WO2009065441A1

WO2009065441A1 - Method and arrangement in fluoroscopy and ultrasound systems

Info

Publication number: WO2009065441A1
Application number: PCT/EP2007/062658
Authority: WO
Inventors: Hagen Spies; Tomas Loock
Original assignee: Sapheneia Commercial Products Ab
Priority date: 2007-11-21
Filing date: 2007-11-21
Publication date: 2009-05-28

Abstract

The present invention relates to a method and an arrangement enhancing image quality of ultrasound and fluoroscopic image sequences. A noise reduction scheme is provided, where the original image data is filtered by a 3D-filter (F) consisting of spatial FIR and temporal IIR parts.

Description

TITLE

Method and arrangement in fluoroscopy and ultrasound systems

TECHNICAL FIELD

[0001] The present invention relates to a method and an arrangement in fluoroscopy and ultrasound imaging systems, and particularly connected to processing images of fluoroscopic and ultrasound sequences as well as a method for such processing. The invention further relates to a computer-readable medium containing computer program handling of fluoroscopic and ultrasound images.

BACKGROUND OF THE INVENTION [0002] Fluoroscopy is a diagnostic procedure that uses special x-ray equipment to obtain live pictures of a body. The computer displays these pictures as detailed images of objects, such as organs, bones and other tissues. Fluoroscopy may e.g. be used to detect or confirm the presence of a stenosis inside the body of a person. During a fluoroscopic exam, the person may be treated with a contrast agent that makes certain structures more visible. Fluoroscopy may also be used to visualize the position and action of various instruments used for treatment or diagnose.

[0003] The fluoroscopy system converts x-ray attenuation measurements into intensity values that indicate the tissue density along a particular ray of sight. A detailed description of how a fluoroscopy system generates image data is not included here. This is, however, a well-known technique.

[0004] One of the main difficulties in the interpretation of fluoroscopy images is the presence of noise. In order to achieve a good image quality, it is necessary to expose the person being scanned to high radiation doses and a major hinder for the use of lower radiation doses is the accompanied increase in noise. To reduce the effect of noise, several image processing filters have been proposed. One prior art approach is found in document US 6 647 093, which discloses a method for detail enhancement of X-ray images based on a pattern matching strategy. Another method, disclosed in US 6 763 129, provides spatial noise reduction by filtering along detected directions. [0005] Ultrasound is another diagnostic procedure that uses high frequency sound waves to generate images of the structures within the body. Here the reflections of the sound waves are recorded and converted into dynamic image sequences. Due to randomly distributed scatter the resulting images contain a substantial amount of speckle noise. Many methods for speckle reduction have been reported, e.g. F. Zhang et al, Nonlinear Diffusion in

Laplacian Pyramid Domain for Ultrasonic Speckle Reduction, IEEE Transaction on Medical Imaging, 26(2):200-21 1 , February 2007, and the presented method provides a novel such method that is particularly suitable for real time speckle reduction.

[0006] Common to both described diagnostic methods is the creation of an image sequence that can contain a considerable amount of noise. As both techniques visualize dynamic processes within the body it is necessary to exploit both the spatial and the temporal coherence of structures when attempting to reduce the noise while keeping these structures.

[0007] Drawbacks of the methods described above are that they do not utilize the advantages of spatio-temporal image processing to enhance image quality by reducing noise in the images.

[0008] Filters for image processing comprise finite impulse response (FIR) and infinite impulse response (NR) filters. The FIR filter consists of a fixed number of filter coefficients without feedback. The NR filter on the other hand includes a feedback loop in the filtering. Such filters are also called recursive filters. In the context of real time imaging system such recursive filters have the large advantage that they do not introduce a time delay when applied to image sequences.

[0009] Recursive (time-lag) filters are commonly used for noise suppression on medical image sequences. Document US 5 872 602 describes a recursive filter for fluoroscopic imaging that is controlled by an image pixel brightness and the amount of motion detected. Another document, US 5 296 937, describes another method where the coefficients of a recursive filter are controlled by the difference in the image pixel values of the current frame and the last output image.

