Edge enhancement algorithm for low-dose X-ray fluoroscopic imaging
Introduction
X-ray fluoroscopy plays an important role in medical imaging, and this technique has been widely used in patient clinical examinations and interventional procedures (angiography). To protect the patients and staff during clinical examinations, the radiation exposure must be reduced in X-ray medical imaging systems. However, reducing the radiation level degrades the image quality, resulting in low-contrast images with noise and blurring due to quantum noise, which reduces the accuracy of clinical tests [1], [2], [3], [4]. To reduce the quantum noise, X-ray fluoroscopy devices usually have a time-average form of noise reduction pre-processor. However, even with this pre-processor, signal-dependent noise still exists in the edge region owing to the low-dose environment. In the region of motion of edge details, such as the clinical device and the patient's organ, blurring and artifacts are exacerbated by the time-averaging pre-processor. As a result, the degradation in the edge region makes clinical diagnoses difficult because there is insufficient information about the fine details of the image. Therefore, to obtain high-quality X-ray images in low-dose environments, an edge enhancement algorithm is required.
In digital image processing, sharpening techniques are a typical method for improving the edge details of blurred objects [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18]. These methods have been widely researched in the enhancement of medical images [5], [6], [7], [8], [9], [10], [11], [12]. However, enhancing medical images using sharpening method remains a challenging problem. Most sharpening methods use high-pass filtering to enrich the high-frequency content of the image because the desired detailed information is usually contained in high-frequency image components. Unsharp masking is widely used to improve image sharpness [13], [14], [15], [16], [17]. Moreover, unsharp masking has been recently studied as a nonlinear high-pass filter for increased flexibility [15], [16]. This approach is simple and effective and yields a sharper image. However, these methods cause two critical problems when applied to low-dose X-ray imaging: (1) overshoot and (2) artifacts and noise.
The overshoot problem may occur during the sharping process [18]. Because of differences between image areas, pixel values change quickly in some areas but slowly in others. Thus, if sharpening is performed in the same manner in both a relatively flat area and a rapidly changing area, such as edge detail, an overshoot problem is likely to occur in the rapidly changing area. This problem reduces the sharpness of the image and results in less than ideal results.
On the other hand, when the radiation dose is minimized in X-ray fluoroscopy, strong quantum noise inevitably arises in the image. Because noise can be unpredictable, it is treated as a random error in terms of probability and statistics. Although the X-ray device's own pre-processor can stabilize the signal-dependent noise, the quantum noise still exists in the image. Therefore, in the process of improving the sharpness, noise can be amplified to such an extent that artifacts are generated. Moreover, because the signal-dependent quantum noise is generally more distributed in the edge detail region, small noise errors become large artifacts during the sharpness improvement process.
Various methods for resolving the above problems have been proposed [3], [4], [5], [6], [7], [8], [9]. However, most of these proposed methods make a compromise between noise suppression and enhancement: noise is removed at the cost of reducing the enhancement effect. Moreover, even after those methods have been applied, the overshoot problem remains. To overcome these shortcomings and deficiencies, in this paper, we propose a method for removing the noise and artifacts while enhancing the edge details and suppressing the overshoot problem.
The contribution of this work is twofold. First, a method that incorporates region adaptive high-pass filtering (HPF) with transient improvement (TI) is proposed to enhance the sharpness of the degraded images without overshoot. The basic idea underlying TI is similar to conventional image sharpening, but there is a fundamental difference between the two algorithms. Conventional unsharp-based enhancement algorithms add high-frequency components to the original image to enhance edge details or enhance blurred details. Moreover, unsharp-based enhancement algorithms allow overshoot and undershoot to exaggerate the image detail. In contrast, the TI algorithm slowly transitions the blurred images in the sharpening process without causing overshoot problems, and creates sharp and natural edge transitions without artifacts. Consequently, the TI algorithm has been successfully applied to a variety of image processing applications [20], [21], [22]; however, it is not commonly used in fluoroscopic imaging. Thus, our proposed method is novel in that it combines region adaptive HPF and the TI algorithm to improve sharpness without causing the overshoot problem.
