Deep Learning-Based Segmentation of Perfusion MRI

Jointly supervised with Gabriel Hoffmann and Christine Preibisch (TUM Universitätsklinikum).

Abstract:

This project aims to develop and validate deep learning models for the automatic segmentation of perfusion territories in super-selective arterial spin labeling (ss-ASL) data and for the segmentation of individual watershed areas from contrast-agent-based time-to-peak (TTP) maps. The project will leverage the nnUNET framework, which is known for its robustness and flexibility in medical image segmentation tasks. Currently, expert segmentations are highly time-consuming and labor-intensive. Automating these segmentations with a reliable deep learning model will significantly enhance research efficiency and accuracy.