MVision AI Segmentation

MVision AI Segmentation is a software tool utilizing pre-trained deep learning models to automate the initial contouring of anatomical structures on CT images for radiation therapy planning. Its purpose is to streamline the delineation process, providing a template for clinicians to review and refine, thereby aiming to reduce manual effort and expedite treatment preparation without performing diagnostic functions.

MVision AI Segmentation is a software-only medical device designed to assist in radiation therapy treatment planning by automating the contouring of anatomical structures on CT images. It utilizes pre-trained, static deep artificial neural network models to generate initial segmentation templates for various regions of interest across multiple anatomical sites, including the head and neck, brain, breast, lung, abdomen, and pelvis in adult patients. These automatically generated contours are intended to serve as a starting point for medical professionals, who must then visualize, review, modify, and approve them within their existing clinical workflows before any further use. The system can be deployed either on-premises, integrating with local healthcare IT networks via DICOM TCP/IP, or in a cloud environment using secure HTTPS for de-identified data transfer. The core technology relies on machine learning algorithms trained on hundreds of CT scans per region of interest, adhering to established segmentation consensus guidelines. While no new animal or clinical trials were conducted for this device, performance evaluation was based on "golden datasets" collated from diverse sources to ensure the models' generalizability and robustness, particularly for the US patient population. Validation studies conducted by radiotherapy experts indicated that the software significantly reduces the initial time and effort required for contouring, thereby allowing clinicians to proceed more quickly to subsequent stages of treatment planning. It is important to note that MVision AI Segmentation is not designed for detecting lesions or tumors, nor is it intended for use in real-time adaptive planning workflows.

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. How does the performance of the deep learning models vary across different CT scanner manufacturers and imaging protocols, and what is the minimum image quality required for optimal segmentation?; What specific "segmentation consensus guidelines" were the models trained to comply with, and how is inter-observer variability in manual contouring addressed during model training and validation?; Can the system's pre-trained models be fine-tuned or adapted with local institutional data to further improve accuracy for specific patient populations or rare anatomical variations, and if so, what mechanisms are in place?; What are the quantitative metrics (e.g., Dice Similarity Coefficient, Hausdorff Distance) achieved for each of the supported anatomical regions of interest when compared against the "golden dataset," and how do these compare to the predicate device or manual contouring?; Considering the cloud deployment option, what are the detailed data de-identification processes and specific cybersecurity measures beyond TLS encryption to ensure patient data privacy and system integrity, especially against evolving cyber threats?