Deep Learning-Based Enhancement of Low-Field MRI Using CNN-GAN for High-Quality Image Reconstruction

Abstract

Magnetic Resonance Imaging (MRI) is one of the most common non-invasive imaging modalities with the capacity to visualize internal human anatomical structures such as soft tissues and organs with high resolution. High-field MRI (i.e., 1.5 T and above) provides better image quality, but its high cost and expensive infrastructure limit its availability, especially in low-resource settings. Low-field MRI systems are a more cost-effective solution; however, they are hindered by a relatively poor Signal-to-Noise Ratio (SNR), spatial resolution, and increased image artifacts that can undermine diagnostic accuracy. Existing techniques for MRI image enhancement have evolved through the use of classical and deep learning approaches, mainly based on Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs) of different types. These techniques suffer from major downsides, including a loss of fine anatomical details, poor generalization to datasets from different imaging modalities, and an inability to adequately preserve modality-specific characteristics. To address these limitations, we propose a novel hybrid deep learning-based image enhancement framework that implements both CNN and GAN architectures to generate images with quality comparable to that of high-field (1.5 T) MRI from low-field MRIs. Our method uses a multi-stage pipeline, including preprocessing, feature extraction, adversarial learning, and reconstruction, to reduce noise and enhance structural detail without introducing anatomically implausible structures. Experimental results show that the proposed model can achieve a Peak Signal-to-Noise Ratio (PSNR) of 32.5 dB, a Structural Similarity Index (SSIM) of 0.92, and a Root Mean Square Error (RMSE) of 4.1 on low-field MRI images, representing significant improvements over the baseline results. These findings demonstrate that the proposed methodology improves image fidelity and offers greater similarity to high-field MRI, supporting its use for improved diagnostic applications in low-resource settings.