Developing quantitative image analysis pipelines for scoring histological panoramic images : Testing Rab24 as a possible biomarker for cancer

Abstract

Biomarkers are highly essential to improve diagnosis, confirm the diseases' development, and monitor the treatment. Biomarker discovery requires analysis of a large quantity of data which is aided by computational tools. One of the methods widely used in the search of new biomarkers is immunohistochemistry of tissue samples.

Numerous tools are available to detect different cell types in tissues in histological sections; still, the need for more advanced and quantitative analysis is growing. The most successful paradigms to meet these novel needs are using deep learning-based networks. Rab24, an atypical member of the Rab protein family, plays a role in the late steps of endosomal degradation, in mitochondrial plasticity, and in the clearance of autolysosomes in basal autophagy. Rab24 has been connected to neurodegeneration and cancer. It has been shown to be overexpressed in hepatocellular carcinoma (HCC) and to enhance HCC's malignant phenotype. These findings together indicate that Rab24 might be a potential biomarker for cancer, and its modulation might be used as a strategy for cancer therapy.

This project was undertaken to investigate the expression of Rab24 in different types of human cancers. Rab24 was detected by immunohistochemical staining in cancer tissue samples embedded in paraffin. For the evaluation of expression levels, detailed image analysis pipelines were developed to combine an open-source software called QuPath with a deep learning network, StarDist, in order to setup a robust quantitative cell detection compatible with histological panoramic images. Based on our current analysis, 5 cancer types, including angiosarcoma, stomach gastrointestinal stromal tumor (GIST), rectal neuroendocrine carcinoma (NEC), liposarcoma and fibrosarcoma were selected as potential candidates for further investigation.