Revolutionizing MRI Reconstruction from Compressed Sensing k-space measurements: An AI-Driven Feature Extraction Approach

Magnetic Resonance Imaging (MRI) has become one of the most effective imaging techniques for diagnosing illnesses and injuries. However, it takes longer time for acquisition due to hardware limitations, and improving scanning time using the standard method reduces the quality of the MRI image. Compressed sensing (CS) is a modern method that can reduce overall scanning time while maintaining image quality. Because the CS reconstruction algorithm uses fewer measurements than the Nyquist rate to reconstruct high-quality images, choosing a suitable reconstruction technique is difficult. This paper proposes a new way to reconstruct MRI scans by using a Residual U-Net and CS theory. We use preprocessing and adjust the structure of deep learning networks to obtain high-quality pictures from low data in k-space. Researchers found that the suggested technique worked well for data obtained from a 2D MRI. Numerous tests were performed using an MRI dataset and the findings were compared with current approaches. Two prominent assessment metrics, peak signal-to-noise ratio and structure similarity index measures, were used to evaluate the suggested algorithm. The experimental findings demonstrate the effectiveness of the proposed approach for reconstructing high-quality MRI from CS k-space data in terms of faster acquisition without sacrificing quality.