Automatic estimation method of adult myocardial overall longitudinal strain index based on echocardiogram

Abstract

An automatic estimation method for an adult myocardial overall longitudinal strain index based on an echocardiogram comprises the following steps that a segmentation model is established, and accurate segmentation of a heart area is achieved; processing the heart region obtained by segmentation, reserving a myocardial region, and calculating endpoints of internal and external outlines of the myocardial according to the characteristics of the myocardial outlines to obtain internal and external outline lines of the myocardial region; estimating a cardiac apex position according to the myocardial gravity center; determining a center line of the myocardial region; dividing the center line of the myocardial region, and dividing the myocardial region into a plurality of myocardial segments; the myocardial segment displacement is calculated with the myocardial segment as the unit, and a myocardial strain curve is obtained. According to the method, framing, myocardial region segmentation, myocardial region extraction, myocardial inner and outer contour and center line calculation, cardiac apex position estimation and GLS index estimation can be automatically carried out on an ultrasonic video, the influence caused by errors and subjectivity of manual analysis is reduced to a great extent, and meanwhile the interference influence caused by image quality can be overcome to a certain extent.

Description

Technical Field

[0001] The present invention relates to the technical field of medical ultrasonic imaging, and in particular to an automatic estimation method for adult myocardial global longitudinal strain index based on echocardiography. Background Art

[0002] The heart is a core organ in the human body. Accurately and quantitatively assessing the degree of cardiac dysfunction is crucial for the early diagnosis, intervention, and prognosis of heart disease. Among various cardiac imaging devices, ultrasound imaging technology has been widely used to assess cardiac structure and function due to its unparalleled advantages, including non-invasiveness, real-time imaging, and low cost.

[0003] Global Longitudinal Strain (GLS) is defined as the percentage of longitudinal shortening of the myocardium between end-diastole and end-systole, as described in reference [1]. It is an important parameter for measuring myocardial contractility. It has important reference value for diagnosing various heart diseases such as heart failure and myocardial infarction. At present, existing technologies mainly rely on manual operation of ultrasound doctors and combine commercial software to perform myocardial strain analysis, as described in references [2] to [5]. Since the process of manual analysis of myocardial strain is highly dependent on the professional experience and skills of the operator, the analysis results have a certain degree of difference between different observers, which may affect the accuracy and consistency of the diagnosis; and due to the diversity of imaging equipment and the differences in analysis software, the measurement of myocardial strain has the problem of poor measurement value and repeatability between equipment from different suppliers, as described in reference [6]. This limits the widespread application and standardized promotion of strain parameters in clinical practice.

[0004] In medical ultrasound imaging, due to the complexity of the soft tissue structure inside the human body, there are dense scatterers of different sizes and shapes. When the ultrasound beam passes through these tissues, the ultrasound wavefront will be scattered, reflected, and even diffracted, thus forming a series of fine and dynamic granular "spots" on the imaging plane. Therefore, the spots can be tracked by block matching as described in reference [7] or optical flow technology as described in reference [8]. The method of locating the myocardial area is widely adopted. Reference [9] uses the standard PIV (Particle Image Velocimetry) algorithm to track spots and uses the block-based fast Fourier transform (FFT) set cross-correlation to estimate the inter-frame displacement. Reference

[10] proposes an adaptive estimation method for the block size of the block matching speckle tracking template for echocardiography. Reference

[11] proposes a method based on multi-resolution confidence optimization optical flow to estimate the motion of each segment of the myocardial wall, combined with manual segmentation of the myocardial wall and dynamic tracking of the region of interest, to finally obtain the motion information of each segment of the myocardial wall.

[0005] The above-mentioned method of tracking the myocardium in ultrasound images based on traditional computer vision technology is relatively effective when the ultrasound image quality is good. However, in practice, clinical echocardiography is often affected by a variety of adverse factors, such as patient movement, respiratory interference, equipment resolution limitations, and changes in ultrasound beam penetration depth. These factors can easily lead to a decline in image quality, resulting in problems such as blurred heart edges, missing parts of the structure, high noise levels, and irregular deformation of the overall heart contour in the obtained echocardiogram. Figure 1 In this case, the speckle tracking-based method often fails and requires multiple manual corrections, which seriously affects the measurement accuracy and efficiency.

[0006] In recent years, deep learning technology based on multi-layer neural network models has made breakthrough progress in the field of medical image processing and analysis, and more and more related studies have begun to focus on myocardial motion in ultrasound images. Reference

[12] proposed an unsupervised method based on the UNet architecture to estimate canine myocardial motion in short-axis views. Reference

[13] proposed a left myocardial motion tracking method based on displacement flow U-Net (DispFlow_UNet) and biological variational autoencoder (VAE). This method was only experimented on the ACDC (Automatic Cardiac Diagnosis Challenge) dataset, as described in reference

[14] . The above experiment only estimated myocardial motion in short-axis views, did not involve the analysis of relevant sections of the apical two-chamber heart and four-chamber heart, and did not calculate the GLS index. Reference

[15] proposed a point matching scheme to quantitatively estimate left ventricular displacement and strain in canine 4D ultrasound cardiac image sequences, but most adult cardiac ultrasound examinations currently rely on 2D imaging technology and cannot be directly applied in clinical environments.

[0007] In order to solve the problem of myocardial positioning and segmentation, the literature

[16] proposed a deep learning method based on the FlowNet2 method to estimate myocardial motion and use the Kalman filter to automatically calculate the GLS of the apical four-chamber heart view. This method does not segment the myocardium, but uses 180 uniformly distributed points (36 longitudinal and 5 radial) and 280 triangular cells to grid the myocardium and calculate the motion displacement of any point of the myocardium. It is difficult to highlight the abnormal motion of the local area and cannot quickly and accurately locate the lesion area. The literature

[17] proposed a method to estimate the motion of the left ventricular wall using active polynomials, extract and track the endocardial boundary, and calculate the displacement of the myocardial segment. This method uses the 17-segment myocardial segmentation method to diagnose myocardial infarction in the myocardial segments in the apical two-chamber heart view and the four-chamber heart view, removes the apical cap area of ​​the 17th segment, and calculates the displacement of the remaining six segments, which to a certain extent limits the ability to comprehensively evaluate the overall myocardial motion state.

