Evaluation of Super Resolution Models for Generating Synthetic 3T T1W Images from 1.5T T1W Scans

3T MRI provides higher-resolution brain images than 1.5T MRI, but is more expensive and has limitations. We explored deep learning-based super-resolution methods to enhance 1.5T MRI images to resemble 3T quality. We evaluated these synthetic “upgraded” images using various metrics to compare them with actual 3T images. By employing measures such as Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR) and Learned Perceptual Image Patch Similarity (LPIPS) and Intensity Differences at Pixels (IDP) we were able to quantify the fidelity and perceptual quality of the generated super-resolution 3T images. Our findings highlighted the superior performance of the Super Resolution Generative Adversarial Network (SRGAN), which significantly reduced pixel differences, improved image clarity, and better preserved critical anatomical details compared to other methods. These results can be beneficial for medical imaging, offering a cost-effective and accessible solution to achieve high-quality images without the need for high-field MRI scans.