Conference PaperPDF Available

Denoising Medical X-ray Images Using Block-matching and 3D Filtering

Authors:
  • Iskenderun Technical University

Abstract and Figures

In recent days, machine learning algorithms and method diversity have increased popularity with the spread of image processing applications. The most suitable parameters for noise elimination are calculated using the Block matching algorithm on the chest radiography image. The BM3D algorithm is used to reconstruct chest x-ray medical images. Significant reprimanding performance in staining steps for clinical MRI images was obtained by optimizing cost functions for noise reduction (2,4). In this study, the effect of filtering, which is one of the steps of image processing, on chest x-rays experimented. The various classifiers were trained to evaluate the diagnostic performance of the BM3D. The performance of the classifiers for the chest x-rays was examined using the BM3D filtering method and successful denoising results were observed using the BM3D method.
Content may be subject to copyright.
2010
ISBN: 978-605-68970-3-0
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
1
Conference Proceedings Book
ISBN: 978-605-68970-3-0
Editors :
Assoc. Prof. Dr. Yakup Kutlu
Assoc. Prof. Dr. Sertan Alkan
Assist. Prof. Dr. İpek Abasıkeleş Turgut
Assist. Prof. Dr. Ahmet Gökçen
Assist. Prof. Dr. Gökhan Altan
Technical Editors:
Dr. Ezgi Zorarpacı
Dr. Mehmet Sarıgül
Dr. Levent Karacan
Merve Nilay Aydın
Handan Gürsoy Demir
Hüseyin Atasoy
Kadir Tohma
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
9
2-bit Programmable Metasurface and Antenna Application ................................................ 164
The Effect of Industry 4.0 on Occupational Safety Policies ................................................. 170
Execution of Particle Swarm Optimization Algorithm according to Acceleration Coefficients
and Inertia Weight Strategies .............................................................................................. 171
Nowcasting Applications in the Economy........................................................................... 181
Technologies Used in Real-Time Wireless Sensor Networks .............................................. 191
Applying Capsule Network on Kannada-MNIST Handwritten Digit Dataset ...................... 193
Tuning of PID Parameters by Using Imperialist Competitive Algorithm for a Real-Time
Control System ................................................................................................................... 200
Big Data in Health Care ...................................................................................................... 207
Log Analysis of a Large Scale Network by Using Elastic Stack .......................................... 208
Estimation of Lead and Zinc Adsorption Amount on Activated Carbon Using Artificial
Neural Network .................................................................................................................. 220
Classification of EMG Signals Using Myo Armband .......................................................... 227
Double Robot Arm Movement Planning Using Genetic Algorithm ..................................... 229
Remote Robot Application with Hoverboard ...................................................................... 236
Feature Selection Approach Based on Global Sensitivity .................................................... 244
Denoising Medical X-ray Images Using Block-matching and 3D Filtering ......................... 245
Investigation of Nutrition and Dietetic Application of Artificial Intelligence ...................... 252
Developing the Design Criteria for Augmented Reality Applications on See-Through Head
Mounted Displays............................................................................................................... 261
Feature Extraction and Classification of Gender from Iris images ....................................... 262
Thermal Stress Analysis of Discs Exposed To Parabolic Increasing Temperature Distribution
........................................................................................................................................... 268
A Fuzzy Approach for Multi- Period Sustainable Supplier Evaluation and Order Allocation
Problem .............................................................................................................................. 279
Optimal Tuning of PID Controller Parameters for DC Motor based on Tug of War
Optimization Algorithm...................................................................................................... 280
Estimation of Solar Energy Potential in Iskenderun Region by ANN, MLRM and Empirical
Models ............................................................................................................................... 287
Forecasting Wind Power Generation Using Machine Learning Techniques: A Case Study . 297
Efficient Workload Distribution to Welding Robots............................................................ 298
Common fixed point theorem for Two Pairs of Weakly Compatible Mappings Satisfying a
Larger Generalized (S,T)-Condition ................................................................................... 310
Implementation of KNN Classification Algorithm on Altman Z-Score Model in Data Mining
........................................................................................................................................... 316
Common Fixed Point Theorem for Mappings Satisfying a Larger Generalized Contraction in
Partially Ordered Metric Spaces ......................................................................................... 317
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
245
Denoising Medical X-ray Images Using Block-matching and 3D filtering
Süleyman Serhan Narlı1, Gökhan Altan2
1,2 Department of Computer Engineering, Iskenderun Technical University, Turkey
Emails: serhan.narli@gmail.com , gokhan.altan@iste.edu.tr
Abstract
In recent days, machine learning algorithms and method diversity have increased
popularity with the spread of image processing applications. The most suitable parameters for
noise elimination are calculated using the Block matching algorithm on the chest radiography
image. The BM3D algorithm is used to reconstruct chest x-ray medical images. Significant
reprimanding performance in staining steps for clinical MRI images was obtained by
optimizing cost functions for noise reduction (2, 4).
