Denoising Medical Images Using Deep Learning in IoT Environment

Medical Resonance Imaging (MRI) is a noninvasive, nonradioactive, and meticulous diagnostic modality capability in the field of medical imaging. However, the efficiency of MR image reconstruction is affected by its bulky image sets and slow process implementation. Therefore, to obtain a high-quality reconstructed image we presented a sparse aware noise removal technique that uses convolution neural network (SANR_CNN) for eliminating noise and improving the MR image reconstruction quality. The proposed noise removal or denoising technique adopts a fast CNN architecture that aids in training larger datasets with improved quality, and SARN algorithm is used for building a dictionary learning technique for denoising large image datasets. The proposed SANR_CNN model also preserves the details and edges in the image during reconstruction. An experiment was conducted to analyze the performance of SANR_CNN in a few existing models in regard with peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and mean squared error (MSE). The proposed SANR_CNN model achieved higher PSNR, SSIM, and MSE efficiency than the other noise removal techniques. The proposed architecture also provides transmission of these denoised medical images through secured IoT architecture.     

An extended non‐local means algorithm: Application to brain MRI

Improved adaptive nonlocal means (IANLM) is a variant of classical nonlocal means (NLM) denoising method based on adaptation of its search window size. In this article, an extended nonlocal means (XNLM) algorithm is proposed by adapting IANLM to Rician noise in images obtained by magnetic resonance (MR) imaging modality. Moreover, for improved denoising, a wavelet coefficient mixing procedure is used in XNLM to mix wavelet sub‐bands of two IANLM‐filtered images, which are obtained using different parameters of IANLM. Finally, XNLM includes a novel parameter‐free pixel preselection procedure for improving computational efficiency of the algorithm. The proposed algorithm is validated on T1‐weighted, T2‐weighted and Proton Density (PD) weighted simulated brain MR images (MRI) at several noise levels. Optimal values of different parameters of XNLM are obtained for each type of MRI sequence, and different variants are investigated to reveal the benefits of different extensions presented in this work. The proposed XNLM algorithm outperforms several contemporary denoising algorithms on all the tested MRI sequences, and preserves important pathological information more effectively. Quantitative and visual results show that XNLM outperforms several existing denoising techniques, preserves important pathological information more effectively, and is computationallyefficient.     

A complete automated algorithm for segmentation of tissues and identification of tumor region in T1, T2, and FLAIR brain images using optimization and clustering techniques

Tissues in brain are the most complicated parts of our body, a clear examination and study are therefore required by a radiologist to identify the pathologies. Normal magnetic resonance (MR) scanner is capable of producing brain images with bounded tissues, where unique and segregated views of the tissues are required. A distinguished view upon the images is manually impossible and can be subjected to operator errors. With the assistance of a soft computing technique, an automated unique segmentation upon the brain tissues along with the identification of the tumor region can be effectively done. These functionalities assist the radiologist extensively. Several soft computing techniques have been proposed and one such technique which is being proposed is PSO‐based FCM algorithm. The results of the proposed algorithm is compared with fuzzy C‐means (FCM) and particle swarm optimization (PSO) algorithms using comparison factors such as mean square error (MSE), peak signal to noise ratio (PSNR), entropy (energy function), Jaccard (Tanimoto Coefficient) index, dice overlap index and memory requirement for processing the algorithm. The efficiency of the PSO‐FCM algorithm is verified using the comparison factors.     

Analysis of effect of cycle spinning on wavelet- and curvelet-based denoising methods on brain CT images

The purpose of this paper is to carry out the assessment of effect of cycle spinning on wavelet- and curvelet-based noise reduction methods on brain CT images. In particular, multiscale curvelet- and wavelet-based denoising methods are evaluated with and without cycle spinning. This assessment is focused not only on the noise suppression but also on fine details preservation. The experimental results show that the cycle spinning-based curvelet method outperforms not only other curvelet-based methods but also the wavelet-based methods. The quality assessment parameters taken in this paper are signal-to-noise ratio (SNR), peak-signal-to-noise ratio (PSNR), universal quality index (UQI), structural similarity index metrics (SSIM), and edge keeping index (EKI).     

