A novel structural features-based approach to automatically extract multiple motion parameters from single-arm X-ray angiography

https://doi.org/10.1016/j.bspc.2016.09.012Get rights and content

Highlights

  • A novel multi-motion parameter model of single-arm angiography sequence.

  • Automatic detecting and tracking robust vascular branch points.

  • A novel time-frequency domains iteration separation algorithm on multi-parameters.

  • Providing clinical support for clinicians to quantitatively analyze heart disease.

Abstract

It is essential to extract dynamic information about a patient's heart from a medical X-ray angiography image sequence for a quantitative medical diagnosis. As the motions included in the angiography sequences are the mixture of various motion signals, such as the body's integral translation, respiratory motion, cardiac impulse, and tremor, automatic separation of these signals challenges the effectiveness of the image information processing method.

This paper has proposed an optimal time-frequency domains iteration separation algorithm for multi-motion parameters (TFISA-MMP) to obtain the physiological parameters including the translation. The main procedures of the TFISA-MMP algorithm include three parts. First, the algorithm automatically extracted a set of relatively stable branch points from the coronary artery angiography image, and then automatically tracked these branch points in the sequence image to obtain their motion curves that changed with the time. Second, with the guidance of the multi-motion parameter model, the initial values of each component were estimated based on Discrete Fourier Transformation (DFT). Moreover, the initial values of each motion component were optimized using the global mean square minimum error between the estimated reconstructed signal and original signal and the local mean square minimum error in the frequency domain for each frequency component. Finally, these motion components were estimated by minimizing the residual signal between the original signal and the reconstructed signal via the loop iteration to obtain the estimated optimal motion components, such as two-dimensional (2D) heartbeat, tremor, respiratory motion, and translational motion. Both visible human coronary model simulation experiments and the clinical experiments of single-arm X-ray angiography images of many individuals verified the correctness, validity, and clinical applicability of the separation method.

Introduction

Respiratory motion refers to the rhythmical enlargement and systole of thorax, which can complete inhalation and exhalation. Respiratory motion will spur the integrated translational motion of the human heart in the 3D space [12]. From a clinical perspective, when devices such as CT or MRI are used to image the heart, the heart pulsation will generate “artifacts”, thus reducing imaging quality [7], [13]. Although ECG-gating technology can be used to solve the motion problem caused by the heart beat, the interference caused by the respiratory motion cannot be effectively solved and it is a common practice to require the patients to hold their breath. However, longer imaging time makes it difficult to handle the respiratory motion problem.

In X-ray angiography systems, the coronary artery and blood vessels on the angiography plane will produce a 2D motion under the influence of respiratory motion. On the other hand, during the angiography process, the physician may move the catheter bed to ensure that the angiography sequence cover the whole coronary artery system, which will result in a 2D translation among different frames in one imaging sequence. Therefore, the coronary artery angiography image will record the projection of heart pulsation in the 2D plane, and will simultaneously superpose the 2D motion of the coronary artery on the angiography plane caused by the human body's respiratory motion and the possible integral 2D translational motion of the patient. As a result, if the 3D cardiovascular tree must be correctly reconstructed based on the dynamic 2D angiography image (see Appendix D), the human body's translation or other translations must be separated from one another. In addition, it is also very important to correctly separate a variety of motion parameters for a quantitative diagnosis of the patient's cardiovascular disease, and there is no universal method that can simultaneously separate these movements.

Problems with estimating and separating translational motion: Generally, translation motion is used for image registration among frames. The medical image registration mentioned in Ref. [4] can be generally divided into the internal and external registration. External registration, which must set some obvious marks before angiography, is usually invasive. Both a marking method and a segmentation method are used in the internal registration method. Marking methods are used for registration by selecting some anatomical structural points in the human body that are not always present in the images and must be well understood. Although the segmentation method used the segmented anatomical structure lines for registration using the guidance of the rigid model and deformable model [11], it cannot separate the overall rigid translational motion alone but extracts the motion of the anatomical structure lines. In addition, the blood vessel motion extracted from the angiography image not only includes the translational motion of the patient but also the physiological motion of the blood vessels and respiratory motion. Moreover, the extraction of the translational motion is rarely mentioned in the medical imaging field. The literature [15] proposed the use of a spatially variable gray-level correction method for aligning the contrast image with the mask image, but registration accuracy is affected by the correction error and the method is only a two frame alignment that does not use sequence information.

