Deep learning-based automated scan plane positioning for brain magnetic resonance imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Manual scan planning in clinical MRI is inaccurate, inconsistent, and time-consuming.
• Existing automated methods lack accuracy, stability, and computational efficiency for practical use.

Purpose of the Study:

• Develop and evaluate a reliable, accurate deep learning framework for automatic head MRI scan plane positioning.
• Incorporate prior physical knowledge into the AI model for improved performance.

Main Methods:

• An end-to-end deep learning framework using a cascaded 3D convolutional neural network for landmark detection.
• Multi-scale feature fusion and physically meaningful regression losses (PRL, DRL).
• Data augmentation strategies simulating complex clinical scenarios.

Main Results:

• Achieved high performance on 229 clinical head MRI scans.
• Demonstrated low point-to-point absolute error (0.872 mm) and relative error (0.10%).
• Reported average angular errors of 0.502°, 0.381°, and 0.675° for sagittal, transverse, and coronal planes.

Conclusions:

• The proposed deep learning approach offers high efficiency, accuracy, and robustness.
• Effective for diverse clinical head MRI scans, including variations in positioning, contrast, noise, and pathologies.