Efficient image compression techniques for compressing multimodal medical images using neural network radial basis function approach

Image compression technique is used to reduce the number of bits required in representing image, which helps to reduce the storage space and transmission cost. Image compression techniques are widely used in many applications especially, medical field. Large amount of medical image sequences are available in various hospitals and medical organizations. Large images can be compressed into smaller size images, so that the memory occupation of the image is considerably reduced. Image compression techniques are used to reduce the number of pixels in the input image, which is also used to reduce the broadcast and transmission cost in efficient form. This is capable by compressing different types of medical images giving better compression ratio (CR), low mean square error (MSE), bits per pixel (BPP), high peak signal to noise ratio (PSNR), input image memory size and size of the compressed image, minimum memory requirement and computational time. The pixels and the other contents of the images are less variant during the compression process. This work outlines the different compression methods such as Huffman, fractal, neural network back propagation (NNBP) and neural network radial basis function (NNRBF) applied to medical images such as MR and CT images. Experimental results show that the NNRBF technique achieves a higher CR, BPP and PSNR, with less MSE on CT and MR images when compared with Huffman, fractal and NNBP techniques.     

Robust brain MRI denoising and segmentation using enhanced non‐local means algorithm

Image denoising is an integral component of many practical medical systems. Non‐local means (NLM) is an effective method for image denoising which exploits the inherent structural redundancy present in images. Improved adaptive non‐local means (IANLM) is an improved variant of classical NLM based on a robust threshold criterion. In this paper, we have proposed an enhanced non‐local means (ENLM) algorithm, for application to brain MRI, by introducing several extensions to the IANLM algorithm. First, a Rician bias correction method is applied for adapting the IANLM algorithm to Rician noise in MR images. Second, a selective median filtering procedure based on fuzzy c‐means algorithm is proposed as a postprocessing step, in order to further improve the quality of IANLM‐filtered image. Third, different parameters of the proposed ENLM algorithm are optimized for application to brain MR images. Different variants of the proposed algorithm have been presented in order to investigate the influence of the proposed modifications. The proposed variants have been validated on both T1‐weighted (T1‐w) and T2‐weighted (T2‐w) simulated and real brain MRI. Compared with other denoising methods, superior quantitative and qualitative denoising results have been obtained for the proposed algorithm. Additionally, the proposed algorithm has been applied to T2‐weighted brain MRI with multiple sclerosis lesion to show its superior capability of preserving pathologically significant information. Finally, impact of the proposed algorithm has been tested on segmentation of brain MRI. Quantitative and qualitative segmentation results verify that the proposed algorithm based segmentation is better compared with segmentation produced by other contemporary techniques.     

Image speckle noise denoising by a multi-layer fusion enhancement method based on block matching and 3D filtering

In order to improve speckle noise denoising of block matching and 3D filtering (BM3D) method, an image frequency-domain multi-layer fusion enhancement method (MLFE-BM3D) based on nonsubsampled contourlet transform (NSCT) has been proposed. The method designs an NSCT hard threshold denoising enhancement to preprocess the image, then uses fusion enhancement in NSCT domain to fuse the preliminary estimation results of images before and after the NSCT hard threshold denoising, finally, BM3D denoising is carried out with the fused image to obtain the final denoising result. Experiments on natural images and medical ultrasound images show that MLFE-BM3D method can achieve better visual effects than BM3D method, the peak signal to noise ratio (PSNR) of the denoised image is increased by 0.5?dB. The MLFE-BM3D method can improve the denoising effect of speckle noise in the texture region, and still maintain a good denoising effect in the smooth region of the image.     

Magnetic resonance image denoising using nonlocal maximum likelihood paradigm in DCT‐framework

The data acquired by magnetic resonance (MR) imaging system are inherently degraded by noise that has its origin in the thermal Brownian motion of electrons. Denoising can enhance the quality (by improving the SNR) of the acquired MR image, which is important for both visual analysis and other post processing operations. Recent works on maximum likelihood (ML) based denoising shows that ML methods are very effective in denoising MR images and has an edge over the other state‐of‐the‐art methods for MRI denoising. Among the ML based approaches, the Nonlocal maximum likelihood (NLML) method is commonly used. In the conventional NLML method, the samples for the ML estimation of the unknown true pixel are chosen in a nonlocal fashion based on the intensity similarity of the pixel neighborhoods. Euclidean distance is generally used to measure this similarity. It has been recently shown that computing similarity measure is more robust in discrete cosine transform (DCT) subspace, compared with Euclidean image subspace. Motivated by this observation, we integrated DCT into NLML to produce an improved MRI filtration process. Other than improving the SNR, the time complexity of the conventional NLML can also be significantly reduced through the proposed approach. On synthetic MR brain image, an average improvement of 5% in PSNR and 86%reduction in execution time is achieved with a search window size of 91 × 91 after incorporating the improvements in the existing NLML method. On an experimental kiwi fruit image an improvement of 10% in PSNR is achieved. We did experiments on both simulated and real data sets to validate and to demonstrate the effectiveness of the proposed method. © 2015 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 25, 256–264, 2015     