[0010] US 5 809 105 describes a noise reduction filter for x-ray image sequences that separates structures from the background by the use of edge detection and then applies noise reduction filtering only to the background. This method does not take temporal information into account. [0011] A bilateral filter, [V. Aurich and J. Weule, Non-linear Gaussian filters performing edge preserving diffusion, Proceedings of the DAGM Symposium, pages 538-545, Springer 1995; S. M. Smith and J. M. Brady, SUSAN - a new approach to low level image processing, International Journal of Computer Vision, 23(1 ):45-78, May 1997; and C. Tomasi and R. Manduchi, Bilateral filtering for gray and color images, Proceedings of the International Conference on Computer Vision, pages 839-846, IEEE 1998] is a popular nonlinear filter used to smooth images. The filter output consists of a weighted average of a neighbourhood of pixels, where the weights are determined by Gaussian functions controlled by both the spatial distance of each pixel to the centre of the neighbourhood and the intensity difference between the values at these pixels and the intensity value at the centre of the neighbourhood. A major drawback of this filter is its high computational complexity. To speed up this filter, a separable implementation for video pre-processing has been proposed by T.Q. Pham and L. van Vliet, Separable bilateral filtering for fast video processing, Proceedings International Conference on Multimedia and Expo. IEEE 2005.

[0012] The use of several frequency bands for medical image processing is a widely known technique that often uses the Laplacian pyramid, described by P.J. Burt and E. H. Adelson, The Laplacian pyramid as a compact image code, IEEE Transactions Communications, 31 (4):532-540, April 1983. A recent comparison of some such techniques for radiography can be found in S. Dippel, M. Stahl, R. Wiemker and T. Blaffert, Multiscale Contrast Enhancement Radiographies: Laplacian Pyramid Versus Fast Wavelet Transform, IEEE Transaction on Medical Imaging, 21 (4):343-353, April 2002.

SUMMARY OF THE INVENTION

[0013] The aspects of the present invention relates to a method and an arrangement in fluoroscopy and/or ultrasound imaging systems and particularly to an arrangement allowing an enhancing image quality of fluoroscopic and/or ultrasound sequences as well as a method for such enhancement. Said aspects include a computer readable medium containing a computer program product for said enhancement of the image quality of said types of images.

[0014] Particularly, the present invention discloses a method that utilizes a particular noise reduction scheme based on the difference in intensity between central and surrounding values, whereby the image quality of the image sequences is enhanced. [0015] According to a first aspect of the invention a method is presented including the method steps of the independent method claim.

[0016] According to a second aspect of the invention it is directed to a computer-readable medium containing a computer program product for enhancing image quality of fluoroscopic and ultrasound images.

[0017] Further aspects of the invention are presented in the dependent claims.

[0018] An invention aspect includes a filter in an arrangement of a fluoroscopy and/or ultrasound imaging system for processing signal sequences, representing images of an object, obtained in form of a stream of two dimensional digital arrays containing pixel values forming an input signal to the filter, wherein the filter comprises:

- one or more rows, in parallel, of a series of a one-dimensional spatial FIR-filter in a direction of a first dimension, a one dimensional spatial FIR-filter in a direction of a second dimension, and a temporal IIR-filter.

[0019] According to a further aspect of the invention the filter comprises

- at least one frequency splitter for directing different frequency bands of said input signal to different rows of filters,

- at least one merger for merging the filtered signals from the different rows of filters into a common output signal.

[0020] A further aspect of the invention discloses said filter, wherein the splitter at an input of each row of filters splits a signal introduced to the row of filters into a high frequency band and low frequency band.

[0021] Another aspect of the invention discloses said filter, wherein said low frequency band from a splitter is output as an input signal to an input of an additional splitter at a subsequent row of the parallel arranged row of filters.

[0022] An advantage of the claimed invention is to provide a noise reduction method in medical image sequences that combines the noise reduction capabilities of the bilateral filter, its fast separable implementation, the lack of time delay of the NR filter and the advantages of processing several frequency bands. BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 describes an example of a 3D-filter used in the present invention. Fig. 2 illustrates three example functions as examples of functions that can be used to map the difference in intensity between a central pixel and a surrounding pixel into weights.

Fig. 3 shows an example of a recursive filter that can be used in the method according to the invention.

EMBODIMENTS OF THE INVENTION

[0023] Below, a number of embodiments of the invention are described in support of the enclosed drawings.

[0024] In an arrangement in fluoroscopy and ultrasound imaging systems image sequences are obtained in form of a stream of two dimensional digital arrays containing pixel values. These images are the input F_1n identified in figure 1 . The height and width of these arrays are referred to as the spatial dimensions, herein called first and second dimension, as they correspond to directions on the object that is imaged. For each pixel in these images the neighbourhood surrounding the pixel is considered when processing and in this context the pixel is called central pixel in the following.

[0025] The time between two consecutive images depends on the acquisition settings and defines the temporal resolution. For example, for a fluoroscopy system operating at 30 frames per second the temporal distance is 33 milliseconds.

