Pixelwise Gradient Model with GAN for Virtual Contrast Enhancement in MRI Imaging

Area of Science:

• Medical Imaging
• Computational Biology
• Artificial Intelligence

Background:

• Advanced computational models are vital for enhancing diagnostic accuracy in medical imaging.
• Virtual contrast enhancement (VCE) in MRI aims to simulate contrast agent effects, reducing risks.
• Nasopharyngeal cancer (NPC) diagnosis can benefit from improved MRI techniques.

Purpose of the Study:

• To introduce and evaluate a novel VCE model for MRI, specifically PGMGVCE, for NPC detection.
• To simulate the effects of gadolinium-based contrast agents using AI, thereby minimizing patient risks.
• To optimize the PGMGVCE model through various modifications and normalization techniques.

Main Methods:

• The Pixelwise Gradient Model with GAN for Virtual Contrast Enhancement (PGMGVCE) was developed.
• PGMGVCE utilizes pixelwise gradient methods and Generative Adversarial Networks (GANs) to enhance T1-weighted and T2-weighted MRI images.
• Model performance was optimized by testing hyperparameters, normalization methods (z-score, Sigmoid, Tanh), and training strategies.

Main Results:

• PGMGVCE achieved comparable accuracy to existing models in terms of Mean Absolute Error (MAE), Mean Square Error (MSE), and Structural Similarity Index (SSIM).
• The model demonstrated significant improvements in texture representation, evidenced by enhanced Total Mean Square Variation per Mean Intensity (TMSVPMI), Total Absolute Variation per Mean Intensity (TAVPMI), Tenengrad function per Mean Intensity (TFPMI), and Variance function per Mean Intensity (VFPMI).

Conclusions:

• PGMGVCE offers an innovative and safe approach to virtual contrast enhancement in MRI.
• The study highlights the potential of deep learning models like PGMGVCE in advancing medical imaging.
• This AI-driven VCE method paves the way for more accurate and risk-free diagnostic tools in healthcare.