Automated Detection of Motion Artefacts in MR Imaging Using Decision Forests

Area of Science:

• Medical Imaging
• Machine Learning
• Image Reconstruction

Background:

• Magnetic Resonance (MR) scan acquisition times often exceed patient stillness, leading to motion artifacts.
• Patient motion, including bulk and respiratory motion, significantly degrades MR image quality and diagnostic value through artifacts like ghosting, blurring, and smearing.

Purpose of the Study:

• To investigate the impact of motion on reconstructed MR slices.
• To develop and evaluate a supervised learning method for detecting motion artifacts in MR image reconstruction.

Main Methods:

• A supervised learning approach using random decision forests was employed.
• The study analyzed bulk patient motion in head scans and respiratory motion in cardiac scans.
• Synthetic MR images with introduced motion artifacts were evaluated across Cartesian, radial, and spiral k-space sampling patterns.

Main Results:

• The machine learning model demonstrated capability in learning motion artifact characteristics.
• Detection confidence of motion artifacts varied depending on the k-space sampling pattern used.

Conclusions:

• Supervised machine learning is effective for detecting motion artifacts in MR imaging.
• Understanding the influence of k-space sampling patterns is crucial for accurate artifact detection.