Final Report Summary - IMAGELINK (Multiparametric tumor imaging and beyond: Towards understanding in vivo signals)
The objective of ImageLink was to deepen our knowledge on tumor physiology and pathology using existing and novel tracers for positron emission tomography-imaging (PET) in combination with other imaging modalities. To achieve this, we developed novel PET tracers that enable us to image target structures not accessible by imaging so far and therefore gain valuable additional information about tumors through imaging. In addition, we combined existing imaging modalities in an optimized multiparametric imaging protocol in combination with novel machine learning methods. Finally, we developed a method to gather metabolomic and proteomic information by collecting spatially defined tissue samples, selected based on imaging data for correlation of imaging and omics data (i.e. metabolomics and proteomics).
During the course of this project we were able to develop a senescence-specific PET tracer. The compound was tested extensively in in vitro and in vivo experiments and the promising results encouraged us to perform an initial first-in-man study. Thus, this PET tracer could become the first tool for non-invasive in vivo detection of senescence in humans and non-genetically modified rodents and has high potential to have great impact on clinical care of cancer patients.
A novel multiparametric PET/MR imaging protocol was developed and different methodologies were used to analyze the imaging data. For data analysis with only two parameters, we applied population-based Gaussian mixture models. With this method we were able to monitor the development of viable and necrotic regions within a tumor in a longitudinal manner. For cases where dynamic PET imaging was applied, we developed novel unsupervised machine learning approaches as a complement to standard compartmental modeling. These methods were assessed visually and quantitatively and showed a good agreement between clustering results and histology. Supervised machine learning methods were developed specifically for situations were multiple parameters are acquired with PET and MRI simultaneously. Modified versions of these algorithms are currently applied in two clinical projects.
We have developed an image-guided milling machine (IGMM) to link imaging data with proteomic and metabolomic data. The IGMM enables us to collect tissue samples from a frozen mouse for further analysis. The tissue samples can be selected and excised based on imaging data with high spatial accuracy, while the samples remain completely frozen and therefore metabolically conserved. Therefore, we can combine all available imaging information with the corresponding data generated by metabolomic and proteomic analyses. This methodology will lead to a better understanding of the underlying molecular mechanisms leading to a specific imaging phenotype. Preliminary results show a correlation of FDG-PET uptake and lactate concentration within tissue samples isolated with the IGMM. This is not an unexpected correlation, but analyses with optimized methods are currently ongoing and we expect to find novel correlations between imaging and proteomic and metabolomic data.
Overall we could achieve all major aims of the proposal. The development of a senescence specific PET tracer and the machine learning algorithms developed for tumor segmentation have high potential for clinical application and can significantly improve diagnostics towards personalized medicine. With the IGMM we have developed a tool which will lead to a better understanding of imaging data. Therefore several outcomes of this project have the potential to have a significant impact on clinical care or basic research. Moreover, during this ERC project we acquired a huge amount of complementary and unique multiparametric data, which requires analysis work for approximately another two years.
During the course of this project we were able to develop a senescence-specific PET tracer. The compound was tested extensively in in vitro and in vivo experiments and the promising results encouraged us to perform an initial first-in-man study. Thus, this PET tracer could become the first tool for non-invasive in vivo detection of senescence in humans and non-genetically modified rodents and has high potential to have great impact on clinical care of cancer patients.
A novel multiparametric PET/MR imaging protocol was developed and different methodologies were used to analyze the imaging data. For data analysis with only two parameters, we applied population-based Gaussian mixture models. With this method we were able to monitor the development of viable and necrotic regions within a tumor in a longitudinal manner. For cases where dynamic PET imaging was applied, we developed novel unsupervised machine learning approaches as a complement to standard compartmental modeling. These methods were assessed visually and quantitatively and showed a good agreement between clustering results and histology. Supervised machine learning methods were developed specifically for situations were multiple parameters are acquired with PET and MRI simultaneously. Modified versions of these algorithms are currently applied in two clinical projects.
We have developed an image-guided milling machine (IGMM) to link imaging data with proteomic and metabolomic data. The IGMM enables us to collect tissue samples from a frozen mouse for further analysis. The tissue samples can be selected and excised based on imaging data with high spatial accuracy, while the samples remain completely frozen and therefore metabolically conserved. Therefore, we can combine all available imaging information with the corresponding data generated by metabolomic and proteomic analyses. This methodology will lead to a better understanding of the underlying molecular mechanisms leading to a specific imaging phenotype. Preliminary results show a correlation of FDG-PET uptake and lactate concentration within tissue samples isolated with the IGMM. This is not an unexpected correlation, but analyses with optimized methods are currently ongoing and we expect to find novel correlations between imaging and proteomic and metabolomic data.
Overall we could achieve all major aims of the proposal. The development of a senescence specific PET tracer and the machine learning algorithms developed for tumor segmentation have high potential for clinical application and can significantly improve diagnostics towards personalized medicine. With the IGMM we have developed a tool which will lead to a better understanding of imaging data. Therefore several outcomes of this project have the potential to have a significant impact on clinical care or basic research. Moreover, during this ERC project we acquired a huge amount of complementary and unique multiparametric data, which requires analysis work for approximately another two years.