Second, an artifact and noise reduction (ANR) method that improves the edge quality of the degraded images is proposed. The ANR method uses an edge directional kernel obtained from the gradient covariance to correct the edge artifacts and remove the noise. This kernel, called a steering kernel [19], assigns large weights along the local edge direction to smooth the edge artifacts. Because X-ray fluoroscopic images acquired under low-dose environments are severely degraded, the artifacts and noise are amplified. Even with the improved TI algorithm, these edge effects are perceived more conspicuously during the sharpening process. Thus, the ANR process is performed with the proposed TI method to remove the artifacts and noise while enhancing the edge details. In general, a time-average pre-processor generates the artifacts within the motion region, which deteriorates the quality of image details. The behavior of the proposed ANR method is similar to that of diffusion-based adaptive smoothing methods [26]. However, the proposed method is more powerful than those methods because it provides direct solutions without the iteration process. In addition, it has better performance for preserving the details of the image, while reducing the noise and artifacts along edge direction. A block diagram of the proposed enhancement method is illustrated in Fig. 1. First, the degraded input image is obtained from the X-ray device that pre-processes the frames of low-dose X-ray images. Second, the sharpness of the input image is enhanced by region adaptive HPF and the TI method. Finally, the ANR method is performed on the sharpened image to improve the edge quality.
The remainder of this paper is organized as follows. Section 2 describes the proposed enhancement algorithm, which consists of region adaptive HPF with the TI and ANR methods. Section 3 demonstrates the performance of the algorithm using actual X-ray fluoroscopic images obtained under a low-dose environment and compares the results of our algorithm with results obtained using the conventional method. Finally, Section 4 presents concluding remarks.
Section snippets
Region adaptive high-pass filter with transient improvement
The proposed method consists of two steps. In the first step, a region adaptive HPF is adopted to enhance the slow transition of blurred edges. That is, a region adaptive high-pass filtered signal is added to the blurred signal to reconstruct the high-frequency components; the result is defined as follows: where hHF and F represent the region adaptive HPF and input signal, respectively. Here, a region adaptive Laplacian operator was used as hHF in the proposed method:
Experimental results
The numerical and visual performance of the proposed method was compared experimentally with two recently proposed enhancement methods: the hyperbolic secant square (HBSS) filter based sharpening method [5] and the generalized unsharp masking (GUM) algorithm [17]. The X-ray images for the evaluation were obtained using a clinical angiography prototype system supported by Samsung Electronics and chest phantom (Multipurpose chest phantom N1 “LUNGMAN,” Kyoto Kagaku) to simulate clinical images, as
Conclusion
In this study, an edge enhancement algorithm based on the TI and ANR methods was developed as a means of improving edge quality in low-dose X-ray fluoroscopic images. The developed algorithm combines region adaptive HPF with the TI method and simultaneously applies the ANR method with the TI method to improve the sharpness of edge details without causing overshoot problems. In the proposed ANR method, an edge directional kernel is used to reduce the artifacts and noise while preserving edge
Acknowledgment
This research was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Science, ICT and Future Planning [grant number. 2015R1A2A1A14000912].
Conflict of Interest
The authors declare that they have no conflict of interest.
References (26)
- et al.
A comparison of denoising methods for X-ray fluoroscopic images
Biomed. Signal Process. Control
(2012) - et al.
An enhancement method for X-ray image via fuzzy noise removal and homomorphic filtering
Neurocomputing
(2016) - et al.
Low-dose 2D X-ray angiography enhancement using 2-axis PCA for the preservation of blood-vessel region and noise minimization
Comput. Methods Progr. Biomed.
(2016) - et al.
Sharpening enhancement of computed tomography (ct) images using hyperbolic secant square filter
Optik
(2013) - et al.
Sharpening enhancement of ultrasound images using contourlet transform
Optik
(2013) - et al.
A continuous description of intervertebral motion by means of spline interpolation of kinematic data extracted by videofluoroscopy
J. Biomech.
(2012) - et al.
Vessel enhancement filter using directional filter bank
Comput. Vision Image Underst.
(2009) - et al.
Non-linear polynomial filters for edge enhancement of mammogram lesions
Comput. Methods Progr. Biomed.
(2016) - et al.
Image sharpening by morphological filtering
Pattern Recognit.
(2000) - et al.
Magnetic resonance image enhancement using stochastic resonance in Fourier domain
Magnetic Resonance Imaging
(2010)