# US 2014/0153842 A1 · Method for reducing noise in a sequence of fluoroscopic images by temporal and spatial filtering

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2014/0153842 A1
- Document ID: US20140153842A1
- Status: Published application
- Kind: A1
- Application number: 14/234,292
- Filed: 2012-07-20
- Published: 2014-06-05
- Earliest claimed: 2011-07-22
- Assignees: Thales
- Inventors: Jérémie Pescatore; Carole Amiot
- Web page: https://patentgrep.com/patent/US20140153842A1

## CPC

- G06T 5/70: Denoising; Smoothing
- G06T 2207/10121: Fluoroscopy
- G06T 2207/20012: Locally adaptive
- G06T 2207/20016: Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
- G06T 2207/20182: Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
- G06T 2207/30004: Biomedical image processing
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

In the field of medical imaging using fluoroscopy, a method for reducing noise in a sequence of fluoroscopic images acquired by an X-ray detector comprises for each image (xn) of the sequence: applying temporal filtering to the image (xn) at instant n, the temporal filtering comprising fluoroscopic noise reduction processing; applying spatial filtering to the image (yn) at instant n, the spatial filtering comprising: transforming the image (yn) acquired at the instant n from the spatial domain into the curvelet domain using a curvelet transform, each transformed image being represented by a set of coefficients, thresholding the coefficients of the image (yn) using a thresholding function, the thresholding function cancelling the coefficients below a third predetermined threshold, and preserving or adjusting the coefficients above the third predetermined threshold, transforming the image (zn) whose coefficients have been thresholded from the curvelet domain into the spatial domain using an inverse curvelet transform.

## Claims

### Claim 1

A method for reducing noise in a sequence of fluoroscopic images acquired by an X-ray detector, each image (xn) being formed in the spatial domain by a matrix of pixels, each pixel having a value representative of a signal level, the method comprising the following successive steps for each image (xn):
applying temporal filtering to the image (xn) acquired at an instant n, the application of said temporal filtering comprising the following substeps for each pixel of the image:
determining a difference between the value of the pixel considered in the image (xn) acquired at the instant n and the value of the corresponding pixel in the image (yn-1) acquired at an instant n−1 to which the temporal filtering (22) has been applied,
if the difference is below a first predetermined threshold S1, correcting the value of the corresponding pixel with a first set of parameters,
if the difference is above a second predetermined threshold S2, correcting the value of the corresponding pixel with a second set of parameters,
applying spatial filtering to the image (yn) acquired at the instant n, the application of said spatial filtering comprising the following substeps:
transforming the image (yn) acquired at the instant n from the spatial domain into the curvelet domain using a curvelet transform, each image in the curvelet domain being represented by a set of coefficients,
thresholding the coefficients of the image (yn) using a thresholding function, the thresholding function cancelling the coefficients below a third predetermined threshold, and preserving or adjusting the coefficients above the third predetermined threshold,
transforming the image (yn) whose coefficients have been thresholded from the curvelet domain into the spatial domain using an inverse curvelet transform.

### Claim 2 (depends on 1)

The method as claimed in claim 1, wherein the step of transforming the image (yn) acquired at the instant n into the curvelet domain uses a curvelet transform with 6 scales.

### Claim 3 (depends on 1)

The method as claimed in claim 1, wherein the step of transforming the image (yn) acquired at the instant n into the curvelet domain uses a curvelet transform with 9 scales.

### Claim 4 (depends on 1)

The method as claimed in claim 1 one of the preceding claims, wherein the step of transforming the image (yn) acquired at the instant n into the curvelet domain uses a curvelet transform comprising 16 orientations.

### Claim 5 (depends on 1)

The method as claimed in claim 1 one of the preceding claims, wherein the step of transforming the image (yn) acquired at the instant n into the curvelet domain uses a discrete curvelet transform via USFFT.

### Claim 6 (depends on 1, 2, 3, 4)

The method as claimed in claim 1 one of claims 1 to 4, wherein the step of transforming the image (yn) acquired at the instant n into the curvelet domain uses a discrete curvelet transform via wrapping.

### Claim 7 (depends on 1)

The method as claimed in claim 1 one of the preceding claims, wherein the thresholding function is a hard thresholding function, wherein the coefficients are cancelled if they are below the third predetermined threshold, and preserved otherwise.

