Computational Radiology Research Group

Computational Radiology Research Group (Kleesiek)

© dkfz.de

The research group Computational Radiology develops and validates novel algorithms for the analysis of radiological data. Using machine learning techniques, we identify high-dimensional digital "fingerprints" of pathological and physiological processes.

Research topics include detection and characterization of physiological structures and (oncological) lesions, leading to quantitative imaging markers that support diagnostics and clinical decision making. Next to supervised learning paradigms we also explore unsupervised methods for extracting linear and nonlinear relationships which otherwise would not be accessible to humans. Ongoing projects explore how results from the analysis of imaging data can be integrated with information extracted from radiological and clinical reports using natural language processing (NLP).

Next to algorithms for conventional CT and MRI data we investigate techniques for extracting information from novel MRI sequences. Using special MRI pulse sequences, quantitative information about T1 and T2 times as well as proton density (PD) can be acquired. Based on these measurements various multiparametric contrasts can be calculated. This approach is known as synthetic MRI and has the decisive advantage that the obtained multiparametric imaging data is perfectly aligned. Further, additional information on the temporal course of the MR signal for each spatial location can be utilized to better characterize healthy and non-healthy tissue types. In addition, we also analyze data from metabolic MR imaging, like Sodium and CEST imaging, to discover novel relationship useful for clinical applications.

As part of the digital transformation and introduction of artificial intelligence in radiology, we aim at translating our methods into clinical routine. The described projects are conducted as part of many regional and transregional collaborations.

Selected Publications

Kleesiek, J., Morshuis, J., Isensee, F., Deike-Hofmann, K., Paech, D., Kickingereder, P., Köthe, U., Rother, C., Forsting, M., Wick, W., Bendszus, M., Schlemmer, H.-P., and Radbruch, A. (accepted). Can Virtual Contrast Enhancement in Brain MRI Replace Gadolinium? A Feasibility Study. Investigative Radiology.

Kleesiek, J., Urban, G., Hubert, A., Schwarz, D., Maier-Hein, K., Bendszus, M., and Biller, A. (2016). Deep MRI brain extraction: A 3D convolutional neural network for skull stripping. NeuroImage, 129:460–469.

Kleesiek, J., Petersen, J., Döring, M., Maier-Hein, K., Köthe, U., Wick, W., Hamprecht, F. A., Bendszus, M., and Biller,A.(2016). Virtual Raters for Reproducible and Objective Assessments in Radiology. Nature Scientific Reports, 6:25007.

Biller, A., Badde, S., Nagel, A., Neumann, J.-O., Wick, W., Hertenstein, A., Bendszus, M., Sahm, F., Benkhedah, N., and Kleesiek, J.(2016). Improved Brain Tumor Classification by Sodium MR Imaging: Prediction of IDH Mutation Status and Tumor Progression. American Journal of Neuroradiology, 37(1):66–73.

Kleesiek, J., Biller, A., Bartsch, A.J., and Ueltzhöffer, K. (2015). Crutchfield Information Metric: A Valid Tool for Quality Control of Multiparametric MRI Data? In Biomedical Engineering Systems and Technologies, 113–125. Springer.

to top