Method for three-dimensional ultrasound imaging of heart
By employing deep learning and sparse sampling reconstruction techniques, the limitations of existing two-dimensional cardiac ultrasound imaging methods are overcome, generating efficient and accurate three-dimensional cardiac images suitable for real-time clinical diagnosis. This approach reduces equipment costs and computational complexity while improving imaging quality.
Patent Information
- Application Number
- CN202510837723.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-20
- Publication Date
- 2025-11-21
AI Technical Summary
Existing two-dimensional cardiac ultrasound imaging methods cannot fully present the three-dimensional structure and dynamic movement of the heart. They rely on the operator's experience, resulting in unstable image quality and making it difficult to meet the accurate diagnosis of complex cardiac lesions. Furthermore, existing three-dimensional ultrasound imaging technology is expensive, computationally complex, and suffers from insufficient noise processing.
Using deep learning and sparse sampling reconstruction techniques, multiple standard cross-sectional ultrasound video files are acquired, and image registration and reconstruction are performed. Deep convolutional neural networks and long short-term memory networks are combined to identify the cross-sectional type and cardiac time. Sparse sampling and low-rank matrix factorization methods are used for three-dimensional reconstruction, and image registration algorithms and filters are used to remove artifacts, generating high-quality three-dimensional images of the heart.
It achieves low-cost 3D cardiac imaging, improves imaging accuracy and efficiency, reduces computational complexity, is suitable for real-time clinical diagnosis, improves imaging resolution by 50%, shortens computation time to within 5 seconds, and improves signal-to-noise ratio by 40%.
Smart Images
Figure CN120983079A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical device technology, and in particular to a method for three-dimensional ultrasound imaging of the heart. Background Technology
[0002] Cardiac ultrasound imaging is a non-invasive imaging technique widely used in the diagnosis of cardiovascular diseases. Traditional two-dimensional ultrasound imaging relies on standard sections (such as the long-axis, short-axis, and four-chamber views of the ventricle), requiring the operator to manually adjust the ultrasound probe to obtain different perspectives of the heart. However, this method has significant limitations: two-dimensional images only provide a single planar view and cannot fully present the three-dimensional structure and dynamic motion of the heart; manual operation is highly dependent on the operator's experience, resulting in unstable image quality and low acquisition efficiency; furthermore, two-dimensional imaging is difficult to meet the precise diagnostic needs of complex cardiac lesions (such as valvular defects and congenital heart disease).
[0003] In recent years, three-dimensional ultrasound imaging technology has been developed, such as stereomicroscopy-based ultrasound imaging or CT / MRI fusion imaging, which can provide three-dimensional structural information of the heart. However, these technologies still face challenges: stereomicroscopy imaging requires high-density sampling, has high computational complexity, and is difficult to achieve real-time imaging; CT / MRI fusion imaging equipment is expensive and has limited ability to capture dynamic real-time images of the heart; in addition, existing three-dimensional ultrasound imaging methods still have shortcomings in terms of automation, imaging accuracy, and noise processing, which limits their widespread clinical application.
[0004] Therefore, there is an urgent need for an efficient, accurate, and automated three-dimensional cardiac ultrasound imaging method. This method should optimize the acquisition, registration, and reconstruction processes of cross-sectional images to overcome the limitations of traditional methods, improve imaging quality and real-time performance, and provide more reliable imaging support for the diagnosis of heart diseases. To address these issues, this invention proposes a three-dimensional ultrasound imaging method based on deep learning and sparse sampling reconstruction to achieve efficient and accurate three-dimensional cardiac imaging. Summary of the Invention
[0005] The purpose of this application is to overcome the shortcomings of the existing technology and provide a low-cost method for three-dimensional ultrasound imaging of the heart.
[0006] To achieve the above technical objectives, this application provides a method for three-dimensional echocardiography, comprising:
[0007] The ultrasound probe is placed in a first position to acquire a first video file at a first standard section of the heart. The first video file includes a movie file of at least one cardiac cycle.
[0008] The ultrasound probe is placed in the second position to acquire a second video file at a second standard section of the heart. The second video file includes a movie file of at least one cardiac cycle.
[0009] The ultrasound probe is placed in the third position to acquire a third video file at the third standard section of the heart. The third video file includes a movie file of at least one cardiac cycle, wherein the first standard section, the second standard section, and the third standard section are different standard sections.
[0010] Extract the first, second, and third cross-sectional images of the first, second, and third video files at the same cardiac moment within the cardiac cycle, respectively;
[0011] Register the first cross-sectional image, the second cross-sectional image, and the third cross-sectional image;
[0012] Based on the registered first, second, and third section images, a three-dimensional ultrasound image of the heart is generated.
