Denoising method and system for preserving clinically significant structures in reconstructed images using adaptively weighted anisotropic diffusion filter
Embodiments and processes of computer tomography perform tasks associated with denoising a reconstructed image using an anisotropic diffusion filter and adaptively weighting an iterative instance of the diffused image based upon the product of a weight value and a difference between the iterative instance of the diffused image and the original image. In general, the adaptive weighting is a negative feedback in the iterative steps.
1. A medical imaging apparatus comprising:
a processing circuitry configured to execute anisotropic diffusion for medical image data comprising a plurality of signal values, each signal value corresponding to coordinates in a space, the processing circuitry also configured to generate a plurality of weights, wherein the weight for each signal value depends on an edge value corresponding to the coordinates of the respective signal value,
wherein the processing circuitry repeatedly executes an anisotropic diffusion filter for the medical image data,
the anisotropic diffusion filter includes a combination of an anisotropic diffusion term and an inhomogeneous term,
the anisotropic diffusion term applies anisotropic diffusion to a current image, and
the inhomogeneous term applies the weight to a difference between the current image and an initial image, wherein the anisotropic diffusion filter is given by
u ( t n+1 )= u ( t n )+( t n+1 )−( t n )[∇⋅( D∇u ( t n )− u 0 )]
where u(t n+1 ) is a signal value at the (n+1)th iteration, u(t n ) is a signal value at the nth iteration, u 0 is an initial signal value before application of the anisotropic diffusion filter, D is a diffusion coefficient and W is a weight corresponding to the initial signal value.
2. The apparatus of claim 1 , wherein the medical image data comprises volume data, and the signal value comprises a voxel value.
3. The apparatus of claim 1 , wherein the inhomogeneous term in the anisotropic diffusion filter is a term that multiplies a difference between reference medical image data of the initial image and the medical image data smoothed by the anisotropic diffusion in a current instance in iteration by the weight for the signal value, the reference medical image data comprises medical image data before application of the anisotropic diffusion.
4. The apparatus of claim 1 , wherein the processing circuitry is further configured to generate the weights by using Laplacian edge detection.
5. The apparatus of claim 1 , wherein the processing circuitry is further configured to generate the weights by using Sobel edge detection.
6. The apparatus of claim 1 , wherein the processing circuitry is further configured to generate the weights by using curvature edge detection.
7. The apparatus of claim 1 , wherein the weight is invariant in the anisotropic diffusion repeated by a predetermined number of iterations.
8. The apparatus of claim 1 , wherein the medical image data comprises medical image data smoothed in advance before the anisotropic diffusion.
9. The apparatus of claim 1 , wherein the inhomogeneous term in the anisotropic diffusion filter is a term that multiplies a difference between reference medical image data of the initial image and the medical image data smoothed by the anisotropic diffusion in a current instance in iteration by the weight for the signal value, the reference medical image data comprises the medical image data sharpened based upon medical image data before the anisotropic diffusion and a predetermined parameter.
10. The apparatus of claim 9 , wherein the sharpened medical image data is given by
S
(
u
0
)
=
(
w
2
w
-
1
u
0
-
1
-
w
2
w
-
1
u
)
where S(u 0 ) is a signal value of the sharpened medical image data, u 0 is an initial signal value before application of the anisotropic diffusion filter, u is a signal value at the nth iteration, and w is the predetermined parameter.
11. The apparatus of claim 1 , wherein the processing circuitry is further configured to store a predetermined time scale,
the anisotropic diffusion is executed at a specific time interval, and
a predetermined number is a number obtained by multiplying the time scale by the specific time interval.
12. The apparatus of claim 11 , wherein the specific time interval is smaller than a reciprocal of twice the number of dimensions of the medical image data.
13. A medical image processing method comprising:
providing medical image data comprising a plurality of signal values, each signal value corresponding to coordinates in a space;
generating a plurality of weights, wherein the weight for each signal value depends on an edge value corresponding to the coordinates of the respective signal value; and
repeatedly executing an anisotropic diffusion filter for the medical image data,
wherein the anisotropic diffusion filter includes a combination of an anisotropic diffusion term and an inhomogeneous term,
the anisotropic diffusion term applies anisotropic diffusion to a current image, and
the inhomogeneous term applies the weight to a difference between the current image and an initial image, wherein the anisotropic diffusion filter is given by
u ( t n+1 )= u ( t n )+( t n+1 )−( t n )[∇⋅( D∇u ( t n )− u 0 )]
where u(tn+1) is a signal value at the (n+1)th iteration, u(tn) is a signal value at the nth iteration, u0 is an initial signal value before application of the anisotropic diffusion filter, D is a diffusion coefficient and W is a weight corresponding to the initial signal value.