An official website of the United States government
This paper introduces a new image smoothing filter based on a feed-forward convolutional neural network (CNN) in presence of impulse noise. This smoothing filter integrates a very deep architecture, a regularization method, and a batch normalization process. This fully integrated approach yields an effectively denoised and smoothed image yielding a high similarity measure with the original noise free image. Specific structural metrics are used to assess the denoising process and how effective was the removal of the impulse noise. This CNN model can also deal with other noise levels not seen during the training phase. The proposed CNN model is constructed through a 20-layer network using 400 images from the Berkeley Segmentation Dataset (BSD) in the training phase. Results are obtained using the standard testing set of 8 natural images not seen in the training phase. The merits of this proposed method are weighed in terms of high similarity measure and structural metrics that conform to the original image and compare favorably to the different results obtained using state-of-art denoising filters.
more »
« less
Denoising diffusion weighted imaging data using convolutional neural networks
Diffusion weighted imaging (DWI) with multiple, high b-values is critical for extracting tissue microstructure measurements; however, high b-value DWI images contain high noise levels that can overwhelm the signal of interest and bias microstructural measurements. Here, we propose a simple denoising method that can be applied to any dataset, provided a low-noise, single-subject dataset is acquired using the same DWI sequence. The denoising method uses a one-dimensional convolutional neural network (1D-CNN) and deep learning to learn from a low-noise dataset, voxel-by-voxel. The trained model can then be applied to high-noise datasets from other subjects. We validated the 1D-CNN denoising method by first demonstrating that 1D-CNN denoising resulted in DWI images that were more similar to the noise-free ground truth than comparable denoising methods, e.g., MP-PCA, using simulated DWI data. Using the same DWI acquisition but reconstructed with two common reconstruction methods, i.e. SENSE1 and sum-of-square, to generate a pair of low-noise and high-noise datasets, we then demonstrated that 1D-CNN denoising of high-noise DWI data collected from human subjects showed promising results in three domains: DWI images, diffusion metrics, and tractography. In particular, the denoised images were very similar to a low-noise reference image of that subject, more than the similarity between repeated low-noise images (i.e. computational reproducibility). Finally, we demonstrated the use of the 1D-CNN method in two practical examples to reduce noise from parallel imaging and simultaneous multi-slice acquisition. We conclude that the 1D-CNN denoising method is a simple, effective denoising method for DWI images that overcomes some of the limitations of current state-of-the-art denoising methods, such as the need for a large number of training subjects and the need to account for the rectified noise floor.
more »
« less
- PAR ID:
- 10377352
- Editor(s):
- Yap, Pew-Thian
- Date Published:
- Journal Name:
- PLOS ONE
- Volume:
- 17
- Issue:
- 9
- ISSN:
- 1932-6203
- Page Range / eLocation ID:
- e0274396
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
ABSTRACT PurposeTo achieve high‐resolution, three‐dimensional (3D) quantitative diffusion‐weighted MR spectroscopic imaging (DW‐MRSI) for molecule‐specific microstructural imaging of the brain. MethodsWe introduced and integrated several innovative acquisition and processing strategies for DW‐MRSI: (a) a new double‐spin‐echo sequence combining selective excitation, bipolar diffusion encoding, rapid spatiospectral sampling, interleaved water spectroscopic imaging data, and a special sparsely sampled echo‐volume‐imaging (EVI)‐based navigator, (b) a rank‐constrained time‐resolved reconstruction from the EVI data to capture spatially varying phases, (c) a model‐based phase correction for DW‐MRSI data, and (d) a multi‐b‐value subspace‐based method for water/lipids removal and spatiospectral reconstruction using learned metabolite subspaces, and e) a hybrid subspace and parametric model‐based parameter estimation strategy. Phantom and in vivo experiments were performed to validate the proposed method and demonstrate its ability to map metabolite‐specific diffusion parameters in 3D. ResultsThe proposed method generated reproducible metabolite diffusion coefficient estimates, consistent with those from a standard single‐voxel DW spectroscopy (SV‐DWS) method. High‐SNR multi‐molecular mean diffusivity (MD) maps can be obtained at a 6.9 6.9 7.0 mm nominal resolution with large 3D brain coverage. High‐resolution (4.4 4.4 5.6 mm) metabolite and diffusion coefficient maps can be obtained within 20 mins for the first time. Tissue‐dependent metabolite MDs were observed, i.e., larger MDs for NAA, creatine, and choline in white matter than gray matter, with region‐specific differences. ConclusionWe demonstrated an unprecedented capability of simultaneous, high‐resolution metabolite and diffusion parameter mapping. This imaging capability has strong potential to offer richer molecular and tissue‐compartment‐specific microstructural information for various clinical and neuroscience applications.more » « less
-