Enhanced brain image retrieval using carrier frequency offset compensated orthogonal frequency division multiplexing for telemedicine applications

Wireless technology plays a vital role in the Electronic health (e‐health) applications such as telemedicine and remote patient monitoring. One of the major challenges of telemedicine applications is ensuring adequate quality of service and realizing precise diagnosis. Telemedicine employs image fusion to enhance the quality of medical image for better diagnosis. However, the transmission of medical image needs high data rate and bandwidth. Due to the robustness to multipath fading, high data rate transmission capability, high flexibility, high bandwidth efficiency, and easy implementation using fast Fourier transform, an orthogonal frequency division multiplexing (OFDM) is the suitable technique for medical image transmission in telemedicine applications This paper is focused on improving the quality of the fused brain image transmission over OFDM based telemedicine service using comb type pilot based carrier frequency offset compensation. The simulation results show that the proposed method offers better performance in terms of mean squared error, Euclidean distance, peak signal to noise ratio, bit error rate, carrier to interference ratio, and computational complexity.     

闪光CCD图像的中值-非线性扩散滤波

根据闪光CCD图像的特点,提出了一种中值-非线性扩散滤波(Median-NonlinearDiffusionFiltering,简称MNDF)方法。该方法采用中值预滤波来估计图像的真实边缘,通过求解偏微分方程(PartialDifferentialEquation,简称PDE)来进行非线性扩散滤波,充分发挥了中值滤波和非线性扩散滤波的优势,能更好地消除噪声、保护边缘。实验结果表明,在高斯噪声和脉冲噪声同时存在的情况下,MNDF方法取得的滤波效果较P-M方案和Catte方案要好,信噪比改善因子提高3~5倍,均方误差减小1.3~2.7倍。对闪光照相CCD图像取得了很好的消噪声结果,保护了边缘信息。     

An Improved Jellyfish Algorithm for Multilevel Thresholding of Magnetic Resonance Brain Image Segmentations

Image segmentation is vital when analyzing medical images, especially magnetic resonance (MR) images of the brain. Recently, several image segmentation techniques based on multilevel thresholding have been proposed for medical image segmentation; however, the algorithms become trapped in local minima and have low convergence speeds, particularly as the number of threshold levels increases. Consequently, in this paper, we develop a new multilevel thresholding image segmentation technique based on the jellyfish search algorithm (JSA) (an optimizer). We modify the JSA to prevent descents into local minima, and we accelerate convergence toward optimal solutions. The improvement is achieved by applying two novel strategies: Ranking-based updating and an adaptive method. Ranking-based updating is used to replace undesirable solutions with other solutions generated by a novel updating scheme that improves the qualities of the removed solutions. We develop a new adaptive strategy to exploit the ability of the JSA to find a best-so-far solution; we allow a small amount of exploration to avoid descents into local minima. The two strategies are integrated with the JSA to produce an improved JSA (IJSA) that optimally thresholds brain MR images. To compare the performances of the IJSA and JSA, seven brain MR images were segmented at threshold levels of 3, 4, 5, 6, 7, 8, 10, 15, 20, 25, and 30. IJSA was compared with several other recent image segmentation algorithms, including the improved and standard marine predator algorithms, the modified salp and standard salp swarm algorithms, the equilibrium optimizer, and the standard JSA in terms of fitness, the Structured Similarity Index Metric (SSIM), the peak signal-to-noise ratio (PSNR), the standard deviation (SD), and the Features Similarity Index Metric (FSIM). The experimental outcomes and the Wilcoxon rank-sum test demonstrate the superiority of the proposed algorithm in terms of the FSIM, the PSNR, the objective values, and the SD; in terms of the SSIM, IJSA was competitive with the others.     

