VBrain-OAR

VBrain-OAR is a software tool designed to assist radiotherapy professionals by automating the initial contouring of organs at risk in the brain on T1 contrast-enhanced MRI images and performing rigid image registration of volumetric medical data. It utilizes deep learning neural networks for segmentation and an intensity-based algorithm for registration, aiming to enhance efficiency in treatment planning while ensuring clinical personnel retain final approval of the generated anatomical structures.

VBrain-OAR is a software device designed to assist trained radiotherapy personnel, such as radiation oncologists and medical physicists, in the planning of brain tumor radiation therapy for adult patients. Its primary function is to provide initial automated contours of critical organs at risk (OARs) – specifically the brain stem, eyes, optic nerves, and optic chiasm – on axial T1 contrast-enhanced brain MRI images. Additionally, VBrain-OAR includes an automatic image registration feature that allows for rigid registration of volumetric medical image data, such as MR and CT scans, to align the spatial position and orientation of two images. The core of VBrain-OAR's contouring capability lies in its artificial intelligence algorithm, specifically deep learning neural networks, which process the MRI images to segment the OARs. The system integrates into the clinical workflow via a PACS network, pulling patient scans, executing the contouring and registration algorithms, and exporting the generated contours as DICOM-RT objects. While it aims to streamline the initial contouring phase, the software is intended for informational purposes only; radiotherapy personnel must review, confirm, and modify the generated contours as necessary using external DICOM-RT compatible systems. Non-clinical testing demonstrated clinically acceptable contouring and registration performance across diverse datasets, including varied patient sex and imaging hardware.

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 VBrain-OAR's deep learning algorithm for OAR contouring, quantified by metrics like Dice Similarity Coefficient or Hausdorff distance, compare to expert manual contouring for each of the specified organs at risk (brain stem, eyes, optic nerves, optic chiasm)?; What was the diversity and size of the training dataset used for the deep learning model, and how does its performance vary when applied to MRI data acquired from different scanner vendors or unconventional imaging protocols not extensively represented in the training?; What is the quantifiable impact of VBrain-OAR on the efficiency of the radiation therapy planning workflow, specifically the time reduction achieved in OAR delineation, and how does this affect the overall turnaround time for plan generation?; Are there known scenarios, such as specific post-surgical changes, presence of artifacts, or rare anatomical variants of the OARs, where the AI algorithm's performance is significantly degraded, necessitating more substantial manual correction?; Considering VBrain-OAR performs rigid image registration, what are its limitations in aligning images with significant non-linear distortions or when fusing datasets with different patient positioning, compared to deformable registration techniques?