Contour ProtégéAI+

Contour ProtégéAI+ is software that automatically creates anatomical contours on medical images using machine-learning algorithms. Trained medical professionals use it to process CT and MR images for applications including quantitative analysis, adaptive therapy planning, image registration, and archiving contours for patient follow-up. Results must be reviewed and edited if necessary using appropriate visualization software.

The subject device includes one new 4.3.0 MR Brain model using the existing single 3D U-Net architecture from the predicate, and one updated 5.0.0 CT Male Pelvis model using a new multi-stage architecture with three distinct neural networks for progressive localization and segmentation. Both operate on server-based systems supporting Linux, Windows, or Mac, with cloud or local deployment options.

Testing used Dice coefficient and Mean Distance to Agreement (MDA) metrics, with non-inferiority testing comparing to the MIM Maestro (K071964) reference device. Acceptance criteria referenced American Association of Physicists in Medicine (AAPM) recommendations and prior cleared devices (Radformation AutoContour RADAC V4 K242729, GE HealthCare MR Contour DL K242925). Added Path Length (APL) correlation to editing time was evaluated per Vaassen et al. 2020.

The 4.3.0 model uses identical architecture to the predicate with no design changes. The 5.0.0 model uses a new multi-stage framework but each component is based on the same 3D U-Net core; it was tested using the same procedures and acceptance criteria as the predicate and demonstrated equivalent or superior performance. Both models met acceptance criteria (two or more of: Dice non-inferiority, MDA non-inferiority, or user evaluation ≥3 on 5-point scale), with cumulative APL non-inferior to reference.