MR Contour DL

MR Contour DL is a post-processing software application that uses deep learning algorithms to automatically generate organ-at-risk contours from MR images in the form of DICOM Radiotherapy Structure Sets (RTSS) for radiation therapy planning. The software is designed to contour 37 different anatomical structures in the head/neck and pelvis regions for adult patients, with the contours intended for review and editing by trained medical professionals before use in treatment planning.

Both the proposed device and predicate device employ deep learning algorithms for automated organ segmentation, target adult patients only, use automated workflow with no direct user interface interaction beyond configuration settings, and generate RTSS DICOM outputs for review at workstations. The key difference is modality: MR Contour DL segments MR images while the predicate (Auto Segmentation) segments CT images. Both are deployed on server-based platforms and cover head/neck and pelvis anatomical regions.

The device was developed under ISO 13485 and 21 CFR 820 quality system regulations. Testing used DICE Similarity Coefficient (DSC) and Hausdorff Distance 95th percentile (HD95) as performance metrics. Acceptance criteria were organ-size-dependent DSC thresholds (small 50%, medium 65%, large 80%). Clinical validation employed a Likert scale (1-5) for reader assessment of contour adequacy.

MR Contour DL achieves substantial equivalence through identical fundamental technology (deep learning segmentation), equivalent intended use (workflow tool for initial anatomy segmentation to accelerate radiation therapy planning after user confirmation), and overlapping anatomical coverage (head/neck and pelvis) with the predicate. Performance testing demonstrated that MR Contour DL achieved improved or equivalent HD95 values in 24 of 28 organs analyzed with a mean HD95 of 4.7 mm compared to the predicate device. Clinical reader study showed mean Likert scores of 3.0 or above for all organs, confirming adequacy for radiotherapy planning. The modality difference (MR versus CT) does not raise new safety or effectiveness concerns, as both reference devices demonstrate equivalent performance across both modalities.