Purpose: Magnetic Resonance Imaging (MRI) enables non‐invasive assessment of brain abnormalities during early life development. Permanent magnet scanners operating in the neonatal intensive care unit (NICU) facilitate MRI of sick infants, but have long scan times due to lower signal‐to‐noise ratios (SNR) and limited receive coils. This work accelerates in‐NICU MRI with diffusion probabilistic generative models by developing a training pipeline accounting for these challenges. Methods: We establish a novel training dataset of clinical, 1 Tesla neonatal MR images in collaboration with Aspect Imaging and Sha'are Zedek Medical Center. We propose a pipeline to handle the low quantity and SNR of our real‐world dataset (1) modifying existing network architectures to support varying resolutions; (2) training a single model on all data with learned class embedding vectors; (3) applying self‐supervised denoising before training; and (4) reconstructing by averaging posterior samples. Retrospective under‐sampling experiments, accounting for signal decay, evaluated each item of our proposed methodology. A clinical reader study with practicing pediatric neuroradiologists evaluated our proposed images reconstructed from under‐sampled data. Results: Combining all data, denoising pre‐training, and averaging posterior samples yields quantitative improvements in reconstruction. The generative model decouples the learned prior from the measurement model and functions at two acceleration rates without re‐training. The reader study suggests that proposed images reconstructed from under‐sampled data are adequate for clinical use. Conclusion: Diffusion probabilistic generative models applied with the proposed pipeline to handle challenging real‐world datasets could reduce the scan time of in‐NICU neonatal MRI.more » « less
-
null (Ed.)Deep convolutional neural networks (CNNs) for image denoising are usually trained on large datasets. These models achieve the current state of the art, but they have difficulties generalizing when applied to data that deviate from the training distribution. Recent work has shown that it is possible to train denoisers on a single noisy image. These models adapt to the features of the test image, but their performance is limited by the small amount of information used to train them. Here we propose "GainTuning", in which CNN models pre-trained on large datasets are adaptively and selectively adjusted for individual test images. To avoid overfitting, GainTuning optimizes a single multiplicative scaling parameter (the "Gain") of each channel in the convolutional layers of the CNN. We show that GainTuning improves state-of-the-art CNNs on standard image-denoising benchmarks, boosting their denoising performance on nearly every image in a held-out test set. These adaptive improvements are even more substantial for test images differing systematically from the training data, either in noise level or image type. We illustrate the potential of adaptive denoising in a scientific application, in which a CNN is trained on synthetic data, and tested on real transmission-electron-microscope images. In contrast to the existing methodology, GainTuning is able to faithfully reconstruct the structure of catalytic nanoparticles from these data at extremely low signal-to-noise ratios.more » « less
-
This repository contains processed cardiac diffusion tensor imaging (cDTI) data from a study of a prospectively enrolled cohort mostly consisting of healthy volunteers. All participants were imaged using a standardized 3T cardiac MRI protocol (MAGNETOM Vida Fit, Siemens Healthineers). Standardized imaging protocols and identical radiofrequency coil configurations were used across all participants to maximize consistency and reproducibility. The overarching study was designed to establish a high-resolution reference dataset of myocardial microstructure and biomechanics in healthy subjects using multiparametric cardiac magnetic resonance imaging. The complete imaging repository, including all acquired MRI data, is available at: https://doi.org/10.25740/xt223hz2438. This repository specifically contains the processed cDTI datasets. Each participant directory ending in _Processing includes the intermediate outputs from the complete diffusion processing pipeline implemented in the Cardiac Diffusion in Python (CarDpy) toolbox. Intermediate datasets are organized by processing stage, and processed diffusion images are provided in NIfTI format together with corresponding b-value (.bval), diffusion gradient (.bvec), and header files. Participant directories ending in _Analysis contain the final curated datasets used for the quantitative analyses presented in associated publications. These folders include the fully processed cardiac diffusion tensor imaging (cDTI) datasets, corresponding left ventricular (LV) segmentations, and derived cDTI metric maps. Diffusion imaging data are provided in NIfTI format (.nii), LV segmentations are stored as Nearly Raw Raster Data (.nrrd) files, and quantitative cDTI metrics are provided as MATLAB (.mat) files to facilitate downstream analysis. This resource provides a standardized benchmark dataset for developing and validating cardiac diffusion MRI processing methods, studying myocardial microstructure and biomechanics, constructing computational cardiac models, and investigating healthy myocardial architecture as a reference for future disease studies. All data collection was conducted under Institutional Review Board (IRB) approval with written informed consent obtained from all participants.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript
- Journal Article:
- https://doi.org/10.1371/journal.pone.0274396
