Dilated Deep Residual Network for Image Denoising

Low-dose CT denoising is a challenging task that has been studied by many researchers. Some studies have used deep neural networks to improve the quality of low-dose CT images and achieved fruitful results. In this paper, we propose a deep neural network that uses dilated convolutions with different dilation rates instead of standard convolution helping to capture more contextual information in fewer layers. Also, we have employed residual learning by creating shortcut connections to transmit image information from the early layers to later ones. To further improve the performance of the network, we have introduced a non-trainable edge detection layer that extracts edges in horizontal, vertical, and diagonal directions. Finally, we demonstrate that optimizing the network by a combination of mean-square error loss and perceptual loss preserves many structural details in the CT image. This objective function do not suffer from over smoothing and blurring effects caused by per-pixel loss and grid-like artifacts resulting from perceptual loss. The experiments show that each modification to the network improves the outcome while only minimally changing the complexity of the network.

show abstract

“…Receptive field of the layer L (RF L ) with filter size f × f and dilation rate of r can be computed from the equation 5 [40].…”

Section: Dilated Convolutionmentioning

confidence: 99%

“…To better understand the capability of dilated convolution, Wang et al replaced the standard convolutions in [41] with dilated convolutions with r = 2 and achieved comparable performance in only 10 layers instead of 17 layers [40].…”

Section: Dilated Convolutionmentioning

confidence: 99%

Deep Learning for Low-Dose CT Denoising Using Perceptual Loss and Edge Detection Layer

2019

View full text Add to dashboard Cite

show abstract

“…Moreover, in the innermost bottleneck, we applied four dilated convolutions of rates 2, 4, 8, and 16 to provide additional, broader spatial context to the neurons. Dilated convolutions broaden the receptive field of the network without increasing the network depth (by adding layers of downsampling operations), which is computationally more efficient . Also large and small dilation rates can effectively restore the textures and edges in the image for denoising applications …”

Section: Methodsmentioning

confidence: 99%

“…Dilated convolutions broaden the receptive field of the network without increasing the network depth (by adding layers of downsampling operations), which is computationally more efficient. 23 Also large and small dilation rates can effectively restore the textures and edges in the image for denoising applications. 24 Simple filters can have easily provable properties, like shift invariance meaning the filter does not shift the signal, which is a useful guarantee in denoising.…”

Section: B Dilated U-netmentioning

confidence: 99%

Mitigating inherent noise in Monte Carlo dose distributions using dilated U‐Net

et al. 2019

View full text Add to dashboard Cite

Purpose Monte Carlo (MC) algorithms offer accurate modeling of dose calculation by simulating the transport and interactions of many particles through the patient geometry. However, given their random nature, the resulting dose distributions have statistical uncertainty (noise), which prevents making reliable clinical decisions. This issue is partly addressable using a huge number of simulated particles but is computationally expensive as it results in significantly greater computation times. Therefore, there is a trade‐off between the computation time and the noise level in MC dose maps. In this work, we address the mitigation of noise inherent to MC dose distributions using dilated U‐Net — an encoder–decoder‐styled fully convolutional neural network, which allows fast and fully automated denoising of whole‐volume dose maps. Methods We use mean squared error (MSE) as loss function to train the model, where training is done in 2D and 2.5D settings by considering a number of adjacent slices. Our model is trained on proton therapy MC dose distributions of different tumor sites (brain, head and neck, liver, lungs, and prostate) acquired from 35 patients. We provide the network with input MC dose distributions simulated using 1×106 particles while keeping 1×109 particles as reference.ResultsAfter training, our model successfully denoises new MC dose maps. On average (averaged over five patients with different tumor sites), our model recovers D95 of 55.99 Gy from the noisy MC input of 49.51 Gy, whereas the low noise MC (reference) offers 56.03 Gy. We observed a significant reduction in average RMSE (thresholded >10% max ref) for reference vs denoised (1.25 Gy) than reference vs input (16.96 Gy) leading to an improvement in signal‐to‐noise ratio (ISNR) by 18.06 dB. Moreover, the inference time of our model for a dose distribution is less than 10 s vs 100 min (MC simulation using 1×109 particles).ConclusionsWe propose an end‐to‐end fully convolutional network that can denoise Monte Carlo dose distributions. The networks provide comparable qualitative and quantitative results as the MC dose distribution simulated with 1×109 particles, offering a significant reduction in computation time.

show abstract

“…Further researcher started applying in deep learning. W. Bae, J. Yoo, and J. C. Ye [14]Proposed homology guided manifold simplification and compared with state of art algorithms. Tianyang Wang, Mingxuan Sun, and Kaoning Hu.…”

Section: Related Workmentioning

confidence: 99%

Sparse Residual Learning of Deep Convolution Network for De-Noising Patch Based Block Match Three Dimension Algorithm

Kamalakshi¹

2018

IJRASET

View full text Add to dashboard Cite

This paper introduces a unique approach to de-noise an image based on concepts of Deep Convolution Neural Networks (DCNN) with sparse residual learning sparse reconstruction and batch normalization. The basic concept is modification of existing block match three dimension algorithm in which similar local patches in the input image are integrated into a 3D block. Here first patches are retrieved the features are extracted. The de-noised image is employed as a basic estimate for the block matching, and then de-noising function for the block is learned by a DCNN structure. Most of the residual network has many residual units (i.e., identity shortcuts), our method employs a single scarified residual unit to classify the residual image. Experimental results demonstrate the effectiveness of the sparse residual learning, sparse reconstruction and batch normalization in the tasks of image de-noising. Our experiment results proves that our model provide better efficiency in terms of PSNR. Keywords: Noise, patch, BM3D, sparse residual, sparse reconstruction batch normalization I. INTRODUCTIONRecently, image de-noising methods has gained popularity by a method called patch based method or non local means. This approach is measured an incredible in most of current state-of-the-art methods. The concept employed is to find related patterns that occur randomly all across the image and the image patches that have related patterns can be located far from each other. The patch based approach is a determining work that exploit this NSS prior [1]. The employment of patch based approach has boosted the performance of image de-noising significantly. The best example is the Block Matching and 3D Filtering (BM3D) method [2] which is a very good in performance and highly engineered approach that made the state-of-the-art record in image de-noising stay ahead for almost a decade. In past decade ,Machine learning is gaining popularity and progressively escalating its prominence. Among these deep learning concepts are overtaking shallow learning methods. It is a sort of overhyped. And very potential results have been noticed for image processing applications such as image restoration class. The significant improvement in the performance can be achieved by deep networks is due to their advanced modeling capabilities, deep structure and the adaption of non-linearities that in fact can be combined with qualified learning on large training datasets. Among all the deep learning methods, the convolutional neural networks have shown great performance for image processing tasks because of the reason of its quite easy access to large-scale dataset and the advances in deep learning methods. The proposed work is a modification of BM3D[2] [3][4] where CNN with sparse residual Learning ,sparse reconstruction and batch normalization It also adopts the residual learning formulation .

show abstract

Dilated Deep Residual Network for Image Denoising

Cited by 83 publications

References 27 publications

Deep Learning for Low-Dose CT Denoising Using Perceptual Loss and Edge Detection Layer

Deep Learning for Low-Dose CT Denoising Using Perceptual Loss and Edge Detection Layer

Mitigating inherent noise in Monte Carlo dose distributions using dilated U‐Net

Sparse Residual Learning of Deep Convolution Network for De-Noising Patch Based Block Match Three Dimension Algorithm

Contact Info

Product

Resources

About