Problems with estimating and separating respiratory motion: As some non-cardiac structures (e.g. Diaphragm muscle) cannot move together with the cardiac structures, the previously reported methods [9], [10] set markers, track them, and finally extract their motion, which can be considered as the approximate respiratory motion. Another commonly used method in X-ray angiography is direct manual tracking of the feature points of non-cardiac structures in the angiography image or recording the motion of these points during angiography. The former is a poor method in practice because we cannot ensure that each frame of the angiography image has the marking points that can meet the condition and one must have a good understanding of the anatomical structure of the human body; while the latter requires a substantial amount of experimental control, which is not applicable in general clinical practice. A previously reported method [8] was used to obtain the respiratory parameters under the double-arm X-ray contrast conditions. The idea for separating the heart pulsation and respiratory motion is that selected two angiography images that are taken at the same time from different projection vision angles are used for a 3D reconstruction and to obtain the 3D spatial distribution of blood vessels. Then, all angiography images captured during one respiratory cycle are reconstructed to obtain a 3D structure sequence that can be decomposed based on the B-Spline approximation and least square fit method to construct independent heart and respiratory motion models. Relatively speaking, a reliable estimate result of respiratory motion can be obtained using this method, but the 3D motion sequence of the coronary artery shall be reconstructed first and then the respiratory motion can be decomposed using a very complex processing procedure. Moreover, due to the restriction of the double-arm X-ray angiography conditions, it is very hard to apply this technique to the single-arm X-ray angiography in medical practice. In addition, during double-arm three-dimensional reconstruction, the time phase of respiratory motion in the angiography images is taken from two different vision angles involved in the reconstruction; thus the previously reported method [8] is not applicable to the single-arm angiography system.

Problems with analysis of short duration mixed signals with complete/incomplete periodic components: Some researchers have tried to use the conventional Fourier method to analyze medical signals. Although these researchers have made great progress in various fields of application (non-signal separation) when the short duration mixed signal not only contains periodic components but also non-periodic components and even incomplete periodic components, the error in the signal separated by the conventional Fourier method is large. Digital vascular angiography image sequences are one example. With respect to breathing, the contrast agent in the image lasts for a quite short duration.

The previously reported DFT method [6] is used to simulate the quasi-periodic heart pulsation for the endoscope image, and thus the purpose is different from the respiratory motion parameter estimate after automatic extraction of feature points in the single-arm angiography image in this paper. A previously reported method [1] was used to reconstruct the heart surface, which uses a two-vision angle camera to take the pictures and uses the jumpy diaphragm model to simulate the heart surface and estimate the parameters of heart surface model based on the Kalman filter or the expanded Kalman filter. Thus, the method is not comparable to the separation of heart pulsation and respiratory motion in our paper. In addition, respiratory motion is also mentioned in some other papers; however, most of these publications simulated or predicted the respiratory signal. The simulated respiratory signals not only have several periods, but are also dedicated to the endoscopic images and the ultrasonic images.

In particular, the angiography time is usually very short, due to the toxicity of the contrast agent in the real clinical application. As a result, the length of the angiography sequences is finite, and the short duration motions of structural feature points in the angiography sequences composed of multiple period components, non-period components and non-complete period components are finite. Therefore, the conventional DFT algorithm cannot accurately separate the components. At present, there are no reports on the separation of respiratory and heart pulsation in a recorded sequence of single-arm angiography images of a finite length.

