Arterial Spin Labeling Images Synthesis From sMRI Using Unbalanced Deep Discriminant Learning

Area of Science:

• Medical Imaging
• Deep Learning
• Neuroscience

Background:

• Medical image acquisition faces challenges like high costs and patient issues.
• Deep learning enables medical image synthesis, but Arterial Spin Labeling (ASL) MRI synthesis remains underexplored.
• ASL images are crucial for diagnosing dementia diseases.

Purpose of the Study:

• To synthesize ASL images from structural MRI for the first time.
• To develop a novel deep learning model for ASL image synthesis.
• To evaluate the quality and utility of synthesized ASL images in dementia diagnosis.

Main Methods:

• A novel unbalanced deep discriminant learning model with ResNet sub-structures was developed.
• The model synthesized ASL images from structural MRI.
• Extensive experiments and statistical analyses were conducted.

Main Results:

• Synthesized ASL images closely resemble real ASL scans.
• The synthesized images performed well in partial volume effect corrections.
• The model demonstrated potential for improving dementia diagnosis using synthesized ASL images.

Conclusions:

• The study successfully synthesized ASL images from structural MRI using a novel deep learning approach.
• The synthesized ASL images are of high quality and suitable for further analysis.
• This technique shows promise for enhancing dementia diagnosis and overcoming medical imaging data limitations.