Limbus Contour

Limbus Contour is a software medical device employing pre-trained machine learning models to automatically segment normal anatomical structures from CT and MR images, aiming to provide optimal contours for radiation treatment planning. Its methodology involves processing DICOM images and interfacing with existing treatment planning systems to enhance the efficiency and consistency of contour generation for trained medical professionals.

Limbus Contour is a software-only medical device designed for trained radiation oncologists, dosimetrists, and physicists to aid in radiation treatment planning. It utilizes pre-trained, expert-curated machine learning models to automatically segment (contour) normal anatomical structures from Computed Tomography (CT) and Magnetic Resonance (MR) DICOM images. The software's purpose is to derive optimal contours for input into treatment planning systems, supporting tasks such as creating, transforming, and modifying contours for adaptive therapy, archiving, and the localization and definition of healthy anatomy. Limbus Contour operates as a stand-alone application, interfacing with the user's operating system to import DICOM images and export segmented DICOM RT-Structure Set files, but relies on existing radiotherapy treatment planning systems (TPS) or DICOM viewers for image display and further modification of generated segmentations. It is not intended for use with digital mammography. The device, specifically version 1.7, segments tissues through its machine learning algorithms followed by post-processing of the model predictions. Key updates from its predicate version include an expanded list of anatomical structures available for automatic contouring, discontinuation of MacOS support (now Windows-only), and an increased RAM requirement to 8GB. Performance was established through comprehensive software verification and validation testing, including manual and automated tests. Specific validation of the automatic contouring function demonstrated that the software meets user needs and performs according to its stated requirements, ensuring the accuracy of the generated contours. This validation was based on benchtop performance testing rather than clinical studies.

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 machine learning architectures and training datasets were utilized for the expanded list of anatomical structures in Limbus Contour v1.7, and how is model performance validated across diverse patient cohorts and imaging protocols?; How does the post-processing algorithm applied to machine learning model outputs enhance contour accuracy, particularly for structures with low contrast or indistinct boundaries?; What quantifiable improvements in contouring time and inter-observer variability have been demonstrated in clinical settings when using Limbus Contour v1.7 compared to manual methods or its predicate version?; Considering Limbus Contour relies on external systems for image display, what are the key integration capabilities and potential compatibility challenges with various existing DICOM viewers and radiation treatment planning systems?; What mechanisms are in place for ongoing performance monitoring and model retraining for Limbus Contour, especially as new imaging techniques or clinical needs emerge?