Abstract:Background and objective: Heart disease is a vital disease that has threatened human health, and is the number one killer of human life. Moreover, with the added influence of recent health factors, its incidence rate keeps showing an upward trend. Today, cardiac magnetic resonance (CMR) imaging can provide a full range of structural and functional information for the heart, and has become an important tool for the diagnosis and treatment of heart disease. Therefore, improving the image resolution of CMR has an important medical value for the diagnosis and condition assessment of heart disease. At present, most single-image super-resolution (SISR) reconstruction methods have some serious problems, such as insufficient feature information mining, difficulty to determine the dependence of each channel of feature map, and reconstruction error when reconstructing high-resolution image. Methods: To solve these problems, we have proposed and implemented a dual U-Net residual network (DURN) for super-resolution of CMR images. Specifically, we first propose a U-Net residual network (URN) model, which is divided into the up-branch and the down-branch. The up-branch is composed of residual blocks and up-blocks to extract and upsample deep features; the down-branch is composed of residual blocks and down-blocks to extract and downsample deep features. Based on the URN model, we employ this a dual U-Net residual network (DURN) model, which combines the extracted deep features of the same position between the first URN and the second URN through residual connection. It can make full use of the features extracted by the first URN to extract deeper features of low-resolution images. Results: When the scale factors are 2, 3, and 4, our DURN can obtain 37.86 dB, 33.96 dB, and 31.65 dB on the Set5 dataset, which shows (i) a maximum improvement of 4.17 dB, 3.55 dB, and 3.22dB over the Bicubic algorithm, and (ii) a minimum improvement of 0.34 dB, 0.14 dB, and 0.11 dB over the LapSRN algorithm. Conclusion: Comprehensive experimental study results on benchmark datasets demonstrate that our proposed DURN can not only achieve better performance for peak signal to noise ratio (PSNR) and structural similarity index (SSIM) values than other state-of-the-art SR image algorithms, but also reconstruct clearer super-resolution CMR images which have richer details, edges, and texture.

A U-Residual Local Window Transformer for Ultrasound Images Super-Resolution Reconstruction

Super-resolution Reconstruction Algorithms Based on Fusion of Deep Learning Mechanism and Wavelet

Unsupervised Super-Resolution Framework For Medical Ultrasound Images Using Dilated Convolutional Neural Networks

Progressive Residual Learning with Memory Upgrade for Ultrasound Image Blind Super-Resolution.

Improved wavelet prediction superresolution reconstruction based on U‐Net

Superresolution Reconstruction Algorithm of Ultrasonic Logging Images Based on High-Frequency Enhancement

Single image super resolution via wavelet transform fusion and SRFeat network

Residual Dense Swin Transformer for Continuous Depth-Independent Ultrasound Imaging

Image Super-Resolution Reconstruction Based on Wavelet Domain

Local Implicit Wavelet Transformer for Arbitrary-Scale Super-Resolution

Super-Resolution Reconstruction of Plane-Wave Ultrasound Image Based on a Multi-Angle Parallel U-Net with Maxout Unit and Novel Loss Function

Improvement of Low-Frequency Ultrasonic Image Quality Using a Enhanced Convolutional Neural Network

A High-Efficiency Super-Resolution Reconstruction Method For Ultrasound Microvascular Imaging

UFSRNet: U-shaped face super-resolution reconstruction network based on wavelet transform

A Lightweight Image Super-Resolution Reconstruction Algorithm Based on the Residual Feature Distillation Mechanism

Dual U-Net residual networks for cardiac magnetic resonance images super-resolution

High-resolution Image Reconstruction for Portable Ultrasound Imaging Devices

Super‐resolution reconstruction algorithm for medical images by fusion of wavelet transform and multi‐scale adaptive feature selection

Super-resolution Reconstruction of MR Image with a Novel Residual Learning Network Algorithm

Super-resolution reconstruction of ultrasound image using a modified diffusion model

Underwater Image Super-Resolution Reconstruction with Local Self-Similarity Analysis and Wavelet Decomposition