Multi-sequence generative adversarial network: better generation for enhanced magnetic resonance imaging images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Magnetic Resonance Imaging (MRI) is crucial for diagnosing brain diseases.
• Contrast-enhanced MRI sequences require contrast agents, which pose risks and are contraindicated for certain populations (e.g., pregnant women, children).
• Adverse reactions and high costs associated with contrast agents necessitate alternative imaging methods.

Purpose of the Study:

• To develop a deep learning model capable of generating contrast-enhanced MRI sequences from non-enhanced ones.
• To evaluate the performance of the proposed model against existing methods like pix2pix.
• To explore the potential of reducing contrast agent usage in clinical MRI.

Main Methods:

• A generative adversarial network (GAN) model with multimodal inputs and end-to-end decoding, based on the pix2pix architecture, was proposed.
• The model's effectiveness was assessed using Normalized Mean Square Error (NMSE), Root Mean Square Error (RMSE), Structural Similarity Index Measure (SSIM), and Peak Signal-to-Noise Ratio (PSNR).
• Statistical analysis was performed to compare the proposed model with pix2pix.

Main Results:

• The proposed model significantly outperformed pix2pix, demonstrating higher SSIM and PSNR values, and lower NMSE and RMSE values.
• Inputting T1-weighted (T1W) and T2-weighted (T2W) images yielded superior results compared to other input combinations.
• The model successfully generated magnetic resonance enhancement sequence images from non-enhanced sequences.

Conclusions:

• The developed deep learning model can generate contrast-enhanced MRI sequences without contrast agents.
• This approach offers significant benefits by reducing contrast agent use, protecting vulnerable populations, and potentially lowering healthcare costs.
• The findings provide a foundation for future research in generating enhanced MRI sequences.