[0008] References:

[0009] Literature [1]: Voigt JU, Pedrizzetti G, Lysyansky P, Marwick TH, Houle H, BaumannR, Pedri S, Ito Y, Abe Y, Metz S, Song JH, Hamilton J, Sengupta PP, Kolias TJ, d'HoogeJ, Aurigemma GP, Thomas JD, Badano LP. Definitions for a common standard for 2D speckle tracking echocardiography: consensus document of the EACVI / ASE / Industry Task Force to standardize deformation imaging. Eur Heart J CardiovascImaging.2015Jan;16(1):1-11.doi:10.1093 / ehjci / jeu184.Epub 2014Dec 18.PMID:25525063.

[0010] Reference [2]: Si Tong, Wu Jing. Consistency analysis of automatic measurement of right ventricular area fraction change using AutoEF software based on speckle tracking echocardiography technology [J]. Imaging Research and Medical Applications, 2023, 7(16): 62-64.

[0011] Reference [3]: Wu Wanzhen. Research on key technologies for automatic cardiac motion analysis of cardiac magnetic resonance images [D]. Harbin Institute of Technology, 2022. DOI: 10.27061 / d.cnki.ghgdu.2022.002297.

[0012] Literature [4]: ​​Lopez L, Colan SD, Frommelt PC, et al. Recommendations for quantification methods during the performance of a pediatric echocardiogram: areport from the Pediatric Measurements Writing Group of the American Society of Echocardiography Pediatric and Congenital Heart Disease Council [J]. Journal of the American Society of Echocardiography, 2010, 23(5): 465-495.

[0013] Reference [5]: Lan Tingyu, He Wen, Du Lijuan, et al. Clinical study on the evaluation of left atrial function in patients with hypertensive disorder complicating pregnancy using two-dimensional speckle tracking technology [J]. Chinese Journal of Ultrasound in Medicine, 2024, 40(07): 762-766.

[0014] Literature[6]:Farsalinos,Konstantinos E.Daraban,Ana M.Unlu,SerkanThomas,James D.Badano,Luigi P.Voigt,Jens-Uwe.Head-to-Head Comparison of GlobalLongitudinal Strain Measurements among Nine Different Vendors The EACVI / ASEInter-Vendor Comparison Study[J].Journal of the American Society ofEchocardiography:official publication of the American Society ofEchocardiography,2015,28(10).

[0015] Literature [7]: Hallett EJ, Bruza P, Jermyn M, et al. Noise&mottle suppressionmethods for cumulative Cherenkov images of radiation therapy delivery. [J]. Physics in medicine and biology, 2024, 69(22): 225015-225015.

[0016] Literature [8]: Shi R, Leng

[0017] Literature[9]:Perrot V,Garcia D.Back to basics in ultrasound velocimetry:tracking speckles by using a standard PIV algorithm[C] / / 2018IEEEInternationalUltrasonics Symposium(IUS).IEEE,2018:206-212.

[0018] Reference

[10] : Yu Jianfeng, Yan Jiayong, Xie Lijian. Adaptive estimation of template block size for echocardiography block matching speckle tracking algorithm [J]. Software, 2022, 43(08): 4-10.

[0019] Reference

[11] : Zhang Hao. Echocardiographic myocardial wall segment assessment based on optical flow method [D]. Zhejiang Sci-Tech University, 2023. DOI: 10.27786 / d.cnki.gzjlg.2023.001073.

[0020] Literature

[12] : Ahn SS,Ta K,Lu A,Stendahl JC,Sinusas AJ,DuncanJS.Unsupervised Motion Tracking of Left Ventricle in Echocardiography.ProcSPIE Int Soc Opt Eng.2020Feb;11319:113190Z.Epub 2020Mar 16. PMID:32994659; PMCID: PMC7521020.

[0021] Reference

[13] : Wang Tiantian, Wang Hui, Zhu Yanchun, et al. Left myocardial motion tracking in cardiac cine magnetic resonance images based on displacement flow U-Net and variational autoencoder [J]. Acta Physica Sinica, 2021, 70(22): 325-335.

[0022] Literature

[14] : Bernard O, Lalande A, Zotti C, et al. Deep learning techniques for automatic MRI cardiac multi-structures segmentation and diagnosis: is the problem solved? [J].IEEE transactions on medical imaging,2018,37(11):2514-2525.

[0023] Literature

[15] : Parajuli N, Lu A, Ta K, Stendahl J, Boutagy N, Alkhalil I, EberleM, Jeng GS, Zontak M, O'Donnell M, Sinusas AJ, Duncan JS. Flow network tracking for spatiotemporal and periodic point matching: Applied to cardiac motion analysis. Med Image Anal.2019Jul;55:116-135.doi:10.1016 / j.media.2019.04.007.Epub 2019Apr 18.PMID:31055125;PMCID:PMC6939679.

[0024] Literature

[16] : Evain E, Sun Y, Faraz K, et al. Motion estimation by deeplearning in 2D echocardiography: synthetic dataset and validation [J]. IEEE transactions on medical imaging, 2022, 41(8): 1911-1924.

[0025] Literature

[17] : Degerli A, Kiranyaz S, Hamid T, et al. Early myocardialinfarction detection over multi-view echocardiography[J]. Biomedical SignalProcessing and Control, 2024,87:105448. Summary of the Invention

[0026] To solve the above technical problems, the present invention provides an automatic estimation method for the global longitudinal strain index of the adult myocardium based on echocardiography. This method is a robust automatic estimation method for the global longitudinal strain index of the myocardium in adult cardiac ultrasound videos based on myocardial image semantic segmentation technology. It automatically implements framing of the ultrasound video, myocardial region segmentation, myocardial region extraction, calculation of the inner and outer contours and centerline of the myocardium, estimation of the apex position, and estimation of the GLS index. This greatly reduces the impact of errors and subjectivity in manual analysis, and can also overcome the interference caused by image quality to a certain extent.

[0027] The technical solution adopted by the present invention is:

[0028] The method for automatically estimating global longitudinal strain index of adult myocardium based on echocardiography comprises the following steps:

[0029] Step 1: Establish a segmentation model to achieve accurate segmentation of the heart area;

[0030] Step 2: Process the segmented heart region, retain the myocardial region, calculate the endpoints of the inner and outer contours of the myocardium according to the characteristics of the myocardial contour, and obtain the inner and outer contour lines of the myocardial region;

[0031] Step 3: Determine the centerline of the myocardial region;

[0032] Step 4: Divide the center line of the myocardial region into multiple myocardial segments;

[0033] Step 5: Calculate the myocardial segment displacement based on the myocardial segment to obtain the myocardial strain curve.

[0034] In step 1, first, the input ultrasound video is framed to decompose the continuous video into a series of static image frames; then, the framed image frames are named according to frame numbers and arranged in sequence; then, the processed ultrasound video is placed in a pre-trained segmentation model to automatically segment the required heart area.

