Image & Data Analysis

Competences

 

Software

MEVIS draw

MEVIS draw is a free software provided by Fraunhofer MEVIS for creating and editing segmentations on 3d medical images such as CT or MRI.

 

Machine Learning

With increasing availability of large amounts of medical data, there is a development towards data-driven medicine. We have been using machine learning for image segmentation, tissue and cell classification, and other tasks, and we are extending our expertise towards deep learning technologies.

 

 

 

Radiomics

Radiomics discovers relationships between image features and clinical data. We apply techniques from statistics and machine learning to predict patient outcomes such as survival, therapy response, or side effects. The images are either characterized with engineered features or are directly fed into a neural network.

Automatic Segmentation

Fully automatic extraction of structures from images makes it possible to prepare results and measurements as soon as images arrive, so that the results are already available when human operators open a case. Our methods for robust, automatic segmentation can also be applied to large databases of images in order to derive population statistics.

Object-based Image Analysis

In many applications, such as multi-object segmentation, we employ a representation based on objects instead of pixels.The features of these objects and their relations to each other allow us to classify visible structures in a way that is similar to human vision.

Interactive Segmentation

In medical applications, one has to cope with extremely high variability, and pathologies may even create one-of-a-kind images. Therefore, it is unreasonable to expect automatic segmentation to work perfectly in all situations. Hence, we develop interactive tools, with a strong focus on usability and efficiency, minimizing the amount of work and time spent for the users.

 

Visualization and Interaction

We employ an extensive toolbox for 2D/3D visualization and interaction, including outstanding volume rendering capabilities using efficient, intuitive user interfaces.

Efficient Image Processing in MeVisLab

Our core technology allows us to implement processing pipelines that work with images larger than the available memory. Large volumes are divided into smaller regions for local processing, and the results are combined. This enables automated extraction of quantitative image features from large datasets.

Statistical Shape Models

Statistical Shape Models (SSM) are used to capture the typical shapes of organs, but also typical shape variations. We use SSM in order to guide automatic segmentation approaches when facing difficult contrast situations.

Registration Know-How

Our extensive competences in the area of image registration are put to use in many advanced image analysis tasks. For instance, registration enables the accurate analysis of dynamic contrast enhancement or the use of atlases for guiding segmentation procedures.

Clinical Applicability

We maintain outstanding knowledge of medical applications, their clinical context and workflows, and the medical state of the art by closely cooperating with a large network of clinical partners. This allows us to develop robust and applicable solutions that meet the actual needs of doctors, and to work with clinical routine images and their variability.