In this study, the effect of filtering, which is one of the steps of image processing, on chest x-
rays experimented. The various classifiers were trained to evaluate the diagnostic
performance of the BM3D. The performance of the classifiers for the chest x-rays was
examined using the BM3D filtering method and successful denoising results were observed
using the BM3D method.
Keyword(s): BM3D filtering, Medical Image Processing, Noise Reduction, Chest x-rays.
1. Introduction
In recent days, medical image processing has a very important place and many diseases
can be detected by using radiography. Noise is an unpredictable distortion that disturbs a
signal that causes random fluctuations of observed variables. Generally speaking, it is a very
important issue in any data collection and processing system, especially imaging techniques.
The chest X-Ray dataset is the largest public x-ray data set to date (5). Chest X-ray
examination is one of the most common and cost-effective medical imaging examinations.
However, the clinical diagnosis of chest x-ray can be difficult and can sometimes be believed
to be more difficult than the diagnosis by chest CT imaging. This database presents an
improved version of the data set used in recent studies (6). This data set was extracted from
the clinical PACS database at the Central Clinical Center of the National Institutes of Health
and consists of ~60% of all anterior chest x-rays in the hospital. Therefore, this data set is
expected to be significantly more representative for actual patient population distributions and
realistic clinical diagnostic difficulties than previous patient x-ray data sets.
Image processing consists of several steps, each of which has a major impact on
classification performance. In this study, a two-dimensional lung radiography image was
examined over gray color which is one-dimensional color space.
Noise reduction with BM3D has been used in many areas, one of which is salt and pepper
noise. One of the simplest noise models for digital images is salt and pepper noise. These
noises are eliminated by using a 3x3 median filter (7). The BM3D filter was used to eliminate
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
246
noise from noisy color filter array (CFA) data. A modification of the BM3D filter was
proposed for CFA data using cross-color correlations (8). In many important application
areas, images are often affected by non-Gaussian and / or color noise; in this case, white
Gaussian noise (AWGN) based techniques are far behind their promises. Since noise is the
primary cause of reduced image analysis capability in many application areas, BM3D noise
reduction has also been used in correlated noise reduction problems (9). One of the studies
using BM3D filtering is Magnetic Resonance (MR) images because the signal-to-noise ratio
(SNR) is generally low in these images. The quality of MRI images is generally reduced by
several artifacts and noise, which are appropriately modeled by Rician's noise. With the
BM3D approach, noise reduction performance is more useful than other methods (10).
There are many methods for noise reduction, BM3D (Block Matching 3D) method is used
in this study. Block matching is found using similar blocks, similarity measures for each
reference block by selecting a reference block of a certain size from the noisy image. The
matching blocks are stacked into a 3D sequence and exhibit a high degree of correlation due
to the similarity between them. Image processing consists of several steps which affect the
performance of the model at each step (1). Block-matching and 3D (BM3D) filtering is
performed in four steps: 1) finding image patches similar to the given image patch and
grouping them in a 3D block; 2) 3D linear transformation of the 3D block; 3) contraction of
conversion spectrum coefficients; 4) reversing 3D transformation respectively. Therefore, the
3D filter performs a limitation for each 2D image patch in the 3D block, parallelly (3). The
BM3D algorithm is used to reconstruct chest x-ray medical images.
The remaining of the paper structured as follows: Chest x-ray database specifications,
block matching algorithm and BM3D algorithm. The experimental hyperparameters
selections and the noise reduction efficiency of the BM3D on various noise functions on chest
x-ray images.