The necessity of automatic estimation and separation of the multi-motion parameters: In summary, a powerful method is needed to automatically separate translational motion, respiratory motion and heartbeat motion from a short duration two-dimensional imaging sequence to solve these problems. In addition, the patient's heartbeat, which contains useful diagnostic information (such as local tremors), must be extracted as accurately as possible.

Obviously, the structure of the feature point motion on the coronary angiogram contains the overall translational movement, respiratory movement, heartbeat, and pathological tremor of the patient. Therefore, this paper has proposed a multi-motion parameter model of structural branch points and a multi-parameter extracted method for short duration mixed signals with complete/incomplete periodic components. Under the guidance of the model, the method automatically detects the structural branch points of the blood vessels in the coronary artery vessel angiogram and automatically traced these feature points in the angiography sequence to obtain their displacement curves, which varied with time. We can accurately obtain translational motion, respiratory motion, heartbeat motion, etc. By adjusting the frequency and period-compensated frequency-time domains crossing iterative optimization algorithm. Therefore, the method has a more extensive applicability and flexibility when compared with the method that simply traces non-heart structural feature points and can be applied to almost all angiography sequence images (it certainly requires a sufficient, clear blood vessel distribution and two or more heart pulsation cycles, which is easily accomplished), including double-arm X-ray angiography.

This paper is organized as described below. Section 2 discusses the branch points of the cardiovascular tree, the multi-motion parameter model of branch points and the decomposition of mixed discrete signals with complete/incomplete periodic components. Section 3 proposes an algorithm for automatical extraction of cardiovascular feature points of a single angle image sequence. Section 4 proposes the multi-motion parameter model and frequency amendment-compensated period crossing iterative optimization algorithm, and discusses the various motion signals estimated using the optimal TFISA-MMP. The signals include the heartbeat signal, the respiratory signal, the translational signal and possible heart tremor. Section 5 reports the quantification of the complete experimental results of the simulations and the X-ray angiogram images of clinical patients. Section 6 lists the conclusions and the outlook for future work.

Section snippets

Multi-motion parameter model of feature points

In this section, we mainly introduce the branch points of the vascular tree, the multi-motion parameter model of the feature points based on the branch points, and the decomposition of mixed discrete signals.

Automatic extraction of cardiovascular feature points

Pattern recognition and image processing were comprehensively applied in this paper to propose a method for automatically extracting the branch points from the angiography image, which differs from other references because it can separate the respiratory motion. Using this method, the automatic extraction of branch points in the angiography image can be divided into two phases: (1) the automatic extraction of branch points in each single vision angle sequence image, and (2) the automatic

Estimating the motion parameters under the guidance of the model

The normal heart beats 60–100 times/min in clinic measurement, and the cardiac cycle is 0.6–1.0 s. Cyclic motion with a frequency greater than 140 times/min shall be diagnosed as pathologically abnormal motion, when the period is less than 0.42 s. If the motion frequency is higher than the normal heartbeat based on the coronary artery angiography image and when the amplitude of signal exceeds several pixels, it is reasonable to consider that the signal is a certain useful signal (or signal for the

Simulation experiments using a dynamic model of a visible human heart

A 3D model of the cardiovascular system was used to verify the correctness of the algorithm reported in this paper. The established model was based on the medical physiological research and adopted the anatomical slice picture in the Visible Man data set from the American Visible Human Project [14] (VHP: Visible Human Project). See the references for the process of establishing the model [17], and the model characteristics are summarized below.

Firstly, the contours of the four heart cavities

Conclusions and discussion

The total number of extracted branch points in patients 1, 2, and 3 is 4, 5 and 5, respectively, and the total number of traced points is 678. The number of accurate tracing is 675, and 3 were lost. Thus, 675/678 = 99.557% of the traces were correct. The data further verify the robustness of the proposed method.

