FIELD OF INVENTION
The invention generally relates to image processing and particularly to a system and method of noise reduction in fluoroscopy.using a discrete wavelet transform (DWT).
BACKGROUND
Fluoroscopy is an imaging technique that uses X-rays to obtain real-time moving images of the interior of an object. Fluoroscopy provides continuous X-ray images of anatomies and is widely adopted as support in surgery procedures and diagnoses for real-time operations carried, out on patient for a long period of time. Radiation-induced for long period may cause injuries to the skin and underlying tissues. To avoid the damage, very low X-Ray doses are used. As a result, fluoroscopic images are strongly corrupted by noise. These images exhibit severe signal-dependent quantum noise that is referred to as quantum mottle and generally modelled as Poisson-distribution.
Many denoising algorithms were designed specifically for both signal-dependent noise (AAS, BM3Dc, HHM, TLS) and signal-independent additive noise (AV, BM3D, K-SVD). However, the execution time for existing denoising algorithms is very high and it introduces blurring and therefore, can't be used in real time operations. Many of the existing systems uses FPGA/GPU implementations which are not cost-effective. Hence, there is a need to develop a computationally inexpensive solution.

In the existing system, consecutive 18 frames are averaged to get a noise free image. As 1;8 frame averaging is computationally heavy, the output images are displayed for every 18 frame and the resultant image suffers from trailing effect because of the motion in consecutive frames.
Hence, there is a need to devise a noise reduction algorithm for fluoroscopy images to address the above-mentioned challenges.
The present invention is directed to overcoming one or more of the problems as set forth above.
SUMMARY OF THE INVENTION
Exemplary embodiments of the invention disclose a system and method for performing real¬time fluoroscopy image sequence processing. According to an embodiment, the disclosed system and method receives an image sequence in real time, the image sequence comprising a plurality of images. An input image frame is selected from the received image sequence. A modified input image is generated by applying a weighted frame averaging on the input image and an image frame next to the input image. A discrete wavelet transform (DWT) is applied to the input image to decompose into a first approximation band. The step of applying discrete wavelet transform (DWT) on the input image is repeated for a predefined number of times such that each decomposition uses previously obtained approximation band and generates a new approximation band. The discrete wavelet transform (DWT) is further applied to decompose the modified input image into a modified approximation band, a vertical band, a horizontal band and a diagonal band. An average modified approximation band is generated by computing average of the modified approximation band and the first approximation band. The discrete wavelet transform (DWT) is applied to decompose the average modified

approximation band into a subsequent modified approximation band, vertical band, horizontal band and diagonal band. The step of generating the average modified approximation band,'is repeated followed by decomposition for a predefined number of times, such that each subsequent generation uses previously obtained modified approximation band and corresponding approximation band obtained by the repeating step of DWT on the input image. An inverse DWT is applied to the last obtained modified approximation band and last obtained thresholded vertical, horizontal and diagonal bands to result in a denoised output approximation band. The step of applying inverse DWT is repeated for a predefined number of times to obtain a final denoised output image, sucrv.that each time last obtained denoised output approximation band is used with the thresholded vertical, horizontal and diagonal bands obtained in a step: previous to the corresponding denoised output approximation band.
BRIEF DESCRIPTION OF DRAWINGS
Other objects, features, and advantages of the invention will be apparent from the following
description when read with reference to the accompanying drawings. In the drawings, wherein
like reference numerals denote corresponding parts throughout the several views:
Figure 1 illustrates a high-level block diagram for noise reduction in fluoroscopy, according to
an embodiment of the invention; and
Figure 2 illustrates a detailed block diagram of core processing for noise reduction in
fluoroscopy, according to an embodiment of the invention.
DETAILED DESCRIPTION OF DRAWINGS

The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of exemplary embodiments of the invention. It includes various specific details to assist m that understanding but these are to be regarded as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. In addition, descriptions of well-known functions and constructions are omitted for clarity and conciseness.
According to embodiments of the'invention, a system and method for performing a real-time fluoroscopy image sequence processing is disclosed. Figure I illustrates a high-level block, diagram for noise reduction in fluoroscopy, according to an embodiment of the invention. As illustrated, a modified input image may be generated from denoised image output On-\ of an image frame ln-i, and an image frame ln that occurs next to the image frame In-i. According to an exemplary embodiment, the modified input image may be generated by performing weighted average of 60 percent of ln and 40 percent of On-i. A Discrete Wavelet Transform (DWT) may be applied to the modified input image. According to an embodiment, the generation of the modified input image and DWT occur in a core processing module. According to another embodiment, the output from the core processing module may be fed to a post processing module. According to an embodiment, the post processing module may be a contrast enhancement module. According to an exemplary embodiment, the post processing may be performed using histogram equalization, adaptive histogram equalization or gamma correction.
Exemplary embodiments of the invention disclose a system and method for performing real¬time fluoroscopy image sequence processing. According to an embodiment, the disclosed

