Noise reduction of diffusion tensor images by sparse representation and dictionary learning

Self-supervised structural similarity-based convolutional neural network for cardiac diffusion tensor image denoising

BACKGROUND

Diffusion tensor imaging (DTI) is a promising technique for non-invasively investigating the myocardial fiber structures of human heart. However, low signal-to-noise ratio (SNR) has been a major limit of cardiac DTI to prevent us from detecting myocardium structure accurately. Therefore, it is important to remove the effect of noise on diffusion weighted (DW) images.

PURPOSE

Although the conventional and deep learning-based denoising methods have shown the potential to deal with effectively the noise in DW images, most of them are redundant information dependent or require the noise-free images as golden standard. In addition, the existed DW image denoising methods often suffer from problems of over-smoothing. To address these issues, we propose a self-supervised learning model, structural similarity based convolutional neural network with edge-weighted loss (SSECNN), to remove the noise effectively in cardiac DTI.

METHODS

Considering that the DW images acquired along different diffusion directions have structural similarity, and the noise in these DW images is independent and identically distributed, the structural similarity-based matching algorithm is proposed to search for the most similar DW images. Such similar noisy DW image pairs are then used as the input and target of the denoising network SSECNN, which consists of several convolutional and residual blocks. Through the self-supervised training with these image pairs, the network can restore the clean DW images and retain the correlations between the denoised DW images along different directions. To avoid the over-smoothing problem, we design a novel edge-weighted loss which enables the network to adaptively adjust the loss weights with iterations and therefore to improve the detail preserve ability of the model. To verify the superiority of the proposed method, comparisons with state-of-the-art (SOTA) denoising methods are performed on both synthetic and real acquired DTI datasets.

RESULTS

Experimental results show that SSECNN can effectively reduce the noise in the DW images while preserving detailed texture and edge information and therefore achieve better performance in DTI reconstruction. For synthetic dataset, compared to the SOTA method, the root mean square error (RMSE), peak signal-to-noise ratio (PSNR), and structure similarity index measure (SSIM) between the denoised DW images obtained with SSECNN and noise-free DW images are improved by 6.94%, 1.98%, and 0.76% respectively when the noise level is 10%. As for the acquired cardiac DTI dataset, the SSECNN method could significantly improve SNR and contrast to noise ratio (CNR) of cardiac DW images and achieve more regular helix angle (HA) and transverse angle (TA) maps. The ablation experimental results validate that using the structure similarity-based method to search the similar DW image pairs yield the smallest loss, and with the help of the edge-weighted loss, the denoised DW images and diffusion metric maps can preserve more details.

CONCLUSIONS

The proposed SSECNN method can fully explore the similarity between the DW images along different diffusion directions. Using such similarity and an edge-weighted loss enable us to denoise cardiac DTI effectively in a self-supervised manner. Our method can overcome the redundancy information dependence and over-smoothing problem of the SOTA methods.

Deep Learning-Based Denoising for High b-Value at 2000 s/mm2 Diffusion-Weighted Imaging

Diffusion-weighted imaging (DWI) allows white matter quantification of the white matter tracts of the brain. However, at a high b-value (≥ 2000 s/mm2), DWI acquisition suffers from noise due to longer acquisition times obscuring white matter interpretation. DWI denoising techniques can be used to acquire high b-value DWI without increasing the number of signal averages. We used a residual learning-based convolutional neural network (DnCNN) to reduce noise in high b-value DWI based on the literature review. We applied the proposed denoising method on high b-value, retrospectively collected DWI data with multiple noise levels. Experimental results show an improved image quality after denoising in retrospective DWI (average PSNR before and after denoising: 27.63 ± 1.06 dB and 51.76 ± 1.95 dB, respectively). The prospective DWI included one and two signal averages for denoising. DWI with four signal averages was used as the reference. Representative images show high b-value prospective DW images denoised using the DnCNN. We demonstrated DnCNN for cases of multiple noise levels and signal averages. For the prospective study, the PSNR values for 1-NEX before and after denoising were 27.39 ± 3.75 dB and 27.68 ± 3.75 dB. For 2-NEX, the PSNR values before and after denoising were 27.51 ± 4.18 dB and 27.75 ± 4.05 dB.

High-field mr diffusion-weighted image denoising using a joint denoising convolutional neural network

BACKGROUND

Low signal-to-noise ratio (SNR) has been a major limiting factor for the application of higher-resolution diffusion-weighted imaging (DWI). Most of the conventional denoising models suffer from the drawbacks of shallow feature extraction and hand-crafted parameter tuning. Although multiple studies have shown the promising applications of image denoising using convolutional neural networks (CNNs), none of them have considered denoising multiple b-value DWIs using a multichannel CNN model.

PURPOSE

To present a joint denoising CNN (JD-CNN) model to improve the SNR of multiple b-value DWI.

STUDY TYPE

Retrospective technical development.

POPULATION

Twenty healthy rats and two rats with clinically confirmed focal cortical dysplasia were included to evaluate the performance of the proposed method.

FIELD STRENGTH/SEQUENCE

11.7T MRI, a multiple b-values DWI sequence.

