FIELD OF INVENTION
In recent years significant technological advancements and progress in image 5 processing have been achieved, however, still a number of factors in the visual quality of images, hinder the automated analysis, and disease evaluation. These include imperfections of image acquisition instrumentations, natural phenomena, transmission errors, and coding artifacts, which all degrade the quality of the image in the form of induced noise. Ultrasound imaging is a powerful non-10 invasive diagnostic tool in medicine, but it is degraded by a form of multiplicative speckle noise, which makes visual observation difficult. Speckle is mainly found in echogenic areas of the image in the form of a granular appearance that affects the texture of the image, which may carry important information about the shape of tissues and organs, also in Synthetic Aperture Radar (SAR) images. Texture 15 and morphology may provide additional quantitative information of the area under investigation, which may complement the human evaluation and provide additional diagnostic details. It is therefore of interest for the research community to investigate and apply new image despeckle filtering technique that can increase the visual perception evaluation and further automate image analysis, thus 20 improving the final diagnosis. This technique is specifically designed for medical image processing applications. It should be however noted, that it is not always desirable to remove speckle noise from the images as it can be considered as a natural tissue effect which may provide additional information, especially in the areas of strain imaging and speckle tracking, and methods of ultrasound tissue 25 characterization. This framework preprocesses the ultrasound images for further analysis and assessment by the medical experts in cardiovascular imaging-based diagnosis. The present work uses theories like Bayesian and Bivariate thresholding, method noise thresholding, DWT, homomorphic and non-
3
homomorphic thresholding and invents a new despeckling framework specifically for ultrasound images.
BACKGROUND & PRIOR ART
The ultrasound imaging (sonography) uses high-frequency sound waves to view 5 inside the body. Because ultrasound images are captured in real-time, they can also show the movement of the body's internal organs as well as blood flowing through the blood vessels. Unlike X-ray imaging, there is no ionizing radiation exposure associated with ultrasound imaging. In an ultrasound exam, a transducer (probe) is placed directly on the skin or inside a body opening. A thin layer of gel 10 is applied to the skin so that the ultrasound waves are transmitted from the transducer through the gel into the body. The ultrasound image is produced based on the reflection of the waves off of the body structures. The strength (amplitude) of the sound signal and the time it takes for the wave to travel through the body provide the information necessary to produce an image. The automatic analysis of 15 medical images is somewhat controversial. The advantages are clear: computer analysis is cheaper and results will not vary for a given image. However, medical images are highly variable. Algorithms are inherently tuned to look for specific features and are prone to fail spectacularly when given unexpected input data. The framework is invented to assist trained technologists. It would attempt to detect 20 candidate features but would require human validation of all questionable decisions.
There are many despeckling techniques and filters available in spatial as well as frequency domain to handle speckled SAR images by preserving the major information in the image like edge, texture, homogeneous and non-homogeneous 25 regions of the image. Some standard and conventional method for despeckling purpose in the field of satellite imagery are, frost filter, Kuan filter, Kuwahara filter, lee filter, mean filter and median filter. Other developed schemes under homomorphic and non-homomorphic filtering are effective and adaptive. Some of the major filters discussed here. Linear despeckle filter (DsFlsmv): These filters 30
4
utilize first order statistics such as the variance and the mean of a pixel neighborhood and may be described by a multiplicative noise model. It increases the overall image quality of the image by enhancing edges. Wiener despeckle filter (DsFwiener): The filter DsFwiener uses a pixel-wise adaptive Wiener method. It is also not well suited for statistical analysis as well as for improving 5 the classification accuracy. Linear despeckle filter (DsFlsminsc): The DsFlsminsc is a 2D filter operating in a 5 × 5 pixel neighborhood by searching for the most homogenous neighborhood area around each pixel using a 3 × 3 subset window. The middle pixel of the 5 × 5 neighborhood is substituted with the average gray level of the 3 × 3 mask with the smallest speckle index. The DsFlsminsc filter is 10 very well suited for improving the outcome of the statistical analysis and the classification accuracy, but it does not well preserve edges and the overall image quality. Nonlinear despeckle filter (DsFkuwahara): The DsFkuwahara is a 1D filter operating in a 5 × 5 pixel neighborhood searching for the most homogenous neighborhood area around each pixel. The middle pixel of the 1 × 5 neighborhood 15 is then substituted by the median gray