| Coronary heart disease is one of the most serious diseases affecting human health and with high mortality. Coronary angiography (CAG) and intravascular ultrasound (IVUS) is the most common imaging technique in diagnosis of coronary heart disease. Coronary angiography in the diagnosis of coronary heart disease was considered as the "gold standard" imaging technique, but it is hard to find obvious imaging features for critical lesions, plaque rupture, thrombosis, even through the multi position projection. IVUS is an invasive tomography ultrasound imaging technology which can make up for many deficiencies of CAG because IVUS can get access to real-time cross-sectional images of blood vessels, accurately obtain the information such as the vessel thickness, lumen diameter, plaque composition, vascular area and lumen area stenosis rate, evaluate the disease which is difficult to judge used CAG, determine the plaque progression and reversal. IVUS is known to be the "new gold standard" imaging technique in diagnosis of coronary heart disease.IVUS image sequences are acquired by pulling the catheter equipping an ultrasonic probe in the tip back with a motor running at a constant speed. Coronary artery attached inside the epicardial fat. The movement of the heart cause rotational and translational relative changes in short axis plane between ultrasound probe and vessel, the probe oscillate back and forth along the long axis, and the vessel cyclically expands and contracts due to the pressure changes. Summarizing, IVUS sequences suffer from different dynamic artifacts:Firstly, (1) subsequent frames can be misaligned because the probe is not fixed in short-axis plane, which can be decomposed into rotation and translation. Then, (2) some positions in the vessel are sampled multiple times or non-sampled because the probe moves back and forth along the long axis due to a continuous oscillatory movement. Finally, (3) subsequent frames have different vessel size because of the elastic vessel changed periodicity due to the movement of the heart. Motion artifacts in the longitudinal view of IVUS sequences show a sawing tooth distribution.Motion artifacts of IVUS image sequence are very strong, which have seriously affected coronary examination. The motion artifacts would interfere the visualization of blood vessels, reduce the accuracy of the lumen volume measurement, vessel 3D reconstruction and vascular lesion quantitative accuracy, cut down the stent intervention guidance value, in all, motion artifacts effect the diagnosis and treatment of coronary heart disease. There is no rigorous mathematical model to describe the motion artifacts so far, we need to adopt the computer aided method to suppress or compensate the motion artifacts.Gating method is an effective method for suppression of motion artifacts, the basic principle of gating method is to search an IVUS frame image in the same phase at each cardiac cycle, all searched images composite the gated sequence, the motion artifacts would be suppressed in the gated sets. Gating method is divided into ECG gating method and image-based gating method. ECG gating method can be also divided into online ECG gating and off-line gating method. As we all known, ECG gating method requires the collection of ECG signal is synchronized to the gathering of IVUS image sequence, while the demand of the hardware is very high, which is not feasible in clinical practice, this weakness and the shortcomings of ECG method limited its use in clinical. Therefore, researchers are committed to searching new methods to suppressing motion artifacts based on the IVUS image sequences in recent years, which be called as the image-based gating method. Image-based gating method is modeled on the ECG gating method, useing image processing technology to extract significant characteristics reflecting the IVUS frame, characteristics changed along the IVUS sequence, than formatting a signal which is similar to the ECG signal. The IVUS frames can be divided into two categories as end-diastolic and not end-diastolic frames. The end-diastolic frames are defined as the key frames, which composite the gated sequence.This paper presents two image-based gating methods, spectrum analysis method and manifold learning method.(1) Spectrum analysis method. The basic principle of the spectral analysis method is:the pixel gray value of IVUS frame reflects the structure characteristics of vascular tissue because IVUS is a kind of ultrasonic imaging technique. The same position pixel gray value changes along the sequence represents the change of organizational structure at this point, hidden heart motion information. An image-gated method was proposed based on this principle, which taking the local mean gray value of pixel point as the significant feature. The basic steps are as follows, first of all 1) converting the IVUS image from Cartesian coordinates to polar coordinates to reduce the burden of computer and computation, then filter the IVUS image in polar coordinates to use the local mean pixel gray value instead the gray value, sampling the filtered image at last. Secondly,2) extracting the LMG signals, than designing a filter by analyzed the spectrum amplitude of the LMG signal to select the key point. The LMG signals of key points composed the one-dimensional signal clusters which can reflect the heart movement. Thirdly,3) filterring the one-dimensional signal clusters by Butterworth band-pass filter with a center frequency of heart rate, the signal clusters were disturbed by free breathing, image noise, etc. except for the heart movement. Finally,4) extracting the key frames by selecting the maximum or minimum value of the filtered signal clusters, to composite the gated sequence. The Spectrum analysis method based on the consistency between the pixel gray value and organizational structure, the algorithm is simple and basically real-time. The weakness is hard to accurately retrieve key frames of the IVUS sequences with heart rate or vascular structure changes obviously.