Deep embedding convolutional neural network for synthesizing CT image from T1-Weighted MR image.

Area of Science:

• Medical imaging
• Artificial intelligence
• Image processing

Background:

• Medical image synthesis across modalities is gaining attention.
• Synthesizing computed tomography (CT) from T1-weighted magnetic resonance (MR) images presents significant challenges due to distinct appearances.
• Accurate CT synthesis is crucial for various clinical applications.

Purpose of the Study:

• To develop a novel deep learning method for MR-to-CT image synthesis.
• To address the complexity and appearance gaps between MR and CT modalities.
• To improve the quality and efficiency of synthesized CT images.

Main Methods:

• A novel deep embedding convolutional neural network (DECNN) was proposed.
• Feature maps from MR images are generated and processed through convolutional layers.
• A tentative CT synthesis is embedded back into feature maps iteratively to refine them.
• The method was validated on brain and prostate imaging datasets.

Main Results:

• The DECNN method demonstrated superior performance compared to state-of-the-art techniques.
• Synthesized CT images exhibited enhanced perceptive quality.
• The DECNN achieved efficient run-time costs for CT image synthesis.

Conclusions:

• The proposed DECNN with repeated embedding operations is effective for MR-to-CT synthesis.
• The method offers a promising solution for generating high-fidelity CT images from MR data.
• DECNN provides a balance between perceptual quality and computational efficiency.