system and method receives an image sequence in real time, the image sequence comprising £ plurality of images. An input image may be selected from the received image sequence. & modified input image is generated by applying a weighted frame averaging, on a denoised output of the input image and an image frame next to the input image. According to an exemplary embodiment, the modified input image may be generated by performing weighted
average of 40 percent of denoised output of the input image and 60 percent of the image frame
i i
next to the input image. A discrete wavelet transform (DWT) is applied to the input image to
decompose into a first approximation band. According to an embodiment, the DWT may be
applied to the input image to decompose into multiple frequency bands including a firs*
approximation band, a vertical band, a horizontal band and a diagonal band. The step ofr
applying discrete wavelet transform (DWT) on the input image is repeated for a predefined
number of times such that each decomposition uses previously obtained approximation band
and generates a new approximation band. According to an exemplary embodiment, if
predefined number of decompositions is defined as 2, then the DWT may be applied on the
first approximation band to generate a second approximation band. According to another
exemplary embodiment, if predefined number of decompositions is defined as 3, then the DWT
may be further applied on the second approximation band to generate a third approximation
band.
The discrete wavelet transform (DWT) may be further applied to the modified input image to
decompose the modified input image into a modified approximation band, a vertical band, a
horizontal band and a diagonal band. According to an exemplary embodiment, the
decomposition of the modified input image may be performed using a Harr wavelet
decomposition filter. An average'modified approximation band is generated by computing
average of the modified approximation band and the first approximation band. The discrete
wavelet transform (DWT) is applied to decompose the average modified approximation band

into a subsequent modified approximation band, vertical band, horizontal band and diagonal
band. The step of generating the average modified approximation band may be repeatecl
followed by decomposition for a predefined number of times, such that each subsequent
generation uses previously obtained modified approximation band and corresponding
approximation band obtained by the repeating step of DWT on the input image. According to
an exemplary embodiment, a second average modified approximation band for the second
generation may be computed by average of the corresponding modified approximation band
and the second approximation band. Similarly, a third average modified approximation band
for the third generation may be-., computed by average of the corresponding modified
approximation band and the third approximation band. ';
An inverse DWT may be applied to the last obtained modified approximation band and last obtained thresholded vertical, horizontal and diagonal bands to result in a denoised output approximation band. According to an embodiment, the last obtained modified approximation band may refer to the modified approximation band corresponding to last level of decomposition of predefined levels of decomposition. According to another embodiment, the last obtained thresholded vertical, horizontal and diagonal bands may refer to the thresholded vertical, horizontal and diagonal bands corresponding to last level of decomposition. According to an embodiment, the thresholded vertical, horizontal and diagonal bands may be obtained by applying soft thresholding to the vertical, horizontal and diagonal bands. According to another embodiment, the soft thresholding may be applied to all vertical, horizontal and diagonal bands using Bayesian shrinkage estimate to suppress the noise present in the image. The step of applying inverse DWT is repeated for a predefined number of times to obtain a final denoised output image, such that each time last obtained denoised output approximation band is used with the thresholded vertical, horizontal and diagonal bands obtained in a step previous to the corresponding denoised output approximation band.

Figure 2 illustrates a detailed block diagram of core processing for noise reduction fn fluoroscopy, according to an exemplary embodiment of the invention. A plurality of real time image frames of a fluoroscopy image sequence is received. According to an embodiment, the fluoroscopy image sequence may comprise a plurality of images. An input image frame In is selected from the image sequence and a modified input image (202) may be generated by-applying a weighted frame averaging on the selected input image frame ln and an image frame next to the input image frame ln+i. According to an embodiment, a discrete wavelet transform (DWT) may be applied to the input-image frame In to decompose into multiple frequency bands, including a first approximation band A(n-i)i(204). The DWT may be repeated on the input; image for a predefined number of times. According to an exemplary embodiment, the DWT may be performed three times on the input image. The DWT may be applied on the first approximation band to generate a second approximation band (230). Similarly, the DWT may be applied on the second approximation band to generate a third approximation band (244).
The DWT (206) may be applied to the modified input image (202) to decompose the modified input image into a first modified approximation band Ani(208), a first vertical band Vni(210), a first horizontal band Hni(2l2)anda first diagonal band Dni(214). According to an exemplary embodiment, the decomposition of the modified input image may be performed using a Harr wavelet decomposition filter. An average of the first modified approximation band (208) and the first approximation band A(n-i>i(204) may be computed to generate a first average modified approximation band MAni(2l6).
The DWT (218) may be applied to decompose the first average modified approximation band MAni(2l6) into a second modified approximation band An2(220), a second horizontal band

Hn2(222) and a second diagonal band Dn2(224) and a second vertical band Vn2(226). An average of the second modified approximation band An2(220) and the second approximation band A(n-i)2(230) may be computed to generate a second average modified approximation band MAn2(228).
The DWT (232) may be applied to decompose the second average modified approximation band MAn2(228) into a third modified approximation band An3(234), a third vertical band Vn3(236), a third horizontal band Hn3(238) and a third diagonal band Dro(240). An average of the third modified approximation band An3(234) and the third approximation band A