[0035] In step 1, a TransUNet semantic segmentation model is established by combining transformer and Unet to segment the left ventricle and myocardium.

[0036] In step 2, the position and boundary of the left ventricular myocardial area are determined: the left ventricular myocardium wraps the left ventricle, the inner boundary is the boundary with the left ventricular cavity, and the outer boundary is the outer surface of the heart; the left ventricle is covered and only the myocardial area is retained.

[0037] In step 2, the inner and outer contours of the myocardial area are extracted, including the following steps:

[0038] S2.1: Preprocess the obtained myocardial region segmentation image results:

[0039] Median filtering is used to remove edge jagged effects; then binarization is performed to obtain a segmentation result with smooth edges, which facilitates subsequent contour extraction.

[0040] S2.2: Starting from the upper left corner of the myocardial region segmentation image, scan row by row from left to right, and judge point by point. If the current point is a foreground point and there is at least one background point in its four neighborhoods, it is judged as a contour point and its coordinates are stored. Through this process, the myocardial contour information composed of a series of contour points can be obtained;

[0041] The myocardial contour is operated, and the boundary information of the myocardial contour is used to find the minimum rectangular box that can enclose the contour to narrow the processing range; the minimum rectangular box is divided into two parts, the lower half containing the endpoints of the myocardial contour is retained; in order to accurately locate the coordinates of the left and right endpoints of the inner and outer contours, the retained lower half is divided into two parts again, and the two ends of the myocardial contour are processed separately.

[0042] S2.3: Extract the coordinates of the non-zero pixels in the left contour. Based on all the coordinates, find the maximum x-coordinate a and the maximum y-coordinate b in the left and right regions. Find the coordinates of the points corresponding to these maximum values ​​and store them in the corresponding lists a1_L, b1_L, a1_R, and b1_R. Based on the characteristics of the myocardial contour, the left inner contour endpoint m1 corresponds to the point with the maximum y-coordinate in a1_L, and the left outer contour endpoint n1 corresponds to the point with the minimum x-coordinate in b1_L. Similarly, the right inner contour endpoint m2 corresponds to the point with the minimum x-coordinate in b1_R, and the right outer contour endpoint n2 corresponds to the point with the maximum y-coordinate in a1_R.

[0043] S2.4: After finding the endpoints of the inner and outer contours, use the left endpoints m1 and n1 as the starting point and the right endpoints m2 and n2 as the ending point, and traverse all the points on the contour in order to find the inner and outer contour lines.

[0044] In step 3, the center line of the myocardial region is calculated. The center line is located inside the myocardial region and is not affected by the blurred external contour. The details are as follows:

[0045] First, the inner and outer contours are evenly divided into multiple key points to simplify the centerline calculation process; then, the distances between the corresponding two key points on the inner and outer contours are calculated one by one in sequence, and the median of these distances is taken as the basis for the corresponding position points on the centerline, so that the coordinates of the corresponding multiple key points on the centerline can be accurately calculated; finally, these calculated points on the centerline are connected in sequence to find the centerline.

[0046] In step 4, a 16-segment segmentation method is selected to divide the myocardial area.

[0047] In step 4, it can be known that the direction of the apex is far away from the center of gravity according to the characteristics of the myocardial region, so the direction of the apex is found by the center of gravity of the myocardial region.

[0048] Determine the distance between the center of gravity and the inner and outer contours within a certain range, find points a and b on the inner and outer contours in the apex direction, and then find the midpoint m between points a and b. However, due to the problem of point division, the midpoint m may not be on the center line of the found myocardial area. Further determine point m and find the midpoint on the center line of the myocardial area that is closest to point m. Use midpoint as the apex point to facilitate subsequent myocardial segment division.

[0049] The center line of the myocardial area is divided into two parts through the apex point, and then these two parts are divided into three segments respectively, and different colors are given to distinguish the segments.

[0050] In step 5, myocardial strain is calculated by calculating the displacement change of each myocardial segment over a complete cardiac cycle, using the end-diastolic frame as the reference frame. By comparing the reference frame with the myocardial position at other times during the cardiac cycle, the strain value of each myocardial segment at different stages, i.e., the relative change in its length, can be accurately calculated.

[0051] The longitudinal ventricular length L at each time step t is calculated using the centerline of the myocardial region to estimate the Lagrangian strain. The strain calculation formula is shown in formula (1):

[0052]

[0053] In formula (1), S L is the strain of each myocardial segment, L(t) is the longitudinal length at a given time t, and L(t0) is the longitudinal length at time t0. Global longitudinal strain (GLS) refers to the relative change in left ventricular myocardial length between end-diastole and end-systole. The calculation formula for global longitudinal strain (GLS) is shown in formula (2):

[0054]

[0055] In formula (2), MLs is the myocardial length at the end of systole, and MLd is the myocardial length at the end of diastole. As the heart contracts, the myocardium shortens, and during the diastole phase, the myocardium stretches relatively, resulting in the myocardial length MLs at the end of systole being smaller than the myocardial length MLd at the end of diastole, thus making the global longitudinal strain GLS a negative value.

[0056] The present invention provides an automatic method for estimating global longitudinal strain in adult myocardium based on echocardiography. The technical effects are as follows: 1) To overcome the impact of factors such as blurred cardiac edges, missing partial structures, strong speckle noise, and irregular deformation of the overall cardiac contour on myocardial region extraction in echocardiography, the present invention proposes a method for extracting myocardial contours based on segmentation results. This method uses deep learning-based segmentation technology to accurately locate and segment the cardiac region in ultrasound images, ensuring the accuracy and integrity of the cardiac contour. The segmented cardiac region is then analyzed to extract the myocardial region. An endpoint detection algorithm for the inner and outer contours of the myocardium is then designed based on the characteristics of the myocardial region. This algorithm can automatically and accurately calculate the inner and outer contours of the myocardial region, laying a solid foundation for subsequent centerline calculation. This solution can, to a certain extent, ignore the interference of echocardiographic quality, thereby improving the accuracy of myocardial contour extraction in echocardiography. 2) To address the problem of poor ultrasound image quality caused by interference from various factors in echocardiography, the present invention proposes a stable and reliable method for extracting myocardial contours. This method uses deep learning-based semantic segmentation technology to accurately locate and segment the cardiac region in ultrasound images, ensuring the accuracy and integrity of the cardiac contour. The segmented heart region is then analyzed to extract the myocardial region. Based on the characteristics of the myocardium, an endpoint detection algorithm for the inner and outer contours of the myocardium is designed, which automatically and accurately calculates the inner and outer contours of the myocardium. This method reduces the impact of ultrasound image quality on the myocardial region and accurately extracts the inner and outer contours of the myocardium.

