Method for reducing noise in medical image

Patent metadata

Field	Value
Publication number	US-9269128-B2
Application number	US-201214402800-A
Country	US
Kind code	B2
Filing date	May 23, 2012
Priority date	May 23, 2012
Publication date	Feb 23, 2016
Grant date	Feb 23, 2016

Abstract

There is provided a method of reducing noise in a medical image while maintaining structure characteristics within the medical image. The method of reducing noise in the medical image includes: a step of estimating the amount of noise at each pixel by using a look-up table that previously stores physical properties of an image capturing device; a step of calculating a structure direction and a signal coherence at each pixel; a step of performing anisotropic smoothing by using an anisotropic smoothing filter kernel that reflects the structure direction and the signal coherence; and a step of obtaining a structure direction and a signal coherence from an image on which the anisotropic smoothing is performed, obtaining an anisotropic smoothing filter kernel that reflects the obtained structure direction and the signal coherence, and performing filtering that reflects statistical properties of the estimated amount of noise.

First claim

I claim:

1. A method of reducing noise in a medical image, the method comprising:
a step of estimating the amount of noise in at each pixel position of an image by using a look-up table that is previously stored in order to suppress noise in the medical image;
a step of extracting a first signal structure direction and a first signal coherence at each pixel position;
a step of obtaining a first anisotropic smoothing filter kernel that reflects the extracted signal structure direction and the signal coherence;
a step of obtaining an anisotropic-smoothed intermediate image by performing filtering using the first anisotropic smoothing filter kernel; and
a step of extracting a second signal structure direction and a second signal coherence at each pixel of the anisotropic-smoothed intermediate image, and obtaining a second anisotropic filter kernel that reflects the second signal structure direction and the second signal coherence, and performing filtering on an original image by applying an additional weight that reflects the estimated amount of noise in the position of each pixel to surrounding pixels within a range of the filter kernel.

2. The method of reducing noise in the medical image of claim 1 ,
wherein the look-up table that previously stores physical properties of an image capturing device is a two-dimensional look-up table that has two terms including a radiation dose for an object during irradiation of X-ray and a signal average value at each pixel in an image, and provides a noise standard deviation at each pixel by inputting a radiation dose value which is provided from the image capturing device and a signal average value at each pixel in the image.

3. The method of reducing noise in the medical image of claim 2 ,
wherein when the radiation dose value is not provided from the image capturing device, the radiation dose value is generated by referring to the look-up table that previously stores information including an X-ray tube voltage, the product of a tube current and radiation time, a distance between an X-ray focal point and a detector, and a used X-ray energy filter.

4. The method of reducing noise in the medical image of claim 1 ,
wherein in the step of extracting the signal structure direction and the signal coherence at each pixel position, the signal structure direction is determined as a vertical direction of a gradient vector that is normalized at each pixel, and the signal coherence is determined by obtaining absolute values of gradient values of the normalized gradient vector and a vertical direction vector of the gradient vector and dividing a difference value between the absolute values of the gradient values by the sum of the absolute values.

5. The method of reducing noise in the medical image of claim 1 ,
wherein in the step of extracting the signal structure direction and the signal coherence at each pixel position, the signal structure direction is determined as a second eigenvector of the Hessian matrix for each pixel, and the signal coherence is determined by obtaining absolute values of two eigenvalues of the Hessian matrix and dividing a difference value between the absolute values of the two eigenvalues by the sum of the absolute values.

6. The method of reducing noise in the medical image of claim 4 ,
wherein in the step of extracting the signal structure direction and the signal coherence at each pixel position, the signal structure direction and the signal coherence are determined as the signal structure direction and the signal coherence according to claim 4 when a ratio of an absolute value of a gradient value obtained at each pixel to an absolute value of a first eigenvalue of the Hessian matrix is obtained and the obtained ratio is greater than a predetermined threshold value, and the signal structure direction and the signal coherence are determined as the signal structure direction and the signal coherence according to claim 5 when the ratio is smaller than the predetermined threshold value.

7. The method of reducing noise in the medical image of claim 1 ,
wherein in the step of obtaining the anisotropic smoothing filter kernel that reflects the signal structure direction and the signal coherence that are extracted at each pixel, a kernel that follows a two-dimensional anisotropic Gaussian function using the structure direction and the signal coherence is obtained, and lengths of a major axis and a minor axis of the anisotropic Gaussian function are respectively determined as a size of the kernel and a value obtained by multiplying the size of the kernel and the signal coherence, and a rotation angle of the anisotropic Gaussian function is determined as the structure direction.

8. The method of reducing noise in the medical image of claim 1 ,
wherein in the step of obtaining the anisotropic-smoothed intermediate image by performing filtering using the first anisotropic smoothing filter kernel, anisotropic smoothing filtering is repeatedly performed by the number of times that is determined by multiplying a predetermined proportional coefficient and the sum of the estimated amounts of noise in the respective pixels in the image in claim 1 .

9. The method of reducing noise in the medical image of claim 1 ,
wherein in the step of applying the additional weight that reflects the estimated amount of noise in the position of each pixel to the surrounding pixels within the range of the filter kernel, a ratio is obtained by dividing an absolute value of a difference value between a pixel value of the anisotropic-smoothed intermediate image at the central pixel of the kernel and the pixel value of the original image at surrounding pixel by the estimated amount of noise, and a value obtained by dividing the sum of the ratio and a predetermined constant β by 1 is applied as a weight for the surrounding pixel.

10. The method of reducing noise in the medical image of claim 1 ,
wherein in the step of applying the additional weight that reflects the estimated amount of noise at each pixel position to the surrounding pixels within the range of the filter kernel, a value of the Gaussian function in which the estimated amount of noise at the central pixel of the kernel is set as a standard deviation, the pixel value of the anisotropic-smoothed intermediate image is set as an average value, and the pixel value of the original image at the surrounding pixel is set as a variable is obtained, and the obtained value is set as the weight for the surrounding pixel.

11. The method of reducing noise in the medical image of claim 1 ,
wherein in the step of performing filtering on the original image by applying the additional weight that reflects the estimated amount of noise at each pixel position to the surrounding pixels within the range of the filter kernel and outputting the filtered image as a final output, the filtering is repeatedly performed by the number of times determined by multiplying a predetermined proportional coefficient and the sum of the estimated amounts of noise for the respective pixels in the image, which are obtained from the original image, and a value obtained by multiplying a predetermined reduction coefficient and the estimated amount of noise of the preceding filtering is used as the estimated amount of noise at each pixel whenever the filtering is repeated.

12. The method of reducing noise in a medical image of claim 1 ,
wherein in the step of performing filtering on the original image by applying the additional weight that reflects the estimated amount of noise at each pixel position to the surrounding pixels in the range of the filter kernel and outputting the filtered image as a final output, a small amount of n

Assignees

• Kim Jong Hyo
• Snu R&Db Foundation

Inventors

• Kim Jong Hyo

Classifications

• A61B6/5258

  involving detection or reduction of artifacts or noise · CPC title

• G06T5/20

  using local operators · CPC title

• G06T2207/10072

  Tomographic images · CPC title

• G06T2207/10116

  X-ray image · CPC title

• G06T2207/20182

  Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering · CPC title