Title:	Multimodal breast cancer image data registration and analysis
DNr:	Berzelius-2026-161
Project Type:	LiU Berzelius
Principal Investigator:	Ida-Maria Sintorn <ida.sintorn@it.uu.se>
Affiliation:	Uppsala universitet
Duration:	2026-05-30 – 2026-12-01
Classification:	10207
Keywords:

Abstract

The overall purpose of the project is to create a framework for registration and combined analysis of 4 different layers of in situ image informatics (high resolution multimodal image data of consecutive sections of a sample containing histology, transcriptomics, proteomics and metabolomics information), with the biological goal to map out the breast cancer tumor micro-environment, understand the regulatory signaling network, and identify early stage progression markers and treatment targets. We will develop and train ML models to a) learn missing layers in multimodal imaging. The 4 types of data are from consecutive tissue slices and are typically distorted at sample preparation. We will develop a model that learns the appearance of one layer from the others in order to use the created layer as template to b) do subsequent patch-wise registration of our multimodal data. For our types of data the distortions are typically a mix of continuous and discontinuous, and thus other DL based registration approaches working on full images (JEPA, subspace registration, etc.) used in other applications fail.