Sujitkumar Gavali: ML-based Detection and Quantification of MR Artifacts Utilizing Physics-based Data Simulators

BCCN Berlin / Technische Universität Berlin

Abstract

An artifact in a Magnetic Resonance Imaging (MRI) image degrades its diagnostic value or confuses with the pathology thus requiring a new patient scan. Troubleshooting
and rectifying an MR artifact is a time consuming process and requires technical expertise, which makes it a more complex problem to solve. There are studies on Machine
learning (ML) based detection and correction of a particular type of an MR artifact. However, detection and quantification of the MR artifact on a broader level is critical to solve the image quality (IQ) problems at the vendor manufacturing unit and at hospitals. This thesis explores the detection and quantification of MR artifacts on a broader scale for the real datasets generated using MRI Scanners and for the simulated datasets generated utilizing a Physics-based data simulator. The thesis also outlines a proof of concept for the root cause analysis of the MR artifacts. The approach involves using technical MR system parameters alongside MR DICOM images as multi-modal input for the ML algorithm, which can help identify the root causes of MR artifacts.

 

Additional information:

Master thesis defense

 

Organized by:

Prof. Dr. Matthias Böhm & Prof. Dr. Fatma Deniz