MR Contour DL

The GE Healthcare MR Contour DL is a post-processing application utilizing deep learning algorithms to generate initial segmentations of organs at risk from MR images, aiming to accelerate radiation therapy planning workflows. Its development involved rigorous non-clinical bench testing, assessing segmentation accuracy with metrics like DICE score and HD95, and a clinical reader study to affirm its safety, effectiveness, and substantial equivalence for clinical application.

GE Healthcare's MR Contour DL has been reviewed and determined to be substantially equivalent to legally marketed devices, allowing it to be marketed for its intended use. This device is a post-processing application that uses deep learning algorithms to generate contours of organs at risk from MR images, producing a Radiotherapy Structure Set (RTSS) DICOM. It is designed to aid in radiation therapy planning by providing initial segmentations that accelerate workflow, with the understanding that trained medical professionals must verify and correct the contours as needed. The device's development followed stringent quality system regulations, including comprehensive design control testing. Non-clinical bench testing assessed its segmentation accuracy using metrics like DICE score and Hausdorff Distance (HD95) against a database of clinical MR exams, demonstrating performance that was either improved or equivalent to predicate devices for most organs. A clinical reader study further evaluated the generated contours, with qualified radiotherapy practitioners confirming their adequacy for radiation therapy planning. These assessments collectively indicated that the device introduces no new questions of safety or effectiveness, affirming its suitability for clinical application.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. What specific deep learning architecture and training data were utilized for MR Contour DL, and how might these impact generalizability across diverse patient populations and MR scanner types?; Beyond DICE score and HD95, what clinical validation metrics were used in the reader study to assess the "adequacy" of contours, and what was the inter-observer variability among the qualified radiotherapy practitioners?; How does the workflow acceleration provided by MR Contour DL translate into tangible improvements in clinic efficiency and patient throughput compared to manual contouring, and are there any documented cases of workflow disruption?; What is the process for updating or retraining the deep learning model within MR Contour DL, and how frequently are such updates expected to occur to maintain performance and adapt to new clinical data?; Are there specific anatomical regions or pathologies where MR Contour DL's segmentation performance was noted to be less accurate or required more significant manual correction by clinicians, and how will these limitations be addressed?