An efficient medical image super resolution based on piecewise linear regression strategy using domain transform filtering

Abstract

Summary High quality medical images are expected for the detailed analysis in medical diagnostic system. However, spatial resolution of medical images mostly suffers from the factors such as medical equipment and time constraints. Despite these limitations, a well‐designed super‐resolution (SR) algorithm will help to improve the resolution of the medical images for medical diagnosis. When comparing to upgrade medical equipment, the adaptation of SR algorithms as a post‐processing method after medical imaging gives the benefits of lower cost and superior performance. This article proposes a piecewise linear regression system for learning image specific low‐to‐high resolution mapping via domain transform filtering and weighted least squares (WLS) optimization framework. We initially employed a WLS optimization framework for gradually coarsen the original input images and extract the multi‐scale information by constructing edge‐preserving multi‐scale decompositions. Then, with the aim of adequately preserving the edges of the medical images, we utilize a recursive filtering in transform domain. The Hadamard patterns generated from low‐resolution training patches are then used to classify the LR–HR training patch pairs into different classes. In the end, the piecewise linear regression system is utilized to learn the mapping relationship between LR space to HR space for each class, which is subsequently utilized to obtain desired HR image. In the context of quantitative metrices and qualitative analysis, our proposed method generates high‐quality medical images as compares to other existing state‐of‐the‐art SR methods.

Special Issue: Recent advances in quantum computing and quantum neural networks

“…In the article by Lepcha et al, 5 a piecewise regression model is proposed for learning the images through the mapping of resolution images from low to high. This is performed by utilizing filtering of the domain transform and an optimized framework of weighted least squares.…”