Pan-cancer tumor microenvironment exploration with vision transformers

Projekt: Stiftelser och fonder


Beskrivning (abstrakt)

Tumor microenvironment holds spatial information relevant for understanding cancer biology and development of targeted therapies. Fluorescent microscopy enables capturing this information from cancer tissue sections in high-throughput manner using organelle or protein specific markers. Recent developments in machine learning opens new possibilities in data-driven discoveries from this large-scale imaging data. In this project, we aim to study and develop novel deep learning methods based on vision transformers that are able to capture spatial associations between fluorescent markers, cells and tissue compartments. These methods will be used to discover novel associations from large-scale pan-cancer microscopy image dataset including 17 different solid tumor types and over 30 000 tissue microarray samples. Data normalization and self-supervised learning approaches will be studied to enable utilization of the whole pan-cancer dataset for representation learning. These models will be fine-tuned for each cancer type separately to discover spatial associations relevant for cancer prognosis. Finally, findings from all cancer types will be collected together for pan-cancer analysis of tumor microenvironment similarities and differences. The project advances knowledge in computer science and cancer biology. Developed novel deep learning methods and vision transformer models will be shared openly for cancer research community. Project discoveries are expected to create new knowledge of cancer biology that can be further developed as biomarkers for prognostic diagnostics or targeted therapies. The project will be carried out in the Institute for Molecular Medicine Finland (FIMM), HiLIFE, University of Helsinki.
Kort titelPan-cancer tumor microenvironment explor
Gällande start-/slutdatum01/01/202331/12/2024


  • Syöpäsäätiö sr Cancerstiftelsen sr: 50 000,00 €