2. Material and Methods
a) Database
A chest x-ray image produces an image of the chest, lung, heart, airways and blood
vessels. Using a chest x-ray image, a trained radiologist can diagnose pneumonia,
pneumothorax, interstitial lung disease, heart failure, bone fracture, hiatal hernia and the like.
The main advantage of X-ray is its low cost and simplicity. The chest X-ray dataset includes
112,120 frontal view X-ray images of 30,805 patients with text-mined fourteen disease image
labels (each image having a multilateral) extracted from associated radiological reports using
natural language processing. Commonly used fourteen thoracic pathologies include
Atelectasis, Consolidation, Infiltration, Pneumothorax, Edema, Emphysema, Fibrosis,
Effusion, Pneumonia, Pleural-Thickening, Cardiomegaly, Nodule, Mass, and Hernia, an
extension of 8 common disease patterns listed in CV 2017 (5).
Chest X-ray is a non-invasive procedure that takes only a few minutes and typically occurs
within half an hour (Figure 1).
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
247
Figure 1- Several Chest X-Ray images
Noise reduction was performed on Chest X-Ray images and the performance of the BM3D
method was measured.
b) Block matching and 3-Dimensional
The similar degree of two images or image blocks is evaluated essentially from the
following two aspects: similarity based on pixels and structure similarity. In the block
matching process, a single-pixel gray value with the neighbor is compared to another
according to the gray distribution. However, when the image details or edges are strong,
measuring the similarity only with the grayscale distribution results in an incorrect grouping.
In this paper, similarity based on pixels is combined with structure similarity (11) to find the
most similar blocks.
We assume that fixed-size blocks in the actual image exhibit mutual correlation. As we can
see in this example, correlation matching is checked at the pixel level and the block closest to
the reference block is selected. Blocks similar to the reference block are arranged in 3D and
the correlation coefficients are taken and hard thresholding is performed. As a result of hard
thresholding, the picture is divided into two parts and the correlation coefficients are
estimated without noise (Figure- 2).
Because of the similarity between the matching blocks, there is a correlation along with the
size of the array in which the blocks are stacked. The only transformation that decodes a 3D
decor rarely represents the actual signal, so that noise can be effectively mitigated by
contracting the 3D conversion coefficients (2).
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
248
Figure 2- Scheme of the BM3D algorithms (3)
The PSNR block computes the peak signal-to-noise ratio, in decibels, between two images.
This ratio is used as a quality measurement between the original and a compressed image. The
higher the PSNR, the better the quality of the compressed, or reconstructed image (Eq. 1).
The mean-square error (MSE) and the peak signal-to-noise ratio (PSNR) are used to
compare image compression quality. The MSE represents the cumulative squared error
between the compressed and the original image, whereas PSNR represents a measure of the
peak error. The lower the value of MSE, the lower the error (Eq. 2).



 




3. Results and Discussion
In this study, chest X-Ray radiography images were compared with the noise that may
occur in gray-tone lung images and the results obtained by using the BM3D method were
compared. BM3D Sigma parameter was changed according to the noise intensity and a certain
Sigma interval was determined based on the best PSNR result (Table 1). After the
determination of the appropriate parameters as a result of the obtained data, the reduction of
the noise generated in the image by using BM3D of the chest radiography images was
successfully performed.
When the intensity of noise is increased by using the BM3D method, the quality of the
resulting image depends entirely on the sigma value (Figure 6, Figure 7).
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
249
Figure 3- (Sigma = 50)
Figure 4- (Sigma = 30)
Figure 5- (Sigma = 10)
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
250
Figure 6- (Sigma = 50)
Figure 7- (Sigma = 8)
Table 1. PSNR values depending on the Sigma parameter
In general, when the Sigma values are at a range of 20-40, the result of the PSNR value
difference is high. Successful results were obtained by eliminating the noise in this range by
using the BM3D method.
Sigma
Noisy PSNR (dB)
Denoised PSNR (dB)
50
14.129
32.52
40
16.068
33.59
30
18.56
34.81
20
22.08
36.25
10
28.109
38.48
International Conference on Artificial Intelligence towards Industry 4.0 (ICAII4.0) Nov 14-16, 2019
Iskenderun, Hatay / TURKEY
251
In further studies, this method can be used to eliminate device-induced (current-induced
measurement noise) noise generated in lung radiography images obtained from medical
devices.