Denoising and segmentation of MR images using fourth order non-linear adaptive PDE and new convergent clustering

At present, digital image processing plays a vital role in medical imaging areas and specifically in magnetic resonance imaging (MRI) of brain images such as axial and coronal sections. This article mainly focused on the MRI brain images. The existing methods such as total variation (MC), parallel MRI, modified pyramidal dual-tree direction filter, adaptive dictionary selection algorithm, classifier methods, and fuzzy clustering techniques are poor in image eminence and precision. Thus, this article presents a novel approach consisting of denoising followed by segmentation. The objective of these proposed methods was visual eminence improvement of medical images to examine tumor extent using an adaptive partial differential equation (APDE)-based analysis with soft threshold function in denoising. The fourth order, nonlinear APDE was used to denoise the image depending on gradient and Laplacian operators associated with the new adaptive Haar-type wavelet transform. A second approach was the new convergent K-means clustering for segmentation. The convergent K-means procedure diminishes the summation of the squared deviations of structures in a cluster from the center. The significance of these proposed methods was to compute their performances in terms of mean squared error, peak signal-to-noise ratio, structure similarity, segmentation accuracy, false hit, missed-term, and elapsed time. The results were analyzed with the MATLAB software.     

A fully automated hybrid methodology using Cuckoo‐based fuzzy clustering technique for magnetic resonance brain image segmentation

This article aims at developing an automated hybrid algorithm using Cuckoo Based Search (CBS) and interval type‐2 fuzzy based clustering, so as to exhibit efficient magnetic resonance (MR) brain image segmentation. An automatic MR brain image segmentation facilitates and enables a radiologist to have a brief review and easy analysis of complicated tumor regions of imprecise gray level regions with minimal user interface. The tumor region having severe intensity variations and suffering from poor boundaries are to be detected by the proposed hybrid technique that could ease the process of clinical diagnosis and this tends to be the core subject of this article. The ability of the proposed technique is compared using standard comparison parameters such as mean squared error, peak signal to noise ratio, computational time, Dice Overlap Index, and Jaccard T animoto C oefficient Index. The proposed CBS combined with interval type‐2 fuzzy based clustering produces a sensitivity of 0.7143 and specificity of 0.9375, which are far better than the conventional techniques such as kernel based, entropy based, graph‐cut based, and self‐organizing maps based clustering. Appreciable segmentation results of tumor region that enhances clinical diagnosis is made available through this article and two of the radiologists who have hands on experience in the field of radiology have extended their support in validating the efficiency of the proposed methodology and have given their consent in utilizing the proposed methodology in the processes of clinical oncology.     

基于无监督学习的高分辨光声重建

光声断层成像(Optoacoustic Tomography,OAT)是一种新兴的生物医学成像技术,在基础医学研究与临床实践中具有重要作用。针对现有光声断层成像空间分辨率较低的问题,提出了一种结合物理点扩散函数(Point Spread Function,PSF)模型和卷积神经网络(Convolutional Neural Network,CNN)的新型高分辨光声重建网络方法(Physical Attention U-Net,Phys-AU-Net)。该方法采用无监督学习策略,结合物理PSF模型和基于注意力机制的U-Net网络。其中,物理PSF模型用于完成对衍射受限机制的模拟,基于注意力机制的U-Net网络用于实现对高密度重叠吸收体图像的特征提取。在二者共同作用下,Phys-AU-Net突破了声衍射极限对于OAT成像空间分辨率的限制。实验结果表明,Phys-AU-Net能够有效实现对声衍射受限光声断层图像的高分辨重建,其性能相较于U-Net网络具有较大程度提升,在结构相似性指标(Structural Similarity,SSIM)方面提升了43.5%,在峰值信噪比(Peak Sign...     

基于HSI颜色空间噪声检测的彩色图像去噪算法

目的为了有效去除彩色图像中的椒盐噪声,提高彩色图像质量。方法采用椒盐噪声检测和中值滤波相结合的方法,提出一种基于HSI颜色空间噪声检测的彩色图像去噪算法。将图像转换到HSI颜色空间,根据椒盐噪声在S通道具有极大值或极小值的特点判断出可疑椒盐噪声的位置,在H通道、I通道将可疑椒盐噪声分为噪声点和有用信号,对检测出的噪声像素进行中值滤波去噪。结果采用文中算法去噪后,验证图像主观评价值(Z)为1.30,平均PSNR为37.54,SSIM为0.99,Entropy为7.31,在主客观评价上优于现在常用算法。结论文中提出算法可以为彩色图像椒盐噪声的去噪提供理论基础,具有一定的实际应用价值。     

Attenuating noise from computed tomography medical images using a coefficients‐driven total variation denoising algorithm

