<![CDATA[This study aims to improve image super-resolution techniques by balancing distortion reduction with perceptual quality improvement. It introduces a new framework called Toward Improving Super-Resolution, which focuses on producing high-quality knee magnetic resonance images. The framework uses a lightweight, resolution-independent, feedforward convolutional network with 266,000 parameters, which includes smoothing and denoising preprocessing and Leaky Rectified Linear Unit activations for stable training. The model builds on a baseline deep learning architecture to improve training stability and visual quality while maintaining computational efficiency. The fast Magnetic Resonance Imaging knee dataset was compared to established super-resolution methods like Super-Resolution Convolutional Neural Network, Very Deep Super-Resolution, and Enhanced Deep Super-Resolution. Toward Improving Super-Resolution achieved a peak signal-to-noise ratio of 38.405 ± 0.129 decibels and a structural similarity index of 0.9815 ± 0.0021, surpassing Super-Resolution Convolutional Neural Network, Very Deep Super-Resolution, and Enhanced Deep Super-Resolution. It maintained high performance at scales three and four, demonstrating accuracy and statistical robustness. The study shows that the proposed framework can enhance diagnostic imaging, reduce the need for repeated scans, and speed up clinical decision-making.]]>
Copyrights © 2025
