calf – Software for CEST Analysis with Lorentzian Fitting

Abstract

Analysis of chemical exchange saturation transfer (CEST) MRI data requires sophisticated methods to obtain reliable results about metabolites in the tissue under study. CEST generates z-spectra with multiple components, each originating from individual molecular groups. The individual lines with Lorentzian line shape are mostly overlapping and disturbed by various effects. We present an elaborate method based on an adaptive nonlinear least squares algorithm that provides robust quantification of z-spectra and incorporates prior knowledge in the fitting process. To disseminate CEST to the research community, we developed software as part of this study that runs on the Microsoft Windows operating system and will be made freely available to the community. Special attention has been paid to establish a low entrance threshold and high usability, so that even less experienced users can successfully analyze CEST data.

Deep Learning-Based Denoising of CEST MR Data: A Feasibility Study on Applying Synthetic Phantoms in Medical Imaging

“…The application of NLM as a denoising technique in MRI has been validated via numerous studies and its utility in CEST studies is also well documented [7,10,38,39]. In a parallel development, BM3D has surfaced as a modern extension, mainly in the field of image processing.…”

“…showcased a k-means clustering strategy designed to accelerate Lorentzian evaluations while inherently reducing noise [53]. Further, as discussed, methods such as the Image Downsampling Expedited Adaptive Least-squares (IDEAL) [7,54] have been proposed as effective alternatives for reducing noise during Lorentzian analyses. (4) Noise Model Disparities: As we emphasized in our discussion, the idealized noise models used during our training sessions seemed misaligned with the intricate noise landscapes of in vivo experiments, particularly due to variables like respiratory or intestinal movement-related signal fluctuations.…”

“…NLM: NLM is an adaptive filtering method that exploits image redundancy to mitigate noise [7,38,39]. For any pixel p and its associated search window W, the weight of neighboring pixels is determined using the likeness of their local surroundings.…”

“…While conventional methodologies, including Magnetization Transfer Ratio Asymmetry (MTR asym ) and Lorentzian analyses, are effective metrics for CEST imaging [6,7], they do exhibit susceptibility to image noise [5,8]. Mathematically, MTR asym is defined as the difference between the signals at positive and negative frequency offsets with respect to the water resonance.…”

“…However, one must acknowledge the omnipresence of noise across all image-processing modalities. In response, a slew of noise-reduction algorithms have been introduced in recent years, notable among which are principal component analysis (PCA), which allows noise reduction based on ordering noise and signal components in the Z-spectrum signal curve [8,9], nonlocal mean filtering (NLM), which is based on a weighted signal adjustment of the nearest neighbors [7,10], and block matching combined with 3D filtering (BM3D), which associates 2D blocks of the image with similar areas in the image to reduce noise [10][11][12].…”