[0057] 3) In order to solve the problem of unreliable myocardial centerline extraction in adult ultrasound cardiac images using the PIV method, the present invention proposes a method for automatically calculating the centerline based on the inner and outer contours. This method directly uses the inner and outer contours accurately extracted from the image segmentation results to calculate a smooth and accurate myocardial centerline. Compared with the PIV method, this solution significantly reduces human intervention and improves work efficiency, because PIV relies on spatial cross-correlation and block matching technology. These steps are not only complex but also susceptible to common problems such as blurred heart edges, missing structures and high noise in ultrasound images, resulting in unstable center point positioning and even deviation from the myocardial contour. The method of the present invention effectively avoids these problems by directly analyzing the contour lines, ensuring the accuracy and stability of the centerline calculation. To ensure the accuracy of subsequent myocardial segmentation, the present invention also proposes an apex position estimation algorithm that can quickly and reliably find the apex position based on the inner and outer contours and their center of gravity.

[0058] 4) In order to further evaluate the myocardial motion, the present invention analyzes the myocardial segment motion in the echocardiogram, automatically calculates the displacement of each myocardial segment, and obtains the GLS change curve. This method takes the myocardial centerline as the core and the apex point as the reference point, and adopts a 16-segment segmentation model of the myocardium to divide the centerline into multiple segments. Subsequently, by dynamically tracking the centerlines of these segments, the motion trajectory of each segment of the myocardium can be accurately calculated, and then the GLS change curve can be drawn. In order to eliminate the fluctuations of the GLS curve caused by various interference factors, the present invention has found the most suitable GLS curve filtering method for the present invention through a large number of experimental comparisons. By filtering the GLS change curve, the changes in the myocardial motion state are made clearer and more complete. This method not only provides a clinical intuitive display of the myocardial motion state, but also greatly enhances the accuracy and reliability of diagnosis, and provides strong support for doctors to formulate more personalized and scientific treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0060] Figure 1 This is a low-quality echocardiogram (partial sample image)

[0061] Figure 2 Flowchart of the algorithm for automatic estimation of global longitudinal strain indices of myocardial blood using echocardiography in adults.

[0062] Figure 3 Obtain a schematic diagram for the myocardial region.

[0063] Figure 4 This is a flow chart of the algorithm for extracting the inner and outer contours of the myocardium.

[0064] Figure 5 This is an example of the inner and outer contours of the myocardium.

[0065] Figure 6 This is a diagram of the center line calculation process.

[0066] Figure 7(a) is a schematic diagram of a two-chamber heart and a four-chamber heart using the 16-segment segmentation method;

[0067] Figure 7(b) is a schematic diagram of a two-chamber heart and a four-chamber heart using the 17-segment method.

[0068] Figure 8 Schematic diagram of the apex orientation at different tilt angles.

[0069] Figure 9 This is an example of a myocardial segmentation diagram.

[0070] Figure 10 This is the GLS change curve.

[0071] Figure 11 This is the GLS calculation interface in the software supporting commercial ultrasound equipment.

[0072] Figure 12 Comparison of myocardial strain curves under different filtering methods (partial sample images).

[0073] Figure 13 This is the GLS change filter curve.

[0074] Figure 14 Comparison diagram of the center points of the PIV method and the segmentation-based results (partial sample images).

[0075] Figure 15 Comparison diagram of the center points of the low-quality PIV method and the segmentation-based method (partial sample image). DETAILED DESCRIPTION

[0076] To overcome the effects of factors such as blurred heart edges, missing structures, strong speckle noise, and irregular deformation of the heart's overall contour on myocardial region extraction in echocardiography, the present invention proposes a method for extracting myocardial contours based on segmentation results. This solution uses deep learning-based segmentation technology to accurately locate and segment the heart region in ultrasound images, ensuring the accuracy and integrity of the heart contour. The segmented heart region is then analyzed to extract the myocardial portion. An endpoint detection algorithm for the inner and outer contours of the myocardium is then designed based on the characteristics of the myocardium. This algorithm can automatically and accurately calculate the inner and outer contours of the myocardium, laying a solid foundation for subsequent centerline calculations. This solution can, to a certain extent, ignore the interference of echocardiographic quality and improve the accuracy of myocardial contour extraction in echocardiography.

[0077] To address the problem that speckle tracking technology based on PIV cannot reliably and fully automatically track the centerline of the myocardial region in electrocardiograms, this paper proposes an algorithm for extracting the myocardial centerline based on myocardial segmentation results. By analyzing the inner and outer contours of the myocardium, the myocardial centerline can be accurately calculated and extracted. Furthermore, an algorithm for estimating the cardiac apex position is proposed. This method overcomes the positioning errors caused by PIV technology due to factors such as blurred boundaries and noise interference, ensuring the accuracy of centerline and apex positioning, and providing a foundation for subsequent GLS calculations.

[0078] To further assess myocardial motion and better assist clinical diagnosis, the present invention quantitatively analyzes myocardial segmental motion in echocardiography and proposes a method for analyzing myocardial segmental motion based on the myocardial centerline. This method automatically calculates the GLS curve, reducing the need for ultrasound physicians to perform manual operations and their reliance on their expertise and experience, thereby improving operational efficiency and reducing the risk of human error.

[0079] An automatic estimation algorithm for the global longitudinal strain index of the adult myocardium based on echocardiography. The algorithm uses a combination of transformer and Unet, and trains the network model through a self-built data set to achieve accurate segmentation of the heart area. Then, the segmented heart area is processed, the myocardial area is retained, and the endpoints of the inner and outer contours of the myocardium are calculated according to the characteristics of the myocardial contour. The inner and outer contour lines of the myocardial area are obtained, and then the center line of the myocardial area is determined. Next, the center line of the myocardial area is divided, and the myocardial area is divided into six myocardial segments according to the standard 16-segment segmentation method. Finally, the myocardial segment displacement is calculated in units of myocardial segments to obtain the GLS strain curve. The overall process is as follows Figure 1 shown.