### Claim 8 (depends on 7)

The method as claimed in claim 7, wherein the third predetermined threshold TS is determined by the following successive steps:
creating an image Y of the same dimensions as the image (yn) acquired at the instant n and of uniform value 1 over all pixels,
transforming the image Y using a Fourier transform,
normalizing the amplitude F of the peak in the Fourier transform of the image Y using the relationship:

F

=
6
5

F
M
×
P
where M and P denote the dimensions of the image Y,
transforming the Fourier transform of the image Y from the spatial domain into the curvelet domain using a curvelet transform, each image in the curvelet domain being represented by a set of coefficients cj,l,kY,
determining a norm ĉj,lY for the coefficients cj,l,kY using the relationship:
c
^
j
,
l
Y
=
∑
l


c
j
,
l
,
k
Y

2
M
×
P
where M and P denote the dimensions of the image Y and l denotes the orientations of the curvelets,
determining the predetermined threshold TS using the relationship:
TS=r·ĉj,lY·σ
where σ is the standard deviation of the noise in the image (yn) acquired at the instant n and to which the temporal filtering has been applied, and where the factor r takes the value 0 for the coarsest scale of the curvelets, the value 4 for the finest scale and the value 1 for the other scales.

### Claim 9 (depends on 1)

The method as claimed in claim 1, wherein the thresholding function applies soft thresholding to the coefficients whose modulus lies between the third predetermined threshold T1 and a fourth predetermined threshold T2, hard thresholding to the coefficients whose modulus is above a fifth predetermined threshold T3 and transition thresholding between the soft thresholding and the hard thresholding to the coefficients whose modulus lies between the fourth predetermined threshold T2 and the fifth predetermined threshold T3, the predetermined thresholds T1, T2 and T3 being such that T1<T2<T3.

### Claim 10 (depends on 9)

The method as claimed in claim 9, wherein the transition thresholding is defined by a sigmoid defined as follows:
f

(
c
n
)
=
1
1
+
exp

(
-
λ

(
c
n
-
m
)
)
where cn are the coefficients that need to be thresholded, and where the parameters m and λ are determined as follows:
m
=
ln

(
T
1
T
2
-
T
1
)

T
3
-
T
2
·
ln

(
0.01
)
ln

(
T
1
T
2
-
T
1
)
-
ln

(
0.01
)
λ
=
-
ln

(
0.01
)
T
3
-
m
.

### Claim 11 (depends on 10)

The method as claimed in claim 10, wherein the step of applying spatial filtering furthermore comprises, prior to the thresholding substep:
a substep of correcting the coefficients as a function of their parent coefficient, of their neighbor coefficients and/or of their cousin coefficients, the parent coefficient of a given coefficient cnD(j,l,k) being the coefficient cnD(j−1,l,k) having the same orientation and the same position as the given coefficient on the previous coarsest scale j−1, its neighbor coefficients being the coefficients cnD(j,l,k′) corresponding to the contiguous positions for the position of the given coefficient and having the same scale and the same orientation, the cousin coefficients being the coefficients having the same scale and the same position as the given coefficient.

### Claim 12 (depends on 11)

The method as claimed in claim 11, wherein a first range of values is defined for the values below the third predetermined threshold T1, a second range of values is defined between the third predetermined threshold T1 and the fourth predetermined threshold T2, a third range of values is defined between the fourth predetermined threshold T2 and the fifth predetermined threshold T3 and a fourth range of values is defined for the values above the fifth predetermined threshold T3, the substep of correcting the coefficients applying the following corrections:
when the modulus of the coefficient in question cnD(j,l,k) is situated either in the same range as the modulus of its parent coefficient cnD(j−1,l,k) or in the same range as the average of the moduli of its neighbor coefficients cnD(j,l,k′), the modulus of the corrected coefficient {tilde over (c)}nD(j,l,k) takes the value of the modulus of the coefficient cnD(j,l,k),
when the modulus of the parent coefficient cnD(j−1,l,k) and the average of the moduli of the neighbor coefficients cnD(j,l,k′) are situated in the same range, different from that of the modulus of the coefficient in question cnD(j,l,k), the modulus of the corrected coefficient {tilde over (c)}nD(j,l,k) takes the minimum value from the range in which the modulus of the parent coefficient and the average of the moduli of the neighbor coefficients are situated if the modulus of the coefficient in question cnD(j,l,k) is situated in a range with lower values, and the maximum value from this range otherwise,
when the modulus of the coefficient in question cnD(j,l,k), the modulus of its parent coefficient cnD(j−1,l,k) and the average of the moduli of its neighbor coefficients cnD(j,l,k′) are situated in three separate ranges, the modulus of the corrected coefficient {tilde over (c)}nD(j,l,k) takes the minimum value from the range in which the average of the moduli of the neighbor coefficients is situated if the modulus of the coefficient in question cnD(j,l,k) is situated in a range with lower values, and the maximum value from this range otherwise.

### Claim 13 (depends on 1)

The method as claimed in claim 1, wherein the step of applying the temporal filtering to the image acquired (xn) at the instant n comprises, following the substep of determining the difference between the value of the pixel in question in the image (xn) acquired at the instant n and the value of the corresponding pixel in the image (yn-1) acquired at the instant n−1 to which the temporal filtering has been applied, the following additional substep:
if the difference lies between the first predetermined threshold S1 and the second predetermined threshold S2, correcting the value of the corresponding pixel with a third set of parameters.
