New aspects and tools in hepato-gastrointestinal pathology & tumour microenvironment: molecular and computational pathology approaches

Hepatocellular carcinoma (HCC) is the most common type of primary liver cancer and is the fourth leading cause of cancer-related deaths. HCC is a highly heterogeneous and complex disease with limited treatment success, especially for patients with advanced disease. A better understanding of the tumour microenvironment (TME) of HCC could improve patient stratification strategies and help develop novel treatments. Besides using conventional microscopical analysis, implementing recently emerging computational image analysis techniques can exponentially amplify the data one can gather from the slides. This PhD work aims to explore the TME, investigate novel paths of carcinogenesis, and develop new computational tools for disease diagnosis and in-depth analysis of liver and gastrointestinal pathologies. Techniques from a spectrum of conventional, molecular, and computational pathology are used to achieve these aims. The studies are presented in four separate manuscripts.
In Manuscript I, the spatial distribution and density of tumour infiltrating inflammation in HCC was investigated at the pathologist level and classified into inflamed, immune-excluded, and immune desert categories. The classification is associated with patient outcomes. A deep learning (DL) based image analysis pipeline was then developed to automatically analyse the TME from immunohistochemistry slides to provide accurate and detailed immune cell quantification. In Manuscript II, a DL-based pathology tool [AI(H)] was developed for pathologist-level evaluation of liver needle biopsy samples from autoimmune hepatitis patients. The model can provide accurate, quantifiable, granular results regarding standard hepatitis grading, staging assessment, immune cell classification, quantification, and localisation. In Manuscript III, a possible novel path of hepatocarcinogenesis from a benign lesion, focal nodular hyperplasia (FNH), was explored. A case with a concomitant presence of FNH with hepatocellular carcinoma (HCC) neighbouring each other was analysed, and their phylogenetic relationship was shown. Finally, in Manuscript IV, another essential component of TME, stroma, was studied in colorectal carcinoma (CRC). The tumour-promoting role of FAP overexpression as a cancer-associated fibroblast marker in colorectal carcinoma was investigated on protein and RNA levels. It was concluded that cancer-associated stroma with FAP expression is significantly associated with worse prognosis, high tumour stage and immune suppressive type immune infiltration.
In this thesis, the characteristics of the HCC TME, DL-based image analysis in computational pathology of liver and gastrointestinal pathologies and a novel path of hepatocarcinogenesis were investigated extensively. The work described provides a powerful AI-driven tool for the robust evaluation of the TME to enable an extensive analysis of pathology slides and to augment pathologist practice during a daily diagnostic routine.