The aim of image denoising is to recover a visually accepted image from its noisy observation with as much detail as possible. The noise exists in computed tomography images due to hardware errors, software faults and/or low radiation dose. Because of noise, the analysis and extraction of accurate medical information is a challenging task for specialists. Therefore, a novel modification on the total variational denoising algorithm is proposed in this article to attenuate the noise from CT images and provide a better visual quality. The newly developed algorithm can properly detect noise from the other image components using four new noise distinguishing coefficients and reduce it using a novel minimization function. Moreover, the proposed algorithm has a fast computation speed, a simple structure, a relatively low computational cost and preserves the small image details while reducing the noise efficiently. Evaluating the performance of the proposed algorithm is achieved through the use of synthetic and real noisy images. Likewise, the synthetic images are appraised by three advanced accuracy methods –Gradient Magnitude Similarity Deviation (GMSD), Structural Similarity (SSIM) and Weighted Signal‐to‐Noise Ratio (WSNR). The empirical results exhibited significant improvement not only in noise reduction but also in preserving the minor image details. Finally, the proposed algorithm provided satisfying results that outperformed all the comparative methods.     

一种结合小波变换和维纳滤波的图像去噪算法

目的为了有效消除噪声图像中的椒盐噪声、高斯噪声甚至混合噪声,结合维纳滤波的优势和小波分解各分量的特点,提出一种新的图像去噪算法。方法该算法先将含噪声图像进行小波变换,分离出1个低频分量和3个中高频分量,然后对低频分量进行自适应维纳滤波,对3个中高频分量用Canny算子提取边缘,最后将处理后的4个分量进行重构得到去噪后的图像。结果仿真结果表明,该算法对扫描仪引入的常见噪声均表现出较好的去噪效果,PSNR值均大于20 d B。尤其是对于高斯噪声和混合噪声,新算法去噪后的PSNR结果高于维纳滤波、软阈值小波滤波和文献[9]算法1~8 d B,效果较好。结论结合小波变换和维纳滤波的图像去噪算法,能够较好去除噪声图像的多种类型噪声,是一种较为优秀的去噪算法。     

基于凸包优化的盲源分离在图像去噪中的研究

目的针对高斯-脉冲混合噪声图像中难以有效去除大量奇异点或离群数据的问题,提出一种基于凸包优化的盲源分离方法来去除图像中的混合噪声。方法该方法把混合噪声和原图均看作未知的源信号,依据噪声图像中混合噪声与原图内容的加性关系建立盲源分离的模型,并利用凸包优化的方法构建源信号(凸包极点)的仿射包,然后通过最小化仿射包到凸包(噪声图像)上的投影误差,求解混合噪声和原图2个源信号,实现去噪混合噪声、复原原图的目的。结果实验结果发现,无论高斯-脉冲混合噪声强弱,该方法去噪复原后的峰值信噪比和平均结构相似性分别在39.9129 d B和0.9以上。结论由实验数据证实该方法可有效地从盲源分离的角度去除图像中高斯-脉冲混合噪声、复原原始图像。     

基于自相似性与稀疏表示的超分辨率算法

目的为了解决当前稀疏表示的超分辨率算法效果依赖参与训练的数据的问题,结合图像的自相似性,提出一种基于自相似性与稀疏表示相结合的超分辨率算法。方法算法利用图像的多维自相似性,构建多维图像金字塔,采用改进的相似块搜索策略,得到对应的高低分辨率图像块作为训练样本,然后对样本进行字典训练,最后根据稀疏表示得到超分辨率图像。结果实验结果显示,文中算法在峰值信噪比(PSNR)和结构相似度(SSIM)上优于其他算法,对于实验图像而言,PSNR平均提升了0.5 dB。结论提出的超分辨率算法未引入外部数据库,具有较好的效果,能够用于超分辨率重建。     

基于小波优化EEMD的二氧化硫检测

为解决检测SO₂时系统接收到的荧光信号微弱、易被噪声淹没的问题,提出了小波优化总体经验模态分解(EEMD)的方法对SO₂荧光信号进行去噪。搭建了SO₂紫外荧光检测系统对SO₂浓度进行检测,检测的平均相对误差为0.0011, SO2浓度与荧光强度的线性相关系数为0.9787。利用小波优化EEMD去噪方法对SO₂荧光信号去噪,得到的信噪比为224.7958,均方误差为1.01×10^-7,波形相似系数为0.9973,SO2浓度与荧光强度的线性相关系数为0.9936。相比于EEMD和小波去噪方法,小波优化的EEMD去噪方法对SO2荧光信号的去噪效果更好,能够达到理想的去噪效果。