Enhancing Medical Images Quality Using Vision Transformer Framework

Abstract High resolution images are crucial for precise diagnosis and efficient treatment planning in medical imaging. However, prevalence of low-resolution images remains a significant challenge, often limiting the detail and clarity necessary for reliable clinical evaluations. To address this issue, we applied the Vision Transformer Auto Encoder (ViTAE), a specialized Convolutional Neural Network CNN model designed for image enhancement. The study’s dataset, which included a range of medical imaging scenarios, was gathered locally from a computed tomography (CT) scan lab. Over a series of training epochs, the Vision Transformer Auto Encoder (ViTAE) exhibited consistent improvements in peak signal to noise ratio (PSNR), ultimately achieving PSNR of 43.06 decibels dB and Structural Similarity Index Measure SSIM of 0.983. Our proposed model ViTAE also outperforms the other Information eXtraction from Images IXI dataset having a PSNR of 43.72 decibels (dB) and SSIM 0.984 respectively. By optimizing its convolutional layers to extract and refine features from the input images, model progressively enhanced its ability to reconstruct and clarify images. These results underscore potential of ViTAE to significantly improve quality of medical images, offering a promising solution to overcome limitations of low resolution medical imaging.