(2) Manifold learning method. IVUS sequences have some characteristics such as high dimension, large amount of data, nonlinear. Motion artifacts mainly caused by heart periodic movement lead to slow and continuously change between subsequent frames, and the images in the same phase of the cardiac cycle is very similar in the same sequence. Therefore, we can make such an assumption:IVUS sequences form a continuous smooth manifold on an image space, and images collected in the same phase of cardiac cycle are close in the manifold under the influence of the heart periodic movement. An image represents a point in the high dimensional space containing thousands of image features, while the motion artifacts can be described by several characteristics. The dimension of high-dimensional space is equal to the number of pixels in an IVUS image. The key content of this method is to extract and select features to reflect cardiac motion, and then form a one-dimensional signal to reflecting cardiac motion by these features. The basic steps are as follows:first of all, 1) converting the IVUS image from Cartesian coordinates to polar coordinates. The amount of data reduces 65% as the original one and the time of dealing the sequence is also cut down obviously. The average error between the dimension feature vectors is 0.9% after coordinate transformation, keeping the information integrity of data. Secondly,2) using LE algorithm to reduce the dimension of IVUS image after coordinate transformation. We choose the intrinsic dimension is 4, extracting such 4 low dimensional feature vectors expressing cardiac motion. Thirdly,3) building a distance function approximate ECG signal. It is hard to use low manifold distribution of IVUS image should obey clustering properties to searching key frames because the distribution of IVUS sequences in low dimensional space is irregular. Lastly,4) getting the local minimums’location of the distance function to retrieve the key frames, and then compositing the gated sets. Manifold learning method needs no prior knowledge of the coronary anatomy, raining samples and intervention. It is effective to those sequences with high changes in heart rate. The weakness is it is difficult to distinguish the rotation and translation in the short axis, the oscillation in the long axis and vascular contraction and expansion.Experimental data come from Nanyang Hospital,7 suspected patients with coronary heart disease,13 IVUS image sequences. Spectral analysis method and manifold learning methods were evaluated from the following aspects:firstly,1) comprising longitudinal cut of the original sequences and the gated sequences, showed that the gated sequences maintains the vessel, lumen and plaque in the original sequences trend, gated sequences vascular structure more smooth, saw tooth shape can be suppressed. Secondly,2) Comprising the results of two methods via time of the algorithm, number of key frames, average distance between the key frames, the fraction of the cardiac of the algorithm-selected frames, the effectiveness to deal with the sequences with large heart rate changes. Relative to the manifold learning method, spectrum analysis method algorithm is faster, spectrum analysis method for processing a frame IVUS image for an average of 18ms, and manifold learning algorithm is approximately 217ms. To the sequences with constant heart rate, number of key frames, average distance between the key frames, and the fraction of the cardiac of the algorithm-selected frames of these two methods are similar. To the sequences with varying heart rate, spectrum analysis method is difficult to accurately extract out the control sequence, while the manifold learning method overcomes the difficulties of heart rate variation. Thirdly,3) Manual sketching the EEM interface and lumen intima interface of 12 groups of IVUS image sequences, to calculate vessel volume, lumen volume, mean plaque burden of the original and gated sequences. Statistical results show that, on one hand, both vessel volume and lumen volume measured of the gated sequences are significantly smaller than the original ones, and there is no significant difference on mean plaque burden between original and gated sequences, which meets the need of the clinical diagnosis and treatment. On the other hand, variances of vessel area and lumen area of the gated sequences are significantly smaller than the original sequences, indicating that the gated sequences are more stable than the original ones. All show that, gated sets are helpful to clinical diagnosis and treatment and meeting the clinical needsTo sum up, this paper put forward two kinds of image-based gating method to suppress motion artifacts, spectrum analysis method is basically real-time, but it is difficult to accurately extract key frames to these sets with heart rate changes obviously. Manifold learning algorithm need more time to realize motion artifacts suppression, but it can accurately extract key frames to these sets with heart rate changes obviously. Gated sequences can maintain the key information for clinical diagnosis of coronary heart disease, reducing the amount of clinical data analysis. Gated sequence is much more stable compared with the original sequence, is more conducive to the diagnosis and treatment of coronary heart disease. |
PDF Full Text Request