Limbus Contour

Limbus Contour is a software medical device that utilizes pre-trained machine learning models to automatically segment healthy anatomical structures in CT and MR images for radiation oncology. Its methodology centers on providing accurate contours as input for external radiation treatment planning systems, thereby enhancing efficiency in radiotherapy preparation by automating a critical delineation step, while specifically excluding tumor contouring.

Limbus Contour is a software-only medical device designed for use by trained radiation oncologists, dosimetrists, and physicists. Its primary purpose is to assist these professionals in deriving optimal contours of healthy anatomical structures from Computed Tomography (CT) and Magnetic Resonance (MR) images for input into radiation treatment planning. The software operates in conjunction with existing radiation treatment planning systems (TPS) or DICOM viewing systems, facilitating the loading, saving, and transfer of contours, aiding in applications such as adaptive therapy and archiving for patient follow-up, though it does not display images or contours itself, requiring external software for these functions. The device employs pre-trained and expert-curated machine learning models for the automatic segmentation of normal tissues, with additional post-processing applied to the model prediction outputs to refine the results. Limbus Contour interfaces with the user's operating system to import DICOM image files and export the generated contours as DICOM RT-Structure Set files. It is explicitly not intended for use with digital mammography or for the automatic contouring of tumors or tumor clinical target volumes, focusing solely on the localization and definition of healthy anatomy to streamline the initial stages of radiotherapy planning.

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 datasets were used for training and curating the models for different anatomical structures, and how is model performance monitored post-deployment?; How does the post-processing algorithm refine the initial machine learning model predictions, and can users adjust its parameters?; What were the quantitative performance metrics (e.g., Dice similarity coefficient, Hausdorff distance) achieved during the 'Automatic Contouring – Validation Test' across various anatomical structures and imaging modalities compared to expert-defined contours?; Given its reliance on external Treatment Planning Systems (TPS) or DICOM viewers, what is the typical workflow integration time and compatibility assurance process with commonly used commercial and open-source TPS?; Since the device exclusively contours healthy anatomical structures and not tumors, how does it ensure robust differentiation at the interface between healthy tissue and potential pathological areas to avoid mis-segmentation that could impact subsequent tumor delineation in the TPS?