[0080] 1. Reliable segmentation of the heart region:

[0081] The premise for automatically estimating the overall longitudinal strain index of the myocardium in adult cardiac ultrasound images is to automatically and accurately segment the specific cardiac region in the ultrasound cross-sectional image. The present invention uses the TransUNet semantic segmentation model to segment the left ventricle and myocardium. The purpose of this is to help the segmentation model learn the characteristics of the myocardial region more quickly by adding more supervision information and contrast information, so as to achieve better segmentation. The TransUNet semantic segmentation model is described in the literature

[18] . The present invention is not limited to the TransUNet model, and any network model that can meet the requirements of accurate segmentation of the ventricle and myocardium is applicable.

[0082] First, the input ultrasound video is framed, breaking the continuous video into a series of static image frames. The framed ultrasound images are then named and arranged sequentially according to frame numbers to ensure data consistency and traceability, facilitating subsequent automated image processing. The processed video is then fed into a pre-trained segmentation model to automatically segment the desired cardiac region.

[0083] 2. Obtaining left ventricular myocardium:

[0084] To calculate the GLS index for the left ventricular myocardium, the location and boundaries of the left ventricular myocardium must be clearly defined. The left ventricular myocardium surrounds the left ventricle, with the inner boundary being the boundary with the left ventricular chamber and the outer boundary being the outer surface of the heart.

[0085] Process the segmentation results of the first step, the segmentation results are as follows Figure 3 As shown in sub-figure a of , the left ventricle is covered and only the myocardial area is retained, as shown in Figure 3As shown in sub-figure b, after the myocardial area is segmented, a more detailed analysis of the myocardial area will be performed. The present invention simultaneously segments the myocardium and ventricles to allow the model to obtain more supervisory information during training, ensuring the reliability of myocardial segmentation. Ventricular segmentation is an auxiliary operation. Multiple experiments have found that this approach is more effective than simply segmenting the myocardium. On the other hand, the curve of the change in ventricular area can provide reference information for the subsequent selection of the maximum diastole and maximum systole of the heart in the present invention.

[0086] 3. Extraction of inner and outer contours:

[0087] After obtaining the myocardial region, the myocardial contour is used to extract the inner and outer contours of the myocardium. The algorithm for extracting the inner and outer contours of the myocardium is as follows:

[0088]

[0089] Figure 4 The flowchart of the above algorithm is shown in Figure 1. First, the myocardial segmentation image is preprocessed by using a median filter to remove the edge jagged effect, and then binarized to obtain a segmentation result with smooth edges, which is convenient for subsequent contour extraction.

[0090] Next, starting from the upper left corner of the image, the image is scanned row by row from left to right, making point-by-point judgments. If the current point is a foreground point and has at least one background point in its four neighborhoods, it is considered a contour point and its coordinates are stored. This process yields myocardial contour information, consisting of a series of points. The myocardial contour is then manipulated, using its boundary information to find the minimum rectangle that encloses it, thus narrowing the processing range. The minimum rectangle is then divided into two parts, retaining the lower half containing the myocardial endpoints. To precisely locate the left and right endpoints of the inner and outer contours, the retained lower half is again divided equally into left and right halves, processing the two ends of the myocardium separately.

[0091] Next, extract the coordinates of the non-zero pixels in the left contour. Based on all these coordinates, find the maximum x-coordinate a and maximum y-coordinate b in the left and right regions. The coordinates of the points corresponding to these maximum values ​​are stored in the corresponding lists a1_L, b1_L, a1_R, and b1_R. Based on the characteristics of the myocardial contour, the left inner contour endpoint m1 corresponds to the point with the maximum y-coordinate in a1_L, and the outer contour endpoint n1 corresponds to the point with the minimum x-coordinate in b1_L. Similarly, the right inner contour endpoint m2 corresponds to the point with the minimum x-coordinate in b1_R, and the outer contour endpoint n2 corresponds to the point with the maximum y-coordinate in a1_R.

[0092] Finally, after finding the endpoints of the inner and outer contours, we use the left endpoints m1 and n1 as the starting point and the right endpoints m2 and n2 as the ending point, and traverse all the points on the contour in order to find the inner and outer contours. Figure 5 Results of finding inner and outer contours in 2CH (two-chamber view) and 4CH (four-chamber view) images.

[0093] 4. Centerline calculation:

[0094] Given the inherent characteristics of ultrasound images, such as noise, artifacts, complex reflections from tissue interfaces, and the upward and downward movement of the body due to breathing, the myocardium may not be fully displayed. In particular, the outer contour of the myocardial region can be blurred. This uncertainty limits the reliability of accurate analysis based solely on the inner and outer contours. To overcome this problem, the present invention adopts a more reliable approach: calculating the centerline of the myocardial region. This centerline lies within the myocardial region and is not affected by the blurred outer contours, thus ensuring its accuracy and reliability in subsequent analysis.

[0095] Before calculating the center line, the present invention first divides the inner and outer contours into 200 key points each evenly to simplify the center line calculation process. Subsequently, the distances between the two corresponding key points on the inner and outer contours are calculated one by one in sequence, and the median of these distances is taken as the basis for the corresponding position points on the center line. Through this method, we can accurately calculate the coordinates of the 200 corresponding points on the center line. Finally, these calculated points on the center line are connected in sequence to find the center line. This process not only ensures the accuracy of the center line calculation, but also improves the efficiency and effect of the overall processing. The calculation process is as follows Figure 6 shown.

[0096] 5. Myocardial segment division:

[0097] Among the many methods for segmenting the left ventricular myocardium, the commonly used segmentation methods include the 16-segment method and the 17-segment method. The 16-segment method was proposed by the American Society of Echocardiography (ASE) in 1989. As recorded in the literature

[19] , the left ventricular myocardium is divided into three equal-height myocardial rings: the basal ring, the middle ring, and the apical ring. Each ring occupies one-third of the length of the left ventricle. This balanced division method not only simplifies the analysis process but also ensures the comprehensiveness and accuracy of the evaluation. In the analysis of the apical two-chamber heart and apical four-chamber heart sections, the 16-segment method divides the left ventricular myocardial area in these two sections into six segments, as shown in Figure 7(a). In contrast, the 17-segment method proposed by the American Heart Association (AHA) is to capture the myocardial changes in the apical area in more detail, as recorded in the literature

[20] . On the basis of the 16-segment method, the apical cap is added as an independent segment, as shown in Figure 7(b). Considering that the present invention focuses on the assessment of overall myocardial motion, there is no need to consider changes in the apical cap region separately. Furthermore, the 16-segment method is simple to use and its assessment accuracy is sufficient for most clinical needs. Therefore, the present invention selects the 16-segment method for myocardial region segmentation.