Reference(s):
1. Dabov, K., Foi, A., Katkovnik, V., & Egiazarian, K. (2006, February). Image denoising with block-
matching and 3D filtering. In Image Processing: Algorithms and Systems, Neural Networks, and
Machine Learning (Vol. 6064, p. 606414). International Society for Optics and Photonics.
2. Hasan, M. M. 2003. Adaptive Edge-guided Block-matching and 3D filtering (BM3D) Image Denoising
Algorithm. Electronic Thesis and Dissertation Repository.
3. Lebrun, M. (2012). An Analysis and Implementation of the BM3D Image Denoising Method, Image
Processing On Line, 2 (2012), 175213. https://doi.org/10.5201/ipol.2012.l-bm3d.
4. Dai, L., Zhang, Y., & Li, Y. (2013). Bm3d image denoising algorithm with adaptive distance hard-
threshold. International Journal of Signal Processing, Image Processing, and Pattern Recognition, 6(6),
41-50.
5. Open-i: An open access biomedical search engine. https: //openi.nlm.nih.gov.
6. Wang, X., Peng, Y., Lu, L., Lu, Z., Bagheri, M., & Summers, R. M. (2017). Chest x-ray8: Hospital-
scale chest x-ray database and benchmarks on weakly-supervised classification and localization of
common thorax diseases. In Proceedings of the IEEE conference on computer vision and pattern
recognition (pp. 2097-2106).
7. Djurović, I. (2016). BM3D filter in salt-and-pepper noise removal. EURASIP Journal on Image and
Video Processing, 2016(1), 13.
8. Danielyan, A., Vehvilainen, M., Foi, A., Katkovnik, V., & Egiazarian, K. (2009, August). Cross-color
BM3D filtering of noisy raw data. In 2009 international workshop on local and non-local approximation
in image processing (pp. 125-129). IEEE.
9. Matrecano, M., Poggi, G., & Verdoliva, L. (2012). Improved BM3D for Correlated Noise Removal.
In VISAPP (1) (pp. 129-134).
10. Elahi, P., Beheshti, S., & Hashemi, M. (2014, May). BM3D mridenoising equipped with noise
invalidation technique. In 2014 IEEE International Conference on Acoustics, Speech and Signal
Processing (ICASSP) (pp. 6612-6616). IEEE.
11. Wang, M., Wang, H. J., Sun, G. Y., Guo, J. H., Wu, H. Q., Li, L., & Zhu, H. Y. (2013). Three-
dimensional Image Denoising Algorithm Based on Non-local Means. Infrared Technology, 35(4), 238-
241.
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
The chest X-ray is one of the most commonly accessible radiological examinations for screening and diagnosis of many lung diseases. A tremendous number of X-ray imaging studies accompanied by radiological reports are accumulated and stored in many modern hospitals' Picture Archiving and Communication Systems (PACS). On the other side, it is still an open question how this type of hospital-size knowledge database containing invaluable imaging informatics (i.e., loosely labeled) can be used to facilitate the data-hungry deep learning paradigms in building truly large-scale high precision computer-aided diagnosis (CAD) systems. In this paper, we present a new chest X-ray database, namely " ChestX-ray8 " , which comprises 108,948 frontal-view X-ray images of 32,717 unique patients with the text-mined eight disease image labels (where each image can have multi-labels), from the associated radiological reports using natural language processing. Importantly, we demonstrate that these commonly occurring thoracic diseases can be detected and even spatially-located via a unified weakly-supervised multi-label image classification and disease lo-calization framework, which is validated using our proposed dataset. Although the initial quantitative results are promising as reported, deep convolutional neural network based " reading chest X-rays " (i.e., recognizing and locating the common disease patterns trained with only image-level labels) remains a strenuous task for fully-automated high precision CAD systems.
Article
Full-text available
There is a significant recent advance in filtering of the salt-and-pepper noise for digital images. However, almost all recent schemes for filtering of this type of noise are not taking into an account the shape of objects (in particular edges) in images. We have applied the block-matching and 3D filtering (BM3D) scheme in order to refine the output of the decision-based/adaptive median techniques. Obtained results are excellent, surpassing current state-of-the-art for about 2 dB for both grayscale and color images.
