Future proof pathology for predictive medicine and disease prognosis based on tumor heterogeneity. (R-11405)

With the latest technological advances in biomedical digital pathology, it is possible to generate images of tissue sections in high-resolution. However, standard protocols in clinical pathology have not evolved accordingly. In this project, an interdisciplinary team of experts in the fields of cancer biology, pathology, clinical practitioners, imaging- and data-analysis, will develop novel methods to identify the tumor heterogeneity within lung tumors based on high-resolution images of tumor tissue biopsies. First, the fine-grained details in whole slide images will be investigated via convolutional neural networks, supervised machine learning and advanced task adaptation methods to recognize important markers in the image. Secondly, the spatial heterogeneity of cell types within lung tumor tissue will be assessed via spatial point process models, quantifying the spatial distribution of cells, and uncovering interactions between them. The objective is to study the spatial heterogeneity as a novel biomarker for tumor grading and long-term prognosis. The heterogeneity will first be studied in a preclinical mouse model that mimics the histological situation in human lung cancer. Thereafter, the bridge to clinical practice will be made by studying biopsies from lung cancer patients. This allows us to identify subpopulations of patients that are more or less likely to respond to given treatments and predict the outcome of therapeutic approaches. Identifying markers that are common in preclinical and clinical samples, will enable to generate predictive values for therapeutic responses already in the preclinical phase.

Keywords:Artificial intelligence, Oncology

Disciplines:Data mining, Bio-informatics, Histology, Cancer diagnosis, Biostatistics, Cellular interactions and extracellular matrix