We obtain PET data and combine them with histopathological examinations, analyses of genetic mutations relevant to therapy and prognosis, as well as clinical data on survival and response to therapy. The result should be a non-invasive "in vivo pathology" that leads to an individualized therapy algorithm and can continuously monitor its success. To achieve this complex goal, a supervised machine learning (ML) algorithm will be applied. This novel method is designed to identify specific PET patterns caused by mutations in the tumor and to predict the response to therapy.
The results will be validated using preclinical mouse models. Using CRISPR/Cas9, genetic mutations are induced in xenografts, in this case mice to which human tumor cells have been transferred, and then it is investigated whether the PET patterns found in patients can be reproduced in the mouse model. In addition, the imaging approach will be prospectively validated with liquid biopsies, i.e. tests for tumour components and tumour markers in the blood. Blood samples will be taken shortly before PET imaging and clinical data such as survival and therapy response will be collected to correlate them with the PET patterns. The overall goal is to develop an integrative diagnostic algorithm to non-invasively determine and monitor tumor biology and thus the best possible initial therapy.
In order to develop the described pipeline, three tumor entities were selected, which pose different challenges to the method to be investigated: colorectal carcinomas and diffuse large B-cell lymphomas - both investigated using F18-2-fluorodeoxyglucose (FDG)-PET - as well as prostate carcinomas - investigated using prostate specific membrane antigen (PSMA)-PET. In the further course of the study, it is planned to extend the study to breast carcinomas and non-small cell lung carcinomas.