[0013] Furthermore, a deep convolutional neural network, combined with a long short-term memory network, is used to identify the slice type and heart-stopping moments in each frame of the movie file.
[0014] Furthermore, the reconstruction method uses sparse sampling reconstruction technology.
[0015] Furthermore, sparse sampling and low-rank matrix factorization methods are used to reconstruct three-dimensional images of synchronized cardiac moments from different cross-sections, and the sparse signal is solved using the L1 norm optimization method.
[0016] Furthermore, an image registration algorithm is used to register different cross-sectional images belonging to the same moment of heartbeat.
[0017] Furthermore, mixed data is acquired through an ultrasound probe, artifact data is calculated, and the mixed data is filtered out of artifact data to obtain ultrasound data. The ultrasound data is then used to calculate the first section image, the second section image, and the third section image.
[0018] Furthermore, artifact data is obtained by analyzing the spectral characteristics of the received data and identifying frequency components that differ from the ultrasound signal, thus separating the artifact data.
[0019] Furthermore, the amplitude, phase, center frequency, and bandwidth of the artifact data are calculated, and the filter coefficients are calculated based on the amplitude, phase, center frequency, and bandwidth.
[0020] Furthermore, a filter is designed based on the filter coefficients; the filter is a band-stop filter.
[0021] Furthermore, the first labeled section is the long axis section of the left ventricle, the second labeled section is the short axis section, and the third standard section is the four-chamber section.
[0022] The method for three-dimensional ultrasound imaging of the heart disclosed in this application reduces equipment costs by acquiring first, second, and third section images at the same cardiac moment, registering them, and generating a three-dimensional cardiac image. Attached Figure Description
[0023] Figure 1 This is a flowchart of a three-dimensional ultrasound imaging method for the heart.
[0024] Figure 2 A flowchart for removing artifacts from ultrasound images. Detailed Implementation
[0025] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.
[0026] To keep the drawings concise, only the parts relevant to the invention are shown schematically in each figure, and they do not represent the actual structure of the product. Furthermore, for ease of understanding, in some figures, only one of components with the same structure or function is shown schematically, or only one is labeled. In this document, "one" can mean not only "only one" but also "more than one".
[0027] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0028] In this document, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0029] Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.
[0031] See Figure 1 As shown, this application provides a method for three-dimensional ultrasound imaging of the heart, comprising:
[0032] The ultrasound probe is placed in a first position to acquire a first video file at a first standard section of the heart. The first video file includes a movie file of at least one cardiac cycle.
[0033] The ultrasound probe is placed in the second position to acquire a second video file at a second standard section of the heart. The second video file includes a movie file of at least one cardiac cycle.
[0034] The ultrasound probe is placed in the third position to acquire a third video file at the third standard section of the heart. The third video file includes a movie file of at least one cardiac cycle, wherein the first standard section, the second standard section, and the third standard section are different standard sections.
[0035] Extract the first, second, and third cross-sectional images of the first, second, and third video files at the same cardiac moment within the cardiac cycle, respectively;
[0036] Register the first cross-sectional image, the second cross-sectional image, and the third cross-sectional image;
[0037] A three-dimensional cardiac ultrasound image is generated by reconstructing the image based on the registered first, second, and third section images.
[0038] To ensure video quality and timing consistency, the acquisition parameters for the first, second, and third video files were kept consistent. To improve data quality, a Kalman filter was used to preprocess the ultrasound signals during acquisition. The specific steps are as follows:
[0039] Use an ultrasonic probe to acquire mixed data containing raw signals and noise signals;
[0040] Analyze the spectral characteristics of mixed data to extract the amplitude, phase, center frequency, and bandwidth of the noise signal;
[0041] Design a Kalman filter whose state transition matrix is based on the spectral characteristics of the noise signal, with a bandwidth set to 0.5-5kHz;
[0042] The mixed data is filtered to separate noise signals and obtain high-quality ultrasound data.