level of the 1 × 5 mask. The DsFkuwahara filter can be used to improve the classification accuracy of different organs and tissues and to enhance edges, thus also improving the optical perception evaluation. Geometric despeckle filter (DsFgf): The concept in the geometric filtering is that speckle appears in the image as narrow walls and valleys. The 20 geometric filter, through iterative repetition, gradually tears down the narrow walls (bright edges) and fills up the narrow valleys (dark edges), thus smearing the weak edges that need to be preserved. Median (DsFmedian), hybrid median (DsFhmedian) despeckle filters: The filter DsFmedian is a median filter applied over windows of size 5 × 5. This is an extension of the filter DsFhmedian, which 25 was introduced in and later used in and it computes the median of the outputs generated by median filtering with three different windows (cross shape window, x-shape window, and normal window). The filter DsFhmedian preserves the edges and increases the optical perception evaluation. It can thus be used to preserve and enhance edges of various organs in ultrasound images. Anisotropic diffusion filter 30 (DsFad): Perona and Malik introduced the function, that smoothes the original
5
image while trying to preserve brightness discontinuities. DsFad filter usually smooths the image extensively but it was observed that it can also be used to improve the quality of video encoding as well reducing the bandwidth required for transmitting the filtered image over a 3G wireless network. A speckle reducing anisotropic diffusion filter (DsFsrad): Speckle reducing anisotropic diffusion is 5 described is based on setting the diffusion coefficient in the diffusion equation using the local frame gradient and the frame Laplacian. The DsFsrad filter uses two seemingly different methods, namely the Lee and the Frost diffusion filters. the DsFsrad filter may be used to improve the overall image quality. It was furthermore observed to improve the quality of video encoding as well reducing 10 the bandwidth required for transmitting the filtered ultrasound image over a 3G wireless network.
OBJECTIVE OF THE INVENTION
The principal objectives of the present invention are: 15
 Removal of structured speckle noise.
 Feature extraction of Ultrasound images such as Preservation of edges, texture, homogeneous and non-homogeneous regions.
 It is used for imaging soft tissues in organs like spleen, uterus, liver, heart, kidney, brain etc. 20
 It obscures and blurs image detail significantly, degrades the image quality and hence decreases the difficulty for the observer to discriminate fine detail of the image during the diagnostic examination.
 It also reduces the speed and accuracy of ultrasound images processing tasks such as segmentation and registration. 25
6
BRIEF DESCRIPTION OF DRAWINGS
The present invention eliminates the granular structured speckle noise (1) from a speckled ultrasound image (1) using homomorphic (1, 26, 27) and non-homomorphic (1, 2) approach. The parallel smoothing processes are used to despeckle the speckled ultrasound image (1) which includes the transform domain 5 (2, 27) for easy image analysis and later the results are fused using the hybrid organization of method noise thresholding (7, 17, 35) and averaging operation (22, 25). The main cause of noise introduction is most of the time during image acquisition. The speckled image (1) is decomposed using wavelet transform i.e. db2 based 2D-DWT (2, 27), it transforms the image into two components i.e. 10 approximate (12, 20) and detailed parts (3, 28). One step of the parallel process uses homomorphic DWT based filtering (1, 26, 27) and another step also uses non-homomorphic DWT based filtering (1, 2). All the components of both DWT transform (2, 27) are processed using wavelet thresholding (9, 6, 15, 23, 29) followed by method noise thresholding (7, 17, 35) and later the approximate parts 15 (12, 20) are fused using averaging operation (22, 25).
DETAIL DESCRIPTION
The invented framework is designed for reducing the speckle noise present in ultrasound images using parallel smoothing based hybrid approach based on 20 method noise using the concept of image fusion. The two different despeckling mechanism runs in parallel using wavelet thresholding and method noise thresholding. Later the output denoised images are fused using averaging.
One of the parallel despeckling framework uses non-homomorphic wavelet thresholding with method noise. Here DWT (2) is applied to the input speckled 25 ultrasound image (1). The image (1) is decomposed into two components: approximate D1 (3) and detail part A1 (12). Now log transform L (4, 13) is applied to both components (3, 12) for multiplicative nature transformation to the additive. D3 (5) and A3 (14) are the updated components. The wavelet thresholding using bivariate shrinkage rule (6) is applied on the D3 (5) and D5 30
7
(11) is obtained as output. This operation is followed by method noise thresholding. The D3-D5 operation (7) is applied. The resultant D6 (7) is thresholded using Bayesian shrinkage rule (9) and D7 (10) is obtained which is added to D3 (5). Its output is D8 (8) which is enhanced detailed part. A3 (14) is the log-transformed approximate part. The wavelet thresholding using bivariate 5 shrinkage rule (15) is applied on A3 (14) and A5 (16) is obtained.
