MRCAT Head & Neck

MRCAT Head & Neck is a software application for Philips Ingenia MR systems that generates synthetic CT-like images for radiotherapy treatment planning of head and neck soft tissue tumors. It employs an mDixon MR imaging sequence and a machine learning algorithm, featuring a convolutional neural network, to automatically characterize tissue properties and assign Hounsfield Unit values for radiation attenuation estimation.

The MRCAT Head & Neck is a software application designed for Philips' Ingenia, Ingenia Ambition, and Ingenia Elition MR systems. Its primary purpose is to provide operators with tissue property information for estimating radiation attenuation, which is crucial for photon external beam radiotherapy treatment planning. Specifically, MRCAT Head & Neck is indicated for radiotherapy treatment planning in patients diagnosed with soft tissue tumors located in the Head and Neck region; this software is available as an optional component of the Ingenia MR-RT package, facilitating MR-only simulation workflows by generating synthetic CT images. The system operates by automatically generating MRCAT images on the MR console, with embedded image post-processing occurring in the background concurrently with image acquisition using an mDixon imaging protocol (T1-weighted Fast Field Echo 3D dual echo). The core MRCAT algorithm leverages machine learning, specifically a pre-trained and locked convolutional neural network (CNN), for automatic tissue characterization. It segments bones and the body outline from mDixon in-phase and water images, then assigns a continuum of Hounsfield Unit (HU) values to tissues based on their fat and water intensities, thereby creating CT-like density information. The HU values are calibrated using registered CT images, and sanity checks ensure the validity of the generated images, which can be exported in DICOM format. Clinical data indicates that the dose accuracy of MRCAT Head & Neck is similar to CT-based planning, with PTV dose differences being clinically insignificant, affirming its utility for providing spatially accurate radiation attenuation estimates for the head and neck region.

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 machine learning-based bone segmentation in MRCAT Head & Neck compare to manual segmentation or other automated methods in terms of accuracy and efficiency for complex anatomical variations within the head and neck region?; What is the validated range of Hounsfield Unit accuracy across different tissue types (e.g., dense bone, soft tissue, air cavities) in the Head & Neck region using MRCAT, and how does this compare to conventional CT simulation, particularly at challenging tissue interfaces or in the presence of motion?; Given that the Convolutional Neural Network model is locked after training, what is the manufacturer's protocol and timeline for model updates or retraining if new imaging data, hardware changes, or clinical outcomes suggest a need for refinement, especially considering diverse patient populations or evolving treatment techniques?; How robust is the mDixon sequence and MRCAT algorithm in handling common MR artifacts in the head and neck region (e.g., susceptibility from dental implants, motion artifacts), and what are the quantitative impacts of such artifacts on the accuracy of the generated synthetic CT values and subsequent dose calculations?; What specific non-clinical and clinical validation studies were performed to establish the geometric accuracy and dose calculation fidelity of MRCAT Head & Neck across the full range of indicated patient anatomies and tumor types within the head and neck, beyond the comparison to the MRCAT Brain predicate?