[0098] The key to myocardial segmentation is to accurately determine the direction of the cardiac apex. Due to the inherent uncertainty of cardiac ultrasound images, the cardiac region may be tilted to a certain extent, making it impossible to directly determine the direction of the cardiac apex by bisecting the myocardial region or the central axis. The algorithm for extracting the cardiac apex point is as follows:

[0099]

[0100]

[0101] Based on the characteristics of the myocardial region, it can be known that the direction of the apex is usually far away from the center of gravity. Therefore, the present invention proposes to find the direction of the apex through the center of gravity of the myocardial region. By determining the distance between the center of gravity and the inner and outer contours within a certain range, points a and b on the inner and outer contours in the direction of the apex are found, and then the midpoint m between points a and b is found. However, due to the problem of point division, point m may not necessarily be on the found myocardial centerline. Further point judgment is required to find the point midpoint on the myocardial centerline closest to point m. Point midpoint is used as the apex point to facilitate the subsequent myocardial segment division. Figure 8 Schematic diagram showing some of the cardiac apex orientations at different tilt angles.

[0102] In order to facilitate the subsequent GLS curve calculation, the present invention operates on the 200 points on the myocardial centerline determined in the previous processing process, divides the myocardial centerline into two parts by the apex point, and then divides these two parts into three segments respectively, and assigns different colors to distinguish each segment, such as Figure 9 shown.

[0103] 6. Strain curve calculation:

[0104] Myocardial strain is calculated by calculating the displacement change of each myocardial segment over a complete cardiac cycle. This method uses the end-diastolic frame as the reference (i.e., time t=0). By comparing the reference frame with the myocardial position at other times of the cardiac cycle (e.g., systole), the strain value of each myocardial segment at different stages can be accurately calculated, that is, the relative change in its length.

[0105] In the present invention, the longitudinal ventricular length L of each time step t is calculated using the myocardial centerline to estimate the Lagrangian strain. The strain calculation formula is shown in formula (1):

[0106] S L =L(t)-L(t0)×100%

[0107] L(t0)(1);

[0108] In formula (1), S L is the strain of each myocardial segment, L(t) is the longitudinal length at a given time t, and L(t0) is the longitudinal length at time t0. Global longitudinal strain (GLS) refers to the relative change in left ventricular myocardial length between end-diastole and end-systole. The GLS calculation formula is shown in formula (2):

[0109]

[0110] In formula (2), MLs is the myocardial length at end-systole, and MLd is the myocardial length at end-diastole. As the heart contracts, the myocardium shortens, and during diastole, the myocardium stretches relatively. This causes the myocardial length MLs at end-systole to be smaller than the myocardial length MLd at end-diastole, resulting in a negative global longitudinal strain GLS.

[0111] When the ultrasound probe acquires data, the heartbeat of the person being collected, the heartbeat of the collecting physician himself, and the changes in the hand position will cause the size of the myocardial and ventricular areas in the obtained cardiac cross-section image to change. At the same time, the segmentation algorithm itself also has a certain degree of instability. These factors work together to make the GLS curve obtained according to the segmentation results contain inevitable noise, and these noises have a certain degree of randomness. In addition, there may be certain differences in the segmentation results of two adjacent frames of images, resulting in errors in the extracted myocardial contour. This error may cause the displacement or shape of the myocardial contour to be inconsistent, thereby affecting the final GLS calculation results and causing irregular fluctuations in the GLS change curve. This problem also exists in commercial ultrasound equipment products based on speckle tracking technology, so filtering functions will be added to the supporting software of these devices. For example Figure 11 As shown, Figure 11Automated Cardiac Motion Quantification A.I. (aCMQ) provided by Philips A.I. ) technical introduction, aCMQ A.I. Three options for quantification smoothness are provided, and by adjusting the smoothness, clinicians can improve the accuracy and reliability of strain measurements, as documented in

[23] .

[0112] In order to eliminate interference, the present invention adds filtering to the myocardial strain curve for smoothing. Common signal processing filtering methods include average moving filter, weighted average moving filter, Kalman filter, wavelet filter and Savitzky-Golay filter, etc. As recorded in the literature

[24] , these methods have their own advantages in removing noise and maintaining signal characteristics. Average moving filter smoothes the signal by calculating the average value of pixels in the neighborhood, which is suitable for removing simple noise; weighted average moving filter assigns different weights to pixels in the neighborhood when calculating the average value, which can better retain the signal details; Kalman filter can optimize the signal estimation in the presence of noise and uncertainty by establishing a dynamic system model, which is suitable for signal smoothing and prediction; wavelet filter effectively removes high-frequency noise and retains low-frequency components by decomposing the signal into multiple frequency levels; Savitzky-Golay filter uses polynomial fitting method for smoothing, which can not only remove noise, but also retain the derivative information of the signal to a certain extent, thereby better reflecting the changing trend of the signal. As Figure 12 As shown in Figure 3, the myocardial strain curves after applying different filtering methods show different smoothing effects.

[0113] By comparing various filtering results, the present invention found that weighted average moving average filtering can effectively preserve the curve's fluctuation characteristics while smoothing the curve, especially in subtle fluctuations. It can effectively reduce noise without affecting the main motion characteristics of the myocardial strain curve. While other filtering methods can smooth the myocardial strain curve to a certain extent, their effectiveness is still insufficient in areas with more dramatic fluctuations. For example, while wavelet filtering is advantageous in removing high-frequency noise, its overall smoothing effect is poor, and low-frequency noise fluctuations remain noticeable after smoothing. Simple moving average filtering and Savitzky-Golay filtering may over-suppress fluctuations during the smoothing process, resulting in a loss of curve detail. This can easily overlook the actual dynamic changes in motion, especially during critical moments when cardiac motion is rapidly changing. While Kalman filtering can effectively preserve signal trends, its denoising effect is still less than ideal. It lacks processing power for areas with significant fluctuations and cannot completely eliminate the effects of noise. Weighted average moving average filtering can effectively remove noise while fully preserving the key fluctuation characteristics of the myocardial strain curve, thereby accurately reflecting myocardial motion and ensuring signal accuracy and validity. Therefore, the present invention uses weighted average moving average filtering to process the GLS variation curve.

[0114] Figure 13 The figure shows the comparison of the GLS results before and after filtering. It can be seen that the filtered GLS effectively reduces noise interference while retaining the effective information of the original curve to the greatest extent.