In parallel, other smoothing technique uses homomorphic wavelet thresholding. The log transform (26) is applied to the input image (1). Later DWT (27) is applied and it decomposes the image into approximate A2 (20) and detail part D2 (28). The A2 (20) and A3 (14) are fused (21) using averaging operation and A4 10 (22) is obtained from it. A4 (22) is thresholded using Bayesian shrinkage rule (23) and hence A6 (24) is obtained. Method noise thresholding is applied on A5 (16). A5-A6 operation (17) is applied and the output A7 (17) is thresholded using Bayesian shrinkage rule (18) which yields A8 (19) as enhanced approximate part.
Now an average of A5 (16) and A6 (24) is taken which is later added to A8 (19) 15 and it yields A9 (25) which is most enhanced approximate component. Wavelet thresholding using bivariate shrinkage rule (29) is applied on D2 (28) and it yields D4 (30) and later method noise thresholding is applied. The operation D2-D4 is applied and D9 (35) is obtained and it is thresholded using Bayesian shrinkage rule (36). It yields D10 (37) and which is added to D4 (30) and D11 (31) is 20 obtained. D11 (31) is then added to D8 (8) and D12 (32) is obtained which is the most enhanced detailed component. Lastly, the approximate A9 (25) and detail D12 (32) is directed to IDWT (33) and later to exponential operation (34). This gives the final despeckled output image (38).
25
Sign: -
Name: - DIWAKAR, MANOJ
Address: DEPARTMENT OF CSE, UIT, UTTARANCHAL UNIVERSITY, DEHRADUN, 248007, INDIA
Sign: -
Name: - SHREE, RAJ
Address: - B-1/1-4, DEPARTMENT OF INFORMATION TECHNOLOGY, BABASAHEB BHIMRAO AMBEDKAR UNIVERSITY, VIDYA VIHAR, LUCKNOW (226025), UP, INDIA
Sign: -
Name: - SINGH, PRABHISHEK
Address: DEPARTMENT OF INFORMATION TECHNOLOGY, BABASAHEB BHIMRAO AMBEDKAR UNIVERSITY, VIDYA VIHAR, LUCKNOW (226025), UP, INDIA
8
Abstract
A DESPECKLING FRAMEWORK FOR ULTRASOUND IMAGES USING HYBRID APPROACH
The presented invention is based on hybrid approach including homomorphic and non-homomorphic filtering for despeckling the ultrasound images using fusion. 5 The framework provides the solution to the problem of the presence of speckle noise in the ultrasound images by parallel processing the same speckled images at the same time using DWT based wavelet decomposition and reducing the speckle noise by thresholding and then fusing them into an enhanced despeckled SAR image. The two parallel processes include homomorphic and non-homomorphic 10 filtering. The speckle reduction framework uses db2 based 2D-DWT to n level of decomposition. The wavelet decomposition is experimented at the 3-to-7 level and fixed at that level where the results are obtained best. Here one single speckled ultrasound image is processed in parallel in the wavelet domain. The first thresholding is performed using Bayesian thresholding method and another 15 thresholding is performed using Bivariate thresholding method while the low-frequency details are fused using average operation based on method noise thresholding. The updated low and high-frequency details are directed to IDWT for enhanced results. The developed framework is validated on simulated as well as on real speckled ultrasound images and the results are verified using structural 20 similarity index, peak signal to noise ratio, an equivalent number of looks and noise variance estimation.

Claim(s)
The despeckling Ultrasound image method comprising:
1. Provides a unique and highly adaptive method of despeckling the speckled Ultrasound images by using thresholding concept in two different 5 approaches. This hybrid method is unique and highly capable of despeckling the Ultrasound image.
2. Uses the concept of log transform in two ways. In parallel, log transform is applied indifferently.
3. Uses 2D-DWT based thresholding where the bivariate concept has been 10 used using db2, in both log environment data.
4. uses the concept of “method noise thresholding” where Bayes concept has been used using db2 based 2D-DWT.
5. provides a unique and intelligent way of filtering the detailed part and approximation part of the speckled Ultrasound image by applying wavelet 15 thresholding followed by method noise thresholding on the detailed and approximation part of the image in both log environment.
6. provides a unique and hybrid combination by applying fusion of both de-speckled log environment approximation and detail parts for significant speckle reduction and edge preservation.