[0043] Preferably, the first labeled section is the long-axis section of the left ventricle, the second labeled section is the short-axis section, and the third standard section is the four-chamber section. During the scan, the scanning physician first places the ultrasound probe in the first position, i.e., the long-axis section of the left ventricle, to acquire the first video file. The first video file includes a movie file of at least one cardiac cycle and is stored on the ultrasound device. A deep convolutional neural network, combined with a temporal analysis network, is used to identify the section type and cardiac time of each frame in the movie file. Preferably, the first video file includes movie files of five cardiac cycles, and five ultrasound images of the same cardiac time can be obtained through the neural network. The scanning physician can manually select the highest quality ultrasound image from the five frames, or artificial intelligence can be used to automatically identify the highest quality ultrasound image from the five frames. The left ventricular long-axis view is used to display the left ventricle, left atrium, aortic root, and mitral valve structures; the short-axis view is used to display ventricular cross-sections; and the four-chamber view is used to simultaneously display the left ventricle, left atrium, right ventricle, and right atrium. Each acquired video file covers at least one complete cardiac cycle (approximately 0.8-1.2 seconds, depending on heart rate), with a frame rate set to 30-60fps to ensure temporal resolution. The number of views can be adjusted according to equipment performance or diagnostic needs; for example, in high-precision scenarios, up to five views can be added (such as additional two-chamber views and aortic short-axis views) to further improve the detail of the 3D reconstruction.
[0044] After the initial scan at the first position, the ultrasound probe is placed at the second position, i.e., the short-axis section, to acquire the second video file. The second video file includes a movie file of at least one cardiac cycle, preferably a movie file of five cardiac cycles. Five ultrasound images at the same cardiac moment can be obtained via a neural network. The scanning physician can manually select the highest quality ultrasound image from the five images, or artificial intelligence can be used to automatically identify the highest quality ultrasound image from the five images.
[0045] After the second scan is completed, the ultrasound probe is placed at the third position, i.e., the four-chamber view, to acquire the third video file. The third video file includes a movie file of at least one cardiac cycle. Preferably, it is a movie file of five cardiac cycles. Five ultrasound images at the same cardiac moment are obtained using a neural network. The scanning physician can manually select the highest quality ultrasound image from the five images, or artificial intelligence can be used to automatically identify the highest quality ultrasound image from the five images.
[0046] As needed, the scanning physician can obtain video files from multiple standard sections of the heart using the method described above. To complete a 3D model of the heart, ultrasound images from three standard sections are sufficient. If more than three standard sections are available, a better 3D image of the heart can be obtained.
[0047] In performing 3D imaging of the heart, a deep convolutional neural network (DCNN) combined with a long short-term memory network (LSTM) is used to identify cross-sectional images of the same cardiac moment in three movie files: the first cross-sectional image, the second cross-sectional image, and the third cross-sectional image. Since the heart is beating, at the same cardiac moment, the heart can be considered to have the same shape, and cross-sectional images obtained at the same cardiac moment can be considered as scanned from a heart of the same shape. Due to the complex structure of the heart, sparse sampling reconstruction technology is used to improve imaging speed. The DCNN adopts a U-Net architecture, containing 5 convolutional layers and 5 deconvolutional layers to extract spatial features of video frames; the LSTM network contains 2 recurrent units to process the temporal features of the video. The input is a sequence of video frames (resolution 256×256 pixels), and the output is the cross-sectional type (major axis, minor axis, four-chamber heart) and the cardiac moment label (systole, diastole, etc.).
[0048] Preferably, sparse sampling and low-rank matrix factorization methods are used to perform three-dimensional reconstruction of synchronous cardiac moment images from different cross-sections, and the sparse signal is solved using the L1 norm optimization method.
[0049] After obtaining the first, second, and third cross-sectional images, an image registration algorithm is used to register the different cross-sectional images belonging to the same cardiac moment. After registration, a three-dimensional image of the heart is generated. Preferably, image registration can be based on feature points or optical flow. To obtain high-precision registration, coarse registration can be performed first using feature points to correct large translations and rotations; then fine registration can be performed using optical flow. Image registration can employ a feature-point-based coarse registration method, initially aligning by detecting the heart's edges or key anatomical landmarks (such as the apex or valves), followed by fine registration using optical flow to correct minor deformations and motion artifacts.
[0050] See Figure 2As shown, during the operation of ultrasound imaging equipment, artifacts may be generated due to electromagnetic interference from surrounding electronic devices. To reduce artifacts, the following method can be used: First, a mixed data is acquired by scanning with an ultrasound probe. Then, the artifact data is calculated. Next, a filter is used to remove the artifact data from the mixed data to obtain the ultrasound data. The ultrasound data is then used to calculate the first, second, and third section images. By removing the artifact data, a higher quality ultrasound image can be obtained. Specifically: a) Acquire mixed data containing the original signal and artifact signals using an ultrasound probe; b) Analyze the mixed data, identify and calculate the characteristics of the artifact signals (such as amplitude, phase, or frequency); c) Filter the mixed data using a Kalman filter to separate the artifact signals and obtain clean ultrasound data. Furthermore, the identification of artifact signals is achieved through the following methods: a) Calculate the spectral characteristics of the mixed data to determine the center frequency and bandwidth of the artifact signals; b) Based on the spectral characteristics of the artifact signals, construct a dynamic model for state estimation of the Kalman filter.
