Multi-Modality MR Image Synthesis via Confidence-Guided Aggregation and Cross-Modality Refinement.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Multi-modality Magnetic Resonance Imaging (MRI) is crucial for disease diagnosis and treatment planning.
• Simultaneous acquisition of multi-modality MRI is challenging due to patient discomfort and cost.
• Synthesizing missing MRI modalities from existing ones is a significant research area.

Purpose of the Study:

• To develop an effective method for synthesizing high-quality target-modality MR images from available source modalities.
• To leverage complementary and correlative information across different MRI modalities.

Main Methods:

• Proposed a novel confidence-guided aggregation and cross-modality refinement network (CACR-Net).
• Implemented a confidence-guided aggregation module to adaptively combine target-modality images based on confidence maps.
• Utilized a cross-modality refinement module to enhance the synthesized image by exploiting correlations between source and aggregated images.

Main Results:

• The CACR-Net effectively synthesizes high-quality and sharp target-modality MR images.
• Experimental results on a benchmark dataset demonstrate superior performance compared to state-of-the-art methods.
• The method successfully utilizes complementary and correlative information from multiple MRI modalities.

Conclusions:

• The proposed CACR-Net offers an effective solution for multi-modality MR image synthesis.
• This approach can potentially reduce scanning costs and patient discomfort while improving diagnostic accuracy.
• The network demonstrates significant advancements in leveraging cross-modality information for image synthesis.