The definition of clinical target volume (CTV) margins around gross tumor volume (GTV) for radiotherapy of esophageal cancer (EC) and many solid tumors is still a challenge hence the currently available in-vivo imaging techniques still fail to detect areas of microscopic tumor extension (MTE). Many parameters of the tumor microenvironment (TME), e.g., tumor cell proliferation, cancer stem cells, hypoxia, kinases, immune architecture and patient-specific parameters are hypothesized as inducers of MTE in esophageal cancer and other tumors. The correlation of these TME biomarkers with MTE before, during or after radiochemotherapy (RCHT) is crucial in the era of image-guided, adaptive high-precision photon or particle therapy. In this thesis, two study cohorts were used to assess some selected TME biomarkers and their predictive value on MTE for an improved CTV definition. The first study used immunohistochemistry analysis for the assessment of TME marker namely HIF-1α, Ki67, p53, CXCR4 and PD1 in a cohort of retrospectively collected formalin-fixed paraffin-embedded (FFPE) blocks of EC patients treated with either neoadjuvant radiochemotherapy plus resection (NRCHT+R) or resection alone (R). The subsequent study employing a multiplex-immunofluorescence technique assessed the expression of various markers, i.e., FAK, ILK, CD44, HIF-1α and Ki67, in a cohort of prospectively prepared FFPE resection specimens of EC patients with implantable fiducial gold markers at the proximal and distal tumor borders illustrating the GTV prior to NRCHT+R and correlated those markers to the MTE. The findings from our first study showed upregulation of HIF-1α, Ki67, p53, CXCR4 and PD1, within squamous cell carcinoma (SCC) and adenocarcinoma (AC) patients treated with R compared to those having undergone NRCHT+R. In the second study higher expression of FAK+, CD44+, HIF-1+, and Ki67+ cells in tumor-nests than in tumor-stroma of both SCC and AC patients was found, although ILK+ cells were higher in tumor stroma. In addition, MTE reaching up to 31 mm beyond the fiducial markers was found in three patients (all cT3N1) with a stronger expression of FAK+, CD44+ and ILK+ cells in tumor-nests in between the fiducial markers (former GTV) and beyond those (former CTV), even after NRCHT. In conclusion, there is thus far no evidence that the TME influences the CTV margin on an individual patient basis, hence differences in the TME between patients with residual tumor cells in the original CTV compared to those without were not detected.

Colorectal cancer is one of the most prevalent malignancies globally and recently introduced digital pathology enables the use of machine learning as an aid for fast diagnostics. This project aimed to develop a deep neural network model to specifically identify and differentiate dysplasia in the epithelium of colorectal polyps and was posed as a binary classification problem. The available dataset consisted of 80 whole slide images of different H&E-stained polyp sections, which were parted info smaller patches, annotated by a pathologist. The best performing model was a pre-trained ResNet-18 utilising a weighted sampler, weight decay and augmentation during fine tuning. Reaching an area under precision-recall curve of 0.9989 and 97.41% accuracy on previously unseen data, the model’s performance was determined to underperform compared to the task’s intra-observer variability and be in alignment with the inter-observer variability. Final model made publicly available at https://github.com/stinabr/classification-of-colorectal-polyps.