was characterized using Sørensen–Dice coefficient (DSC) and Mean Distance to Agreement (MDA) metrics against consensus radiologist segmentations.

IRISeg is substantially equivalent because it serves the identical intended use (kidney CT segmentation for preoperative planning and intraoperative display), operates through equivalent principles (manual and auto-segmentation workflows), and demonstrated comparable performance metrics (DSC ranges 0.87–0.97 across kidney structures) to the predicate device. The narrowing of auto-segmentation to exclude masses does not affect equivalence because users must segment masses manually in both devices; the standalone architecture and new ML implementation represent expected evolution that does not raise new safety or effectiveness questions.