[0026] According to one embodiment of the invention, the method described uses a 3D- filter that consists of two one-dimensional FIR filters in the spatial dimensions and a temporal NR filter, see figure 1. In the figure, illustrating an example of the 3D-filter F, it is indicated that the signal F_1n inputted to the 3D-filter F is split into a high frequency portion and a low frequency portion in a splitter 1 . The high frequency portion (HP) of the signal, forming a first frequency band of the input signal, is then applied to a first row i, of filters, including in series a first spatial filter 1 b for an x-direction, a first spatial filter 1 c for a y-direction and a first temporal filter 1 d, whereupon the portion of the signal passing said first row is output to a merging unit 1 e.

[0027] It is stated here, in the preceding paragraph, that a splitter 1 and a merger 1 e are used. It must be understood, that said merger is necessary only if more than one row of filters (f _! ...J_p-1) exists in the 3D-filter. A first splitter 1 may be used for splitting the input signal into a high frequency band and a low frequency band, but it is no obligation to use a row of filters of the inventory type for both said frequency bands output from the splitter 1. This is valid also for additional splitters 2 to p-1 . Frequency bands not being used for filtering according to the invention may be used for other purposes. Thus a merger 1 e should, as an example, not be necessary if the high frequency band, only, from a first splitter 1 is used for the filtering according to the invention.

[0028] The low frequency portion (LP) of the signal output from the splitter 1 is then applied to a second splitter 2 splitting the signal anew in a high frequency portion and a low frequency portion of the signal outputted from the first splitter 1 . The high frequency portion of a resulting second frequency band signal is then applied to a second row f₂ of filters, equivalent to the first row i_u including a second spatial filter 2b for the x-direction, a second spatial filter 2c for the y-direction and a second temporal filter 2d, whereupon the portion of the second frequency band signal passing said second row f₂ of filters is output to a merging unit 2e.

[0029] The signal processed in the second row f₂ of filters is then applied to the first merging unit 1 e, whereupon the desired filtered signal is let out as output signal F_out.

[0030] Figure 1 further indicates that an appropriate number of rows U , f₂, ...., f_p of filters including filters corresponding to the first and second rows of filter may be included in said filter F to filter out signals of the corresponding number of frequency bands, whereby each added row of filters is coupled in parallel to the previous rows of filters.

[0031] A splitter 1 to p-1 is arranged at the input of each row of filters U , f₂, ...., f_p for performing the generation of additional frequency bands to apply to subsequent filter rows. The input to successive splitters is the signal received from the low frequency band side of a preceding splitter.

[0032] Each signal output from a higher row from its merging unit (2e to 2_P-1) is applied to the merging unit of the row of the closest lower number of rows of filters.

[0033] The filter weights are controlled by the difference in intensities and spatial/temporal distance. First a one dimensional FIR average filter is applied along the two spatial directions: R = IX - Z(I Z₁ -Z₀ I) - Z₁

where I₀ denotes the intensity at the centre of the neighbourhood and /, the values at the pixel with index i. The parameter r is a configurable filter radius and the summation weights are determined by predefined filter coefficients, W₁, and a function f of the absolute difference in intensity between the central pixel and the summed pixel. The used filter coefficients, W₁, may be that of any low-pass filter such as sampled Gaussians, Binomial or cos²-filters. The function of the intensity differences should be monotonically decreasing. This ensures that pixel values with increasing difference contribute less to the result. Examples of such functions are:

Gaussian /(x) = e ²⁰

(1 )

1-0.4 x < ^

ZOO = σ 0.4

0 x > ^-

Linear 0.4 (2)

Exponential /(x) = e ^σ (3)

[0034] These three example functions are illustrated in figure 2 for σ = 0.4.

[0035] Choosing a Gaussian weight leads to a separable Bilateral filter if the spatial weights are also chosen as Gaussian.

[0036] A recursive NR average filter (an example of a recursive filter is shown in fig. 3) is applied in the temporal direction, again with the summation weights composed of fixed coefficients C₁ and the weight function depending on the absolute intensity difference:

Z = Xc₁ - Z(I Z₁ -Z₀ I) - /, (4) [0037] Here n is the maximum used time lag, typically a time lag of 1 to 3 frames is used. As the weight depends on temporal intensity differences the method could be seen as a very crude motion detector, because the main cause of larger intensity differences will be motion.

[0038] The order in which the one dimensional, 1 D-filters, as an example the spatial filters 1 b and 1 c, are applied does change the result slightly. Typically spatial filtering is done before temporal filtering.

[0039] Temporal noise may have other characteristics than spatial noise and it is therefore desirable to be able to use a different weight function for the temporal integration. This may be accomplished by the use of different parameters σ_x, σ_y and σ_t for the weight functions respectively.

