Contour ProtegeAI

Contour ProtégéAI is a software tool utilizing machine learning algorithms to automatically segment anatomical structures in CT and MR medical images for trained professionals, supporting tasks like quantitative analysis and radiation therapy planning. Its scientific basis involves neural networks trained on diverse image datasets, with its segmentation accuracy validated against benchmark technologies using the Dice coefficient in non-inferiority testing.

Contour ProtégéAI is a software accessory for MIM software, designed to assist trained medical professionals by automatically generating contours of anatomical structures from CT and MR medical images. It utilizes machine-learning algorithms to process DICOM 3.0 compliant images, aiding in applications such as quantitative analysis, adaptive therapy, transferring contours to radiation therapy treatment planning systems, and archiving for patient follow-up. The system specifically segments normal structures across various CT anatomical locations and normal structures of the prostate, seminal vesicles, and urethra within T2-weighted MR images. It is important to note that Contour ProtégéAI is not intended for automatic lesion detection or contouring and requires users to review and, if necessary, edit the automatically generated contours using the main MIM software. The device operates on Windows, Mac, and Linux systems and is deployed on a remote server leveraging the MIMcloud service for data management and transfer. The underlying neural networks were trained on extensive datasets of CT and MR images with associated segmentations from multiple large institutions. For performance validation, the system was tested on 286 CT images and 72 MR images, sourced from institutions distinct from the training data. A non-inferiority test, comparing the mean Dice coefficient of contours generated by Contour ProtégéAI against those from a benchmark technology, established that Contour ProtégéAI was non-inferior by a limit of 0.1 Dice, a clinically acceptable difference, demonstrating its segmentation accuracy.

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 types of neural network architectures are utilized by Contour ProtégéAI, and how does the system account for variability in image acquisition parameters across different CT and MR scanners?; How robust is Contour ProtégéAI's segmentation performance when encountering anatomical variations or post-treatment changes, and what is the typical range of Dice coefficients observed for the specified anatomical structures?; What were the key characteristics of the institutions providing training data (e.g., geographic diversity, types of scanners, patient demographics), and how might this influence the model's generalizability?; Beyond the non-inferiority comparison based on the Dice coefficient, were any qualitative assessments by expert clinicians conducted to evaluate the clinical acceptability and utility of the generated contours?; What is the computational time required for Contour ProtégéAI to process a typical CT or MR dataset, and how does this compare to manual contouring time for the supported anatomical regions?