ASSESSMENT

The total variation (TV) and BM3D denoising methods were also performed on the same dataset for comparison. Peak SNR (PSNR) and normalized mean square error (NMSE) were calculated for the assessment of image qualities.

STATISTICAL TESTS

A paired Student's t-test was conducted to compare the diffusion parameter measurements between different approaches. P < 0.01 was considered statistically significant.

RESULTS

Simulation results showed substantial improvement of image quality after JD-CNN denoising (PSNR of original image: 23.15 ± 1.77; PSNR of denoised image: 42.94 ± 2.12). The proposed method outperforms the state-of-the-art methods on high b-value DWIs in terms of PSNR (TV: 33.51 ± 0.83, BM3D: 35.12 ± 0.94, JD-CNN: 46.52 ± 0.98). In addition, the NMSE of the estimated apparent diffusion coefficient (ADC) reduces from 0.72 ± 0.13 to 0.45 ± 0.06 (P < 0.01) with the application of the JD-CNN model.

DATA CONCLUSION

The proposed method is able to remove noise with a wide range of noise levels in multiple b-value DWI and improve the diffusion parameter estimation. This shows potential clinical promise.

LEVEL OF EVIDENCE

2 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2019;50:1937-1947.

Classification of breast mass lesions on dynamic contrast-enhanced magnetic resonance imaging by a computer-assisted diagnosis system based on quantitative analysis

The aim of the current study was to develop a semi-automatic and quantitative method for the analysis of a time-intensity curve (TIC) from breast dynamic contrast-enhanced magnetic resonance imaging. The performance of the proposed method, based on the level set segmentation algorithm, was evaluated by comparison with the traditional method. In the traditional method, the lesion area is delineated manually and the corresponding mean TIC is classified subjectively as one of three washout patterns. In addition, only one quantitative parameter, the maximum slope of increase (MSI), is calculated. In the proposed method, the lesion region was determined semi-automatically and the corresponding mean TIC was categorized quantitatively. In addition to MSI, a number of quantitative parameters were derived from the mean TIC and lesion area, including signal intensity slope (SI_slope), initial percentage of enhancement (E_initial), percentage of peak enhancement (E_peak), early signal enhancement ratio (ESER) and second enhancement percentage (SEP). Wilcoxon signed-rank test and receiver operating characteristic analyses were performed for statistical analysis. For TIC categorization the accuracy was 61.54% for the traditional method and 82.05% for the proposed method. Using the proposed method, mean curve accuracies were 84.0% for SI_slope, 66.7% for MSI, 66.0% for E_initial, 66.0% for E_peak, 68.0% for ESER and 44.9% for SEP. In the lesion region, the accuracies for the aforementioned parameters were 80.8, 65.4, 66.7, 62.2, 69.2 and 57.1%, respectively. Accuracy of the MSI value derived from the traditional method was 63.4%. Compared with the traditional method, the proposed semi-automatic method in the current study may provide results with a higher accuracy to differentiate benign and malignant lesions. Therefore, the proposed method should be considered as a supplementary tool for the diagnosis of breast lesions.

A PSO-Powell Hybrid Method to Extract Fiber Orientations from ODF

High angular resolution diffusion imaging (HARDI) has opened up new perspectives for the delineation of crossing and branching fiber pathways by orientation distribution function (ODF). The fiber orientations contained in an imaging voxel are the key factor in tractography. To extract real fiber orientations from ODF, a hybrid method is proposed for computing the principal directions of ODF by combining the variation of Particle Swarm Optimization (PSO) algorithm with the modified Powell algorithm. This method is comprised of the global searching ability of PSO and the powerful local optimizing of Powell search. This combination can guarantee finding all the diffusion directions without applying sliding windows and improve the accuracy and efficiency. The proposed approach was evaluated on simulated crossing-fiber datasets, Tractometer, and in vivo datasets. The results show that this method could correctly identify fiber directions under a range of noise levels. This method was compared with the state-of-the-art methods, such as modified Powell, ball-stick model, and diffusion decomposition, showing that it outperformed them. As to the multimodal voxels where different fiber populations exist, the proposed approach allows us to improve the estimation accuracy of fiber orientations from ODF. It can play a significant role in the nerve fiber tracking.

Pulsatile flow in ventricular catheters for hydrocephalus

The obstruction of ventricular catheters (VCs) is a major problem in the standard treatment of hydrocephalus, the flow pattern of the cerebrospinal fluid (CSF) being one important factor thereof. As a first approach to this problem, some of the authors studied previously the CSF flow through VCs under time-independent boundary conditions by means of computational fluid dynamics in three-dimensional models. This allowed us to derive a few basic principles which led to designs with improved flow patterns regarding the obstruction problem. However, the flow of the CSF has actually a pulsatile nature because of the heart beating and blood flow. To address this fact, here we extend our previous computational study to models with oscillatory boundary conditions. The new results will be compared with the results for constant flows and discussed. It turns out that the corrections due to the pulsatility of the CSF are quantitatively small, which reinforces our previous findings and conclusions.This article is part of the themed issue 'Mathematical methods in medicine: neuroscience, cardiology and pathology'.