[0115] After testing 175 ultrasound videos, the present invention was able to calculate the GLS (global longitudinal strain) change curve with a success rate of 96%. After inspection, the videos that failed to successfully calculate the GLS change curve were due to poor video quality, which resulted in inaccurate segmentation results of the myocardial area, thereby affecting the calculation of the GLS change curve. Among the videos that successfully calculated the GLS change curve, 90.48% of the GLS change curves were within the normal range. After careful verification, it was found that the segmentation results of those videos whose GLS change curves deviated from the normal range were accurate, and doctors could further use other examinations to determine whether the patient's myocardial function was abnormal. Therefore, compared with the deep learning method, the method proposed in the present invention is not only more applicable, but also reduces the time for model training, improves work efficiency, and also effectively reduces the errors that may occur when doctors manually mark the myocardial area.

[0116] The present invention provides an automatic estimation method for adult myocardial global longitudinal strain index based on echocardiography, which has the following characteristics:

[0117] (1). The present invention proposes a stable and reliable myocardial contour extraction method. First, image segmentation technology is used to accurately locate and segment the heart area in the ultrasound image to ensure the accuracy and integrity of the heart contour. Then, the segmented heart area is analyzed to extract the myocardial part. Then, based on the characteristics of the myocardium, an endpoint detection algorithm for the inner and outer contours of the myocardium is designed. The inner and outer contours of the myocardium can be automatically and accurately calculated, laying a solid foundation for the subsequent centerline calculation. This scheme can largely resist the interference caused by the quality factors of the echocardiogram and improve the accuracy of myocardial contour extraction in the echocardiogram.

[0118] (2). The present invention addresses the problem of unreliable automatic extraction of the myocardial centerline in adult ultrasonic cardiac images using the PIV-based speckle tracking method, and proposes a method for automatically calculating the centerline based on the inner and outer contours. This solution processes the segmentation results, extracts the inner and outer contours of the myocardium, and then calculates the inner and outer contours to obtain the centerline of the myocardium. The PIV-based speckle tracking method uses the properties of spatial cross-correlation to calculate the position of the centerline. Before using PIV technology to track the myocardial centerline, it is necessary to obtain the coordinates of a series of key points on the centerline of the first frame image through the segmentation results, define a small image block around the key point, and use a search algorithm to match around the corresponding key point in the next frame to find the most similar image area. Based on the block matching results, the displacement vector of each key point between consecutive frames is calculated to find the optimal centerline position in the current frame. Figure 14 As shown, Figure 14 The red dots are the points tracked by the PIV method, and the green dots are the points calculated based on the inner and outer contours in the implementation of the present invention. Figure 14 It can be seen that the center lines extracted by the method of automatically calculating the myocardial center line based on the inner and outer contours and the method based on PIV are highly consistent, which further confirms the effectiveness of the method of automatically calculating the center line based on the inner and outer contours.

[0119] However, before using the PIV method, the centerline of the first frame still needs to be manually annotated by eye to determine the coordinates of key points. This process requires human intervention, increasing workload and reducing efficiency. Furthermore, because ultrasound images may have blurred cardiac edges, missing some structures, and high noise levels, these issues significantly affect the PIV algorithm's ability to accurately track and locate the optimal point in subsequent frames. The center point found is prone to significant displacement deviations and deviations from the centerline, affecting the stability of the results. Furthermore, adding other post-processing can introduce uncertainty, further complicating the analysis.

[0120] On the other hand, even when evaluating young and healthy subjects, image quality limitations result in at least 6% of 2D Speckle Tracking Echocardiogram (2DSTE) being at risk of failure during myocardial function assessment, as documented in the literature

[22] .

[0121] In contrast, the method of automatically calculating the centerline based on the inner and outer contours can not only reduce the error caused by human intervention, but also accurately segment the myocardial area even in poor image conditions after segmentation model training, ensuring the accuracy of centerline calculation. Figure 15 As shown, even when image quality is poor and the PIV method cannot accurately identify the centerline, the segmentation-based method can still reliably calculate the centerline, which is extremely important for the subsequent calculation of the GLS curve. To ensure the accuracy of subsequent myocardial segmentation, the present invention also proposes an apex position estimation algorithm that can quickly and reliably locate the apex based on the inner and outer contours and their center of gravity.

[0122] (3) In order to evaluate myocardial motion, the present invention will quantitatively analyze the myocardial segment motion in echocardiography and propose a method for analyzing the motion of each myocardial segment based on the myocardial centerline, aiming to better assist clinical diagnosis. First, the myocardial centerline is divided according to the 16-segment method. After obtaining each segment, the centerline is dynamically tracked to calculate the motion curve of the myocardial segment. Under normal physiological conditions, the GLS value is roughly maintained at around -20%. When the absolute value of GLS deviates from the normal range, especially below 15%, it may indicate a significant decline in the patient's cardiac function and the risk of poor prognosis. Timely attention and interpretation of the GLS value are of great value for clinical decision-making.

[0123] References:

[0124] Reference

[18] : Chen J, Lu Y, Yu Q, et al. Transunet: Transformers make strong encoders for medical image segmentation[J]. arXiv preprint arXiv:2102.04306, 2021. Reference

[19] : Schiller N B, Shah P M, Crawford M, et al. Recommendations for quantitation of the left ventricle by two-dimensional echocardiography[J]. Journal of the American Society of Echocardiography, 1989, 2(5):358-367. Reference

[20] : American Heart Association Writing Group on Myocardial Segmentation and Registration for Cardiac Imaging:, Cerqueira M D, Weissman N J, et al. Standardized myocardial segmentation and nomenclature for tomographic imaging of the heart: a statement for healthcare professionals from the Cardiac Imaging Committee of the Council on Clinical Cardiology of the American Heart Association[J]. Circulation, 2002, 105(4):539-542.

[0125] Reference

[21] : Lang R M, Bierig M, Devereux R B, et al. Recommendations for chamber quantification: a report from the American Society of Echocardiography’s Guidelines and Standards Committee and the Chamber Quantification Writing Group, developed in conjunction with the European Association of Echocardiography, a branch of the European Society of Cardiology [J]. Journal of the American society of echocardiography, 2005, 18(12): 1440 - 1463.

[0126] Reference

[22] : Blessberger H, Binder T. Two dimensional speckle tracking echocardiography: basic principles [J]. Heart, 2010, 96(9): 716 - 722.

[0127] Reference

[23] : Quantification A A C M. Measuring cardiac tissue motion and strain [J]. Reference

[24] : Schafer R W. What is a savitzky - golay filter?[lecture notes][J]. IEEE Signal processing magazine, 2011, 28(4): 111 - 117.