As the various motions included in the angiography image sequence are a combination of various motion signals, such as body translation, respiratory motion, cardiac impulse and tremor,

Acknowledgements

We appreciated the United States National Library of Medicine, which provided us with the Visible Man dataset to construct the heart coronary artery system model. We also express our sincere appreciation to Beijing Chao-yang Hospital, which provided us with the X-ray angiography image sequence to support the smooth development of our scientific research.

All experiments in this paper were supported by the research subject of “digital subtraction angiography intelligent analysis and 3D

References (17)

  • J.B.A. Maintz et al.

    A survey of medical image registration

    Med. Image Anal.

    (1998)
  • T. Bader et al.

    Model-based motion estimation of elastic surfaces for minimally invasive cardiac surgery

  • F. Benmansour et al.

    Tubular structure segmentation based on minimal path method and anisotropic enhancement

    Int. J. Comput. Vis.

    (2011)
  • A. Frangi et al.

    Multiscale vessel enhancement filtering

    Med. Image Comput. Comput. Assist. Interv. LNCS

    (1998)
  • A.V. Oppenheim et al.

    Signal and Systems

    (1997)
  • R. Richa et al.

    Beating heart motion prediction for robust visual tracking

  • C. Ritchie et al.

    Minimum scan speeds for suppression of motion artifacts in CT

    Radiology

    (1992)
  • G. Shechter et al.

    Three-dimensional motion tracking of coronary arteries in biplane cineangiograms

    IEEE Trans. Med. Imaging

    (2003)
There are more references available in the full text version of this article.

Cited by (11)

  • ILU-Net: Inception-Like U-Net for retinal vessel segmentation

    2022, Optik
    Citation Excerpt :

    Retinal vascular diseases, such as diabetic retinopathy [1], retinal vein occlusion [2], hypertensive retinopathy [3], high myopic retinopathy [4], and other chronic eye diseases [5], are commonly occurred in fundus retinal vessels.

  • Simultaneous denoising and enhancement for X-ray angiograms by employing spatial-frequency filter

    2020, Optik
    Citation Excerpt :

    However, the obtained X-ray angiograms are low-quality due to the additive Gaussian white noise (AGWN) [2–4] and uneven intensity, which are not conducive for the CVDs’ diagnosis, as shown in Fig. 1(a). Also, such degraded angiograms are not convenient to extract cardiovascular for reconstructing three-dimensional cardiovascular and analyzing cardiovascular motion [5,6]. Usually, the combination of image enhancement and denoising in one scheme is very difficult, since the goals of the two tasks conflict.

  • Accurate vessel extraction via tensor completion of background layer in X-ray coronary angiograms

    2019, Pattern Recognition
    Citation Excerpt :

    Due to the complex dynamic structures, the background layer seriously disturbs the observation and measurement of vessels. To facilitate the diagnosis and treatment of cardiovascular diseases, automatical extraction of the vessel layer and effective removal of the nonvascular background layer has become a prerequisite to improve the visibility and detection of vessels for various clinical applications, such as 3D reconstruction of coronary arteries [12], 3D/2D image coronary registration [13], coronary artery labeling [14], heart’s dynamic information extraction [15], and myocardial perfusion measurement [4,5,16]. In addition, vessel extraction is usually used as a preprocessing step to remove noises and complex backgrounds from XCA images while emphasizing vessel-like structures for most sophisticated pipeline algorithms including vessel segmentation and vessel centerline extraction.

  • Spatially adaptive denoising for X-ray cardiovascular angiogram images

    2018, Biomedical Signal Processing and Control
    Citation Excerpt :

    However, due to the effect of imaging apparatus or image transmission, the obtained angiographic images, polluted by the additive Gaussian white noise (AGWN) [2], are not conducive for the doctors to diagnose the patient's condition. And it is also not convenient to extract cardiovascular from the noisy images for long-term study and research (such as three-dimensional cardiovascular reconstruction and cardiovascular motion analysis [3]). Therefore, it is imperative to restore the clear angiogram images with integral cardiovascular tree.

View all citing articles on Scopus
View full text