Autoencoder-Inspired Convolutional Network-Based Super-Resolution Method in MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Convolutional Neural Network (CNN)-based super-resolution methods for Magnetic Resonance Imaging (MRI) can lose image information due to pooling layers.
• Developing efficient and effective MRI resolution enhancement techniques is crucial for improved diagnostic accuracy.

Purpose of the Study:

• To introduce an Autoencoder-inspired Convolutional Network-based Super-resolution (ACNS) method for estimating High-Resolution (HR) MRIs from Low-Resolution (LR) MRIs.
• To evaluate the performance of ACNS against existing super-resolution methods in terms of image quality and computational speed.

Main Methods:

• Developed the ACNS method, incorporating a deconvolution layer to extrapolate spatial information via nonlinear mapping between LR and HR MRI features.
• Conducted simulation experiments using virtual phantom images and thoracic MRIs from four volunteers.
• Compared ACNS with Super-Resolution Convolutional Neural Network (SRCNN), Fast Super-Resolution Convolutional Neural Network (FSRCNN), and Deeply-Recursive Convolutional Network (DRCN) using Peak Signal-to-Noise Ratio (PSNR), Structure SIMilarity index (SSIM), Information Fidelity Criterion (IFC), and computational time.

Main Results:

• ACNS achieved comparable PSNR, SSIM, and IFC results to SRCNN, FSRCNN, and DRCN.
• ACNS demonstrated significantly faster average computation speeds: 6x faster than SRCNN, 4x faster than FSRCNN, and 35x faster than DRCN.
• The average computation time per image for ACNS was as low as 0.15 seconds.

Conclusions:

• The ACNS method offers an efficient approach to MRI super-resolution, providing high-quality results.
• ACNS presents a significant advancement in MRI resolution enhancement, balancing image fidelity with computational speed.