RegQCNET: Deep quality control for image-to-template brain MRI affine registration

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence

Background:

• Affine registration is crucial for brain image analysis tasks like segmentation and functional analysis.
• Manual quality control of registration is time-consuming and impractical for large datasets.
• Automated quality control (QC) is essential for efficient processing of massive neuroimaging data.

Purpose of the Study:

• To introduce RegQCNET, a compact 3D convolutional neural network for automated quality control of affine brain image registration.
• To quantitatively predict affine registration error in metric units for defining usable/non-usable images.
• To evaluate the robustness and accuracy of RegQCNET compared to traditional methods.

Main Methods:

• Development of RegQCNET, a 3D convolutional neural network for predicting registration error.
• Testing RegQCNET on lifespan brain images with simulated transformations and intensity variations.
• Evaluating image classification (usable/non-usable) using manual and automatic thresholds derived from computer-assisted models.
• Comparison of RegQCNET accuracy against image correlation coefficient and mutual information.

Main Results:

• RegQCNET accurately estimates affine registration error in metric units.
• The deep learning QC approach is robust to simulated spatial transformations and intensity variations.
• RegQCNET demonstrates strong performance in classifying images as usable or non-usable.
• The proposed method is significantly faster than manual quality control.

Conclusions:

• RegQCNET provides a robust, fast, and accurate automated quality control solution for affine brain image registration.
• This deep learning approach is suitable for integration into large-scale neuroimaging processing pipelines.
• Automated QC using RegQCNET can significantly improve the efficiency and reliability of brain image analysis.