AUTHOR=Schiano Di Cola Vincenzo , Mango Dea M. L. , Bottino Alessandro , Andolfo Lorenzo , Cuomo Salvatore TITLE=Magnetic resonance imaging enhancement using prior knowledge and a denoising scheme that combines total variation and histogram matching techniques JOURNAL=Frontiers in Applied Mathematics and Statistics VOLUME=9 YEAR=2023 URL=https://www.frontiersin.org/journals/applied-mathematics-and-statistics/articles/10.3389/fams.2023.1041750 DOI=10.3389/fams.2023.1041750 ISSN=2297-4687 ABSTRACT=Introduction

Brain perfusion-weighted images obtained through dynamic contrast studies play a critical and clinical role in diagnosis and treatment decisions. However, due to the patient's limited exposure to radiation, computed magnetic resonance imaging (MRI) suffers from low contrast-to-noise ratios (CNRs). Denoising MRI images is a critical task in many e-health applications for disease detection. The challenge in this research field is to define novel algorithms and strategies capable of improving accuracy and performance in terms of image vision quality and computational cost to process data. Using MRI statistical information, the authors present a method for improving image quality by combining a total variation-based denoising algorithm with histogram matching (HM) techniques.

Methods

The total variation is the Rudin–Osher–Fatemi total variation (TV-ROF) minimization approach, TV-L2, using the isotropic TV setting for the bounded variation (BV) component. The dual-stage approach is tested against two implementations of the TV-L2: the split Bregman (SB) algorithm and a fixed-point (FP) iterations scheme. In terms of HM, the study explores approximate matching and the exact histogram matching from Coltuc.

Results

As measured by the structural similarity index (SIMM), the results indicate that in the more realistic study scenarios, the FP with an HM pairing is one of the best options, with an improvement of up to 12.2% over the one without an HM.

Discussion

The findings can be used to evaluate and investigate more advanced machine learning-based approaches for developing novel denoising algorithms that infer information from ad hoc MRI histograms. The proposed methods are adapted to medical image denoising since they account for the preference of the medical expert: a single parameter can be used to balance the preservation of (expert-dependent) relevant details against the degree of noise reduction.