The increasing availability of multimodal PET/MRI systems leads to the generation of large amounts of imaging data in the laboratory as well as in clinical routine. Multiple functional MRI parameters can be acquired along with the dynamic and static uptake of several PET-tracers. The analysis of such datasets can be overwhelming for researchers and clinicians, and machine learning methods are used to increase the amount of useful information that can be extracted. We aim to find patterns which would allow us to differentiate between different tumor classes, such as benign and malignant, or to differentiate between disease outcomes.
Phenotypic variations commonly exist between different regions within a tumor, and it is essential to elucidate the underlying biological factors. The differences can influence the effectiveness of therapy, or be a predictor for disease progression. Therefore, we are not only interested in the tumor as a whole, but particularly in tumor heterogeneity. These variations also present themselves in imaging parameters measured with PET and MRI.
The different datasets acquired in the clinic, as well as in the Werner Siemens Imaging Center include multi-parametric imaging, histology, and data from various omics techniques. The data are processed and analyzed with a variety of computer algorithms and stored for later access. We are building the complete workflow for processing, storage, analysis, and access, as depicted in Figure 1.
For each type of data, various analysis methods can be used. In figure 2A, different MRI parameters and dynamic FDG-PET images were used to segment a tumor in regions with different biological properties using advanced image analysis and machine learning techniques.
In figure 2B, specific tissue samples were obtained based on image information, in this case differences in FDG uptake. These samples can subsequently be measured with NMR to obtain a metabolic profile of the tissue. Machine learning techniques are subsequently used to analyze such measurements, distinguish different groups of samples and indicate important metabolites.
Figure 2C shows an example of tumor slices with different histological stains. Each stain reveals different information about the biological properties of the tissue. Image analysis methods were used to reveal proliferation (top) and vessel density (bottom) variations within the tumor.
Group Members
PhD student
PostDoc