Automatic Artifact Detection Algorithm in Fetal MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Fetal Magnetic Resonance Imaging (MRI) is susceptible to artifacts like motion, chemical shift, and radiofrequency interference.
• Current artifact detection relies on subjective, time-consuming, and error-prone manual assessment by MRI operators.
• Objective and automated quality assurance methods are needed for reliable fetal MRI interpretation.

Purpose of the Study:

• To introduce RISE-Net, a novel algorithm for consistent, automatic, and objective detection and severity grading of artifacts in 3D fetal MRI.
• To evaluate the performance of RISE-Net against existing state-of-the-art deep learning architectures.

Main Methods:

• Development of RISE-Net, a Convolutional Neural Network (CNN) ensemble utilizing stacked Residual, Inception, Squeeze and Excitation (RISE) blocks.
• A two-stage CNN approach: one for artifact identification and classification, and a second for severity regression.
• Performance comparison with VGG-16, Inception, ResNet-50, ReNet-Inception, SE-ResNet, and SE-Inception architectures.

Main Results:

• The RISE-Net classification network achieved 90.34% accuracy and a 90.39% F1 score, outperforming other tested architectures.
• The severity regression network demonstrated a Mean Squared Error (MSE) of 0.083 across all artifact classes.
• RISE-Net provides a robust method for artifact detection and severity assessment in fetal MRI.

Conclusions:

• RISE-Net offers a significant advancement in fetal MRI quality assurance through automated artifact detection and grading.
• The algorithm's high accuracy and efficiency suggest its potential for rapid implementation in clinical settings.
• This automated approach can enhance the reliability and consistency of fetal MRI interpretation.