Claims

1. An automatic method for estimating global longitudinal strain of adult myocardium based on echocardiography, characterized by The following steps are involved: Step 1: Establish a segmentation model to achieve accurate segmentation of the heart area; Step 2: Process the segmented heart region, retain the myocardial region, calculate the endpoints of the inner and outer contours of the myocardium according to the characteristics of the myocardial contour, and obtain the inner and outer contour lines of the myocardial region; Step 3: Determine the centerline of the myocardial region; Step 4: Divide the center line of the myocardial region into multiple myocardial segments; Step 5: Calculate the myocardial segment displacement based on the myocardial segment to obtain the myocardial strain curve.

2. The method for automatically estimating adult myocardial global longitudinal strain index based on echocardiography according to claim 1, characterized in that: In step 1, first, the input ultrasound video is framed to decompose the continuous video into a series of static image frames; then, the framed image frames are named according to frame numbers and arranged in sequence; then, the processed ultrasound video is placed in a pre-trained segmentation model to automatically segment the required heart area.

3. The method for automatically estimating adult myocardial global longitudinal strain based on echocardiography according to claim 2, characterized in that: The TransUNet semantic segmentation model is established by combining transformer and Unet to segment the left ventricle and myocardium.

4. The method for automatically estimating adult myocardial global longitudinal strain index based on echocardiography according to claim 1, characterized in that: In step 2, the position and boundary of the left ventricular myocardial area are determined: the left ventricular myocardium wraps the left ventricle, the inner boundary is the boundary with the left ventricular cavity, and the outer boundary is the outer surface of the heart; the left ventricle is covered and only the myocardial area is retained.

5. The method for automatically estimating adult myocardial global longitudinal strain index based on echocardiography according to claim 4, characterized in that: In step 2, the inner and outer contours of the myocardial area are extracted, including the following steps: S2.1: Preprocess the obtained myocardial region segmentation image results: Median filtering is used to remove edge jagged effects; then binarization is performed to obtain a segmentation result with smooth edges, which facilitates subsequent contour extraction. S2.2: Starting from the upper left corner of the myocardial region segmentation image, scan row by row from left to right, and judge point by point. If the current point is a foreground point and there is at least one background point in its four neighborhoods, it is judged as a contour point and its coordinates are stored. Through this process, the myocardial contour information composed of a series of contour points can be obtained; The myocardial contour is operated on, and the boundary information of the myocardial contour is used to find the minimum rectangular box that can enclose the contour, thereby reducing the processing range. The minimum rectangular box is divided into two parts, the lower half containing the endpoints of the myocardial contour is retained. In order to accurately locate the coordinates of the left and right endpoints of the inner and outer contours, the retained lower half is divided into two equal parts again, and the two ends of the myocardial contour are processed separately. S2.3: Extract the coordinates of the non-zero pixels in the left contour. Based on all the coordinates, find the maximum x-coordinate a and the maximum y-coordinate b in the left and right regions. Find the coordinates of the points corresponding to these maximum values ​​and store them in the corresponding lists a1_L, b1_L, a1_R, and b1_R. Based on the characteristics of the myocardial contour, the left inner contour endpoint m1 is the point with the maximum y-coordinate in a1_L, and the left outer contour endpoint n1 is the point with the minimum x-coordinate in b1_L. Similarly, the right inner contour endpoint m2 is the point with the minimum x-coordinate in b1_R, and the right outer contour endpoint n2 is the point with the maximum y-coordinate in a1_R. S2.4: After finding the endpoints of the inner and outer contours, use the left endpoints m1 and n1 as the starting point and the right endpoints m2 and n2 as the ending point, and traverse all the points on the contour in order to find the inner and outer contour lines.

6. The method for automatically estimating adult myocardial global longitudinal strain index based on echocardiography according to claim 1, characterized in that: In step 3, the center line of the myocardial region is calculated. The center line is located inside the myocardial region and is not affected by the fuzzy external contour. Specifically, the calculation is as follows: First, the inner and outer contours are evenly divided into multiple key points to simplify the centerline calculation process; then, the distances between the corresponding two key points on the inner and outer contours are calculated one by one in sequence, and the median of these distances is taken as the basis for the corresponding position points on the centerline, so that the coordinates of the corresponding multiple key points on the centerline can be accurately calculated; finally, these calculated points on the centerline are connected in sequence to find the centerline.

7. The method for automatically estimating adult myocardial global longitudinal strain index based on echocardiography according to claim 1, characterized in that: In step 4, a 16-segment segmentation method is selected to divide the myocardial area.

8. The method for automatically estimating adult myocardial global longitudinal strain index based on echocardiography according to claim 8, characterized in that: In step 4, it can be known that the direction of the apex is far away from the center of gravity according to the characteristics of the myocardial region. Therefore, the direction of the apex is found by the center of gravity of the myocardial region, as follows: Determine the distance between the center of gravity and the inner and outer contours within a certain range, find points a and b on the inner and outer contours in the apex direction, and then find the midpoint m between points a and b. However, due to the problem of point division, the midpoint m may not be on the center line of the found myocardial area. Further determine point m and find the midpoint on the center line of the myocardial area that is closest to point m. Use midpoint as the apex point to facilitate subsequent myocardial segment division. The center line of the myocardial area is divided into two parts through the apex point, and then these two parts are divided into three segments respectively, and different colors are given to distinguish the segments.

9. The method for automatically estimating adult myocardial global longitudinal strain index based on echocardiography according to claim 1, characterized in that: In step 5, myocardial strain is obtained by calculating the displacement change of each myocardial segment during a complete cardiac cycle, and the frame at the end of diastole is selected as the reference frame. By comparing the reference frame with the myocardial position at other periods of the cardiac cycle, the strain value of each myocardial segment at different stages, that is, the relative change in its length, can be accurately calculated.

10. The method for automatically estimating adult myocardial global longitudinal strain index based on echocardiography according to claim 9, characterized in that: The longitudinal ventricular length L at each time step t is calculated using the centerline of the myocardial region to estimate the Lagrangian strain. The strain calculation formula is shown in formula (1): In formula (1), S L is the strain of each myocardial segment, L(t) is the longitudinal length at a given time t, and L(t0) is the longitudinal length at time t0; global longitudinal strain GLS refers to the relative change in left ventricular myocardial length between end-diastole and end-systole. The calculation formula of global longitudinal strain GLS is shown in formula (2): In formula (2), MLs is the myocardial length at the end of systole, and MLd is the myocardial length at the end of diastole. As the heart contracts, the myocardium shortens, and during the diastole phase, the myocardium stretches relatively, resulting in the myocardial length MLs at the end of systole being smaller than the myocardial length MLd at the end of diastole, thus making the global longitudinal strain GLS a negative value.

Images