Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

ST: Fachverband Strahlen- und Medizinphysik

ST 1: Computational Methods and Simulation

ST 1.5: Vortrag

Dienstag, 1. April 2025, 14:45–15:00, ZHG003

Autoencoder-based Anomaly Detection in MRI Raw Data — •Jessica Mnischek, Jens Weingarten, Kevin Kröninger, and Marco Schlimbach — TU Dortmund

Tumor distinction in medical imaging remains a challenging task due to the subtle differences between various tumor types and abnormal tissues. Magnetic Resonance Imaging (MRI) is a powerful diagnostic tool widely used in the detection and monitoring of tumors, providing detailed visualization of soft tissues.
To enhance diagnostic accuracy, machine learning methods have been increasingly applied to MRI data. Among these methods, autoencoders have demonstrated potential in detecting subtle differences between healthy and diseased scans. By training on healthy datasets, they learn compact, efficient representations of typical tissue patterns. When applied to diseased datasets, the reconstruction error can highlight anomalies, thereby facilitating the detection of irregularities that may indicate the presence of disease.
This study investigates the application of autoencoders to raw MRI data, which preserves the complete acquired information, including both magnitude and phase components. In contrast, standard MRI analyses mostly rely only on magnitude information. By working directly with raw MRI data, this approach explores the potential of utilizing phase information to more effectively differentiate between healthy and tumor-affected tissues.

Keywords: Machine Learning; Magnetic Resonance Imaging; Tumor Diagnosis; K-Space