[0040] Furthermore there may also be different noise characteristics along the spatial directions. This is particularly true for ultrasound where one direction (typically y) is along the ultrasound beam whereas the other direction correspondents to different directions. However even for a fluoroscopic system there might be different noise characteristics in the two spatial directions that are caused by the way the pixels are read from a sensor in the used equipment. Hence it is sometimes useful to choose different σ_x and σ_y.

[0041] The described 3D-filter can be applied to the data directly. Alternatively, selected frequency bands may be subjected to the filtering. Towards this end, as previously described, a spatial band (or high) pass filter is first applied to each image and the resulting frequency bands are then filtered in the rows (1 to p-1 ) of filters. Finally the results are combined with the residuals, from the last row p of filters, to obtain the result images. The resulting flow chart is illustrated in figure 1 as discussed above.

[0042] The filtering can be applied to ultrasound data before or after scan conversion.

[0043] The 3D-filter according to the claimed method and arrangement can be realized as an algorithm coded into a computer program product for performing said method, wherein said input signal representing said images are provided to the computer for the analysis.

Claims

1 . Method in an arrangement of a fluoroscopy or ultrasound imaging system for processing signal sequences, representing images of an object, obtained in form of a stream of two dimensional digital arrays containing pixel values, characterized in that the method comprises the steps of:

- providing a 3D-filter (F) having one or more rows (U ...f_p), in parallel, of: a one- dimensional spatial FIR-filter (1 b....p_b) in a direction of a first dimension, a one dimensional spatial FIR-filter (1 c....p_c) in a direction of a second dimension, and a temporal IIR-filter (1 d....p_d),

- inputting said signal containing signal sequences to an input (F_1n) of said 3D-filter,

- merging the output of each one of said rows (fi ...f_p) of filters to one output signal (F_out) being output from the 3D-filter (F).

2. The method according to claim 1 , further including the step of:

- arranging the number of rows ^₁...f_p) of filters to equal the desired number of used frequency bands.

3. The method according to claim 2, further including the step of: - using in said temporal IIR-filter (1 d...p_d) filter coefficients based on the intensity difference between pixel values.

4. The method according to claim 3, further including the step of:

- using different filter types and parameters for the filters in a row (f , ...f_p) of filters.

5. The method according to claim 4, further including the step of:

- computing the FIR filter results according to the function:

I) - /,

wherein, I₀ denotes the intensity at the centre of the neighbourhood and /, the intensity values at the pixel with index i. The parameter r is a configurable filter radius and the summation weights are determined by predefined filter coefficients, W₁, and a function f of the absolute value of the difference in intensity between the central pixel and the summed pixel.

6. The method according to claim 1 , further including the step of: - processing said input signal (F_1n) by applying different filter parameters to different frequency bands of the input signal.

7. The method according to any of claims 1 to 6, further including the step of: - selecting a predetermined number of frequency bands, which are subjected to the 3D- filter (F) as the input signal (F_1n).

8. The method according to claim 4, further including the step of:

- setting the function f, when calculating weight functions to be a Gaussian, Linear or Exponential function.

9. A filter (F) in an arrangement of a fluoroscopy or an ultrasound imaging system for processing signal sequences, representing images of an object, obtained in form of a stream of two dimensional digital arrays containing pixel values forming an input signal (F_1n) to the filter, characterized in that the filter (F) comprises:

- one or more rows (fi ...f_p), in parallel, of a series of a one-dimensional spatial FIR-filter (1 b.... P_b) in a direction of a first dimension, a one dimensional spatial FIR-filter

(1 c....p_c) in a direction of a second dimension, and a temporal IIR-filter (1 d....p_d).

10 The filter according to claim 9, wherein the filter (F) further comprises:

- frequency splitters (1 to p-1 ) for directing different frequency bands of said input signal to different rows (fi ...f_p) of filters,

- mergers (1 e....(p-1 )_e) for merging the filtered signals from the different rows (fi ...f_p) of filters into a common output signal (F_out).

1 1. The filter according to claim 9, wherein a splitter (1 to p-1 ) at an input of each row of filters splits a signal input to said splitter into a high frequency band and a low frequency band, wherein at least one of said frequency bands is introduced to the row (fi ...f_p) of filters.

12. The filter according to claim 1 1 , wherein said low frequency band from a splitter (1 to p- 1 ) is output as an input signal to an input of a subsequent splitter (1 to p-1 ).

13. A computer program product for performing the method according to claim 1 , wherein said computer program product is coded with an algorithm for performing the executions of the steps treating the input signal (F_1n).