Home

The Open Ground Truth Training Network project (openGTN)

The openGTN project investigates automatic segmentation of medical images acquired with Magnetic Resonance Imaging (MRI). MRI is a powerful technique to visualize and quantify human anatomy and function, with the ultimate goal to support diagnosis and treatment of disease. In the early days of MRI, clinical users primarily interpreted the images visually, but nowadays more and more automatic image segmentation algorithms are used to derive the essential information.

Example MRI brain segmentation openGTN

Image segmentation

The design of accurate and robust MR image segmentation algorithms is a very challenging task, due to the highly varying appearance of anatomical structure in the MR images. This variation can be caused by differences in the scan protocol (scanner parameter settings), but also by patient variations (e.g. differences in weight) and by scanner software and hardware variations.

The development, optimization, and validation of image segmentation algorithms is a tedious and time-consuming process. Especially novel model-based methods that use machine learning approaches, such as deep learning, require large training sets of annotated data. There is a clear need to develop MRI segmentation methods that are as much as possible insensitive to the specific MR image contrast (T1-weighted, T2-weighthed, etc.) and to patient and scanner variations. The development of such algorithms is the focus of openGTN project. Their availability would significantly reduce training and validation efforts and broaden the applicability in clinical practice.

MRI simulation

To facilitate the development of MR image contrast insensitive algorithms, openGTN will develop methods for the generation of large quantities of simulated magnetic resonance image (MRI) data, with annotations of relevant anatomies, needed for the training and validation of image analysis algorithms (registration, segmentation, quantification), especially for machine learning approaches such as deep learning using convolutional neural networks. The project aims to make the simulated MRI data and annotations available in public databases, which can be used for optimizing and benchmarking image analysis algorithms (often called “algorithm challenges”, see e.g. grand-challenge.org).

Funding

This research is funded by the highly prestigious and competitive Marie Curie Innovative Training Networks (ITN) fellowship program (project 764465).

More information can be found here.