Inter-Slice Resolution Improvement Using Convolutional Neural Network with Orbital Bone Edge-Aware in Facial CT...

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• 3D orbital bone modeling from facial CT is crucial for surgical reconstruction.
• Current CT images suffer from aliasing and disconnected thin bones due to slice thickness.
• Improved inter-slice resolution is needed for accurate 3D orbital bone models.

Purpose of the Study:

• To propose a 2D CNN-based method for enhancing inter-slice resolution in facial CT images.
• To improve the accuracy and clarity of 3D orbital bone models for surgical applications.
• To address aliasing effects and disconnected thin bone structures in CT scans.

Main Methods:

• A 2D convolutional neural network (CNN) utilizing sagittal and axial plane spatial information.
• Generation of intermediate slices on the sagittal plane, transformed to axial images.
• Employing orbital bone edge-aware (OBE) loss and feature map difference loss for accuracy and perceptual quality.

Main Results:

• The proposed 2D CNN method outperformed traditional interpolation techniques and 3D SRGAN.
• Generated intermediate slices exhibited clear edges for thin and cortical bones.
• Demonstrated superior performance in both sagittal and axial planes compared to existing methods.

Conclusions:

• The developed 2D CNN method effectively enhances inter-slice resolution in facial CT images.
• The technique significantly improves the visualization of orbital bone structures, aiding surgical planning.
• This approach offers a promising solution for accurate 3D orbital bone reconstruction.