Göttingen 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

ST: Fachverband Strahlen- und Medizinphysik

ST 1: Computational Methods and Simulation

ST 1.4: Talk

Tuesday, April 1, 2025, 14:30–14:45, ZHG003

Improving Brain Tumor Characterization Using Generative Neural Networks and Raw MRI Data — •Marco Schlimbach, Jens Kleesiek, Kevin Kröninger, Moritz Rempe, and Jens Weingarten — TU Dortmund University

Characterizing brain tumors from MRI scans remains a significant challenge in clinical practice. Determining the specific tumor type often requires invasive biopsies, which hold risks for patients. Research efforts aim to improve non-invasive tumor characterization by using machine learning techniques. However, despite significant advancements, the accuracy of these methods has not yet reached the level needed to reliably replace biopsies. Current state-of-the-art algorithms commonly rely on reconstructed MRI images optimized for human interpretation. These images are the result of complex reconstruction pipelines that discard the raw phase information.

This study investigates the diagnostic potential of raw MRI data, which retains phase information and provides a more comprehensive representation of scanned tissue. A novel workflow is introduced to generate synthetic raw MRI data, including both healthy scans and scans with lesions. By leveraging generative machine learning techniques alongside raw MRI data, this approach aims to reveal new features and insights that could enhance non-invasive tumor characterization.

Keywords: AI; MRI; Tumor Diagnosis; k-Space; Generative Neural Networks