[0051] Artifact data can be calculated as follows: Analyze the spectral characteristics of the received data to identify frequency components different from the ultrasound signal, i.e., separate the artifact data. The artifact components can be determined through frequency domain analysis. Specifically, artifact signals usually have specific frequency characteristics; through Fourier transform or short-time Fourier transform, the system can identify artifact components with frequencies different from the ultrasound signal spectrum. Alternatively, artifact components can also be determined through time domain analysis. Specifically, artifact signals may exhibit periodic or burst noise; the system can identify artifact signals through autocorrelation analysis or time domain feature extraction. Once the artifact signal is identified, further extract the following features: amplitude, phase, center frequency, and bandwidth to confirm the artifact signal and prepare for filtering.
[0052] Furthermore, the amplitude, phase, center frequency, and bandwidth of the artifact data are calculated, and filter coefficients are calculated based on the amplitude, phase, center frequency, and bandwidth. Preferably, the filter is designed based on the filter coefficients, and the filter is a band-stop filter.
[0053] This invention utilizes deep learning to automatically identify the section type and cardiac time, combined with sparse sampling and L1-norm optimized reconstruction techniques, significantly improving the accuracy and efficiency of three-dimensional cardiac ultrasound imaging. Compared to traditional two-dimensional ultrasound imaging, this method provides complete three-dimensional structural information; compared to existing three-dimensional imaging technologies, this method has lower computational complexity and faster imaging speed, making it suitable for real-time clinical diagnosis. Experimental results show that this method improves imaging resolution by approximately 50%, reduces computation time to less than 5 seconds, and improves the signal-to-noise ratio by approximately 40%, providing reliable support for the accurate diagnosis of cardiac diseases.
[0054] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.
Claims
1. A method for three-dimensional ultrasound imaging of the heart, characterized in that, The method comprises the following steps: placing an ultrasound probe at a first position to obtain a first video file at a first standard section of a heart, the first video file comprising a movie file of at least one cardiac cycle; placing the ultrasound probe at a second position to obtain a second video file at a second standard section of the heart, the second video file comprising a movie file of at least one cardiac cycle; placing the ultrasound probe at a third position to obtain a third video file at a third standard section of the heart, the third video file comprising a movie file of at least one cardiac cycle, wherein the first standard section, the second standard section and the third standard section are different standard sections; extracting a first section image, a second section image and a third section image of the first video file, the second video file and the third video file at the same cardiac moment in the cardiac cycle, respectively; aligning the first section image, the second section image and the third section image; reconstructing according to the aligned first section image, the second section image and the third section image to generate a three-dimensional ultrasound image of the heart.
2. The ultrasound imaging method of claim 1, wherein, A deep convolutional neural network is used in combination with a long short-term memory network to identify the section type and the cardiac moment of each frame in the movie file.
3. The ultrasound imaging method of claim 1, wherein, The reconstruction method uses a sparse sampling reconstruction technique.
4. The ultrasound imaging method of claim 3, wherein, A sparse sampling and low-rank matrix decomposition method is used to reconstruct three-dimensional images of different sections at the same cardiac moment, and an L1 norm optimization method is used to solve the sparse signal.
5. The ultrasound imaging method of claim 1, wherein, An image alignment algorithm is used to align different section images at the same cardiac moment.
6. The ultrasound imaging method of claim 1, wherein, Mixed data is obtained by an ultrasound probe, and artifact data is calculated. A filter is used to filter out the artifact data from the mixed data to obtain ultrasound data. The first section image, the second section image and the third section image are calculated using the ultrasound data.
7. The ultrasound imaging method of claim 6, wherein, The artifact data is obtained by analyzing the spectral characteristics of the received data, identifying frequency components different from the ultrasound signal, and separating out the artifact data.
8. The ultrasound imaging method of claim 7, wherein, The amplitude, phase, center frequency and bandwidth of the artifact data are calculated, and the filter coefficients are calculated according to the amplitude, phase, center frequency and bandwidth.
9. The ultrasound imaging method of claim 8, wherein, A filter is designed according to the filter coefficients, and the filter is a band-stop filter.
10. The ultrasound imaging method of claim 1, wherein, The first standard section is a left ventricular long-axis section, the second standard section is a short-axis section, and the third standard section is a four-chamber heart section.