A Deep Learning Framework for Image Super-Resolution for Late Gadolinium Enhanced Cardiac MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Research

Background:

• Cardiac MRI offers high in-plane resolution but suffers from low through-plane resolution, leading to anisotropic 3D images.
• Anisotropic images in Late Gadolinium Enhanced (LGE) cardiac MRI hinder accurate diagnosis of myocardial fibrosis in conditions like myocardial infarction and atrial fibrillation.

Purpose of the Study:

• To develop a self-supervised deep learning approach for enhancing the through-plane resolution of LGE cardiac MRI images.
• To improve the diagnostic quality of 3D cardiac MRI by addressing through-plane resolution limitations.

Main Methods:

• A convolutional neural network (CNN) was trained using a self-supervised approach on patches from short-axis LGE MRI images.
• The CNN leverages high-resolution in-plane information to enhance the lower-resolution through-plane data.
• Experiments were conducted on the 2018 atrial segmentation challenge dataset.

Main Results:

• The proposed method achieved a mean Peak Signal-to-Noise Ratio (PSNR) of 36.99 and 35.92 for scale factors of 2 and 4, respectively.
• A mean Structural Similarity Index Measure (SSIM) of 0.9 and 0.84 was obtained for scale factors of 2 and 4, respectively.
• Demonstrated significant improvement in through-plane resolution for LGE cardiac MRI.

Conclusions:

• The self-supervised deep learning method effectively enhances through-plane resolution in LGE cardiac MRI.
• Improved image quality facilitates more accurate detection and assessment of myocardial fibrosis.
• This technique holds potential for advancing cardiovascular disease diagnosis using MRI.