Article
Full-text available
BM3D is a recent denoising method based on the fact that an image has a locally sparse representation in transform domain. This sparsity is enhanced by grouping similar 2D image patches into 3D groups. In this paper we propose an open-source implementation of the method. We discuss the choice of all parameter methods and confirm their actual optimality. The description of the method is rewritten with a new notation. We hope this new notation is more transparent than in the original paper. A final index gives nonetheless the correspondence between the new notation and the original notation.
Article
Full-text available
We present a novel approach to still image denoising based on effective filtering in 3D transform domain by combining sliding-window transform processing with block-matching. We process blocks within the image in a sliding manner and utilize the block-matching concept by searching for blocks which are similar to the currently processed one. The matched blocks are stacked together to form a 3D array and due to the similarity between them, the data in the array exhibit high level of correlation. We exploit this correlation by applying a 3D decorrelating unitary transform and effectively attenuate the noise by shrinkage of the transform coefficients. The subsequent inverse 3D transform yields estimates of all matched blocks. After repeating this procedure for all image blocks in sliding manner, the final estimate is computed as weighed average of all overlapping blockestimates. A fast and efficient algorithm implementing the proposed approach is developed. The experimental results show that the proposed method delivers state-of-art denoising performance, both in terms of objective criteria and visual quality.
Conference Paper
Full-text available
Most of the literature on denoising focuses on the additive-white-gaussian-noise (AWGN) model. However, in many important applicative fields, images are typically affected by non-Gaussian and/or colored noise, in which cases AWGN-based techniques fall much short of their promises. In this paper, we propose a new denoising technique for correlated noise based on the non-local approach. We start from the well-known BM3D algorithm, which can be considered to be the state of the art in AWGN denoising, and modify it in various critical steps in order to take into account the non-whiteness of noise. Experimental results on several test images corrupted by correlated noise confirm the potential of the proposed technique.
Article
Block-matching and 3D filtering (BM3D) denoising algorithm [1] proposed recently has a problem of computational burden especially for low noise level and a sharp performance drop for high noise level. To solve it, an improved version of BM3D is proposed. The solution combines the digital image characteristic with added noise pollution levels, and adaptively selects block-matching threshold in grouping stage. Experimental results demonstrate it outperforms not only in terms of objective criteria of PSNR and running time, but also in visual quality.
Conference Paper
Block-matching and 3D filtering (BM3D) has shown a great success in image denoising. In this work we propose a new denoising approach for Magnetic Resonance Imaging (MRI) based on a modified BM3D algorithm. BM3D is a combination of nonlocal approach, 3D wavelet shrinkage, and 3D Wiener filtering. We improve the wavelet thresholding stage of BM3D using Noise Invalidation Denoising (NIDe) technique. The new approach provides the optimum wavelet threshold automatically and adaptive to the statistical characteristics of the available data. This is an advantage over the existing denoising stage of BM3D that currently uses an adhoc thresholding value. Combining the proposed BM3D approach with Variance Stabilization Transformation (VST) enables the use of the proposed method for Magnetic Resonance (MR) Image denoising. In our Simulations, the proposed method outperforms the state of the art BM3D based MRI denoising methods in the sense of PSNR and SSIM for T1, T2 and PD weighted MR images.
Adaptive Edge-guided Block-matching and 3D filtering (BM3D) Image Denoising Algorithm
  • M M Hasan
Hasan, M. M. 2003. Adaptive Edge-guided Block-matching and 3D filtering (BM3D) Image Denoising Algorithm. Electronic Thesis and Dissertation Repository.
Threedimensional Image Denoising Algorithm Based on Non-local Means
  • M Wang
  • H J Wang
  • G Y Sun
  • J H Guo
  • H Q Wu
  • L Li
  • H Y Zhu
Wang, M., Wang, H. J., Sun, G. Y., Guo, J. H., Wu, H. Q., Li, L., & Zhu, H. Y. (2013). Threedimensional Image Denoising Algorithm Based on Non-local Means. Infrared Technology, 35(4), 238-241.