Automatic analysis method for beat track of engineered heart tissue based on image recognition algorithm

Abstract

The invention relates to the technical field of medical image processing, in particular to an engineered heart tissue pulsation trajectory automatic analysis method based on an image recognition algorithm, which comprises the following steps: S1, multi-modal image fusion: performing space-time registration and feature fusion on acquired multi-modal heart images to generate a fused image sequence; s2, cardiac muscle tissue segmentation: outputting a cardiac muscle tissue segmentation result with a timestamp; s3, motion track modeling: generating three-dimensional track point cloud data in a pulsation period; s4, feature parameter extraction: performing spatial-temporal feature analysis on the track point cloud data, and extracting multi-dimensional motion parameters; and S5, heterogeneity atlas generation: generating a cardiac pulse heterogeneity atlas according to the multi-dimensional motion parameters. According to the method, automatic analysis of the cardiac pulse track and generation of the heterogeneity atlas based on the multi-modal image and space-time modeling are realized, and the precision and the intelligent level of cardiac motion anomaly recognition are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to an automatic analysis method for engineered cardiac tissue pulsation trajectories based on an image recognition algorithm. Background Art

[0002] With the increasing prevalence of cardiovascular diseases, clinical practice has placed higher demands on the accurate quantification and abnormality detection of myocardial functional status. Traditional cardiac function assessment methods mainly rely on a single modality such as echocardiography or cardiac MRI, and perform qualitative analysis through manual interpretation of myocardial thickness changes and ventricular contraction indicators. These methods are difficult to reveal the fine-grained motion characteristics of the myocardium in spatial and temporal dimensions. In recent years, the application of image recognition and three-dimensional modeling technology in medical imaging has gradually deepened, providing a feasible path for the automatic tracking and analysis of the entire myocardial motion process.

[0003] Existing methods still have obvious shortcomings in terms of accuracy and automation. On the one hand, there are problems such as time asynchrony, spatial mismatch, and low information utilization between multimodal cardiac images, which leads to motion modeling relying on single-modal data and lacking a comprehensive picture. On the other hand, automatic segmentation and motion vector estimation of myocardial tissue often adopt a static framing strategy, which cannot stably restore the continuous motion trajectory of the myocardium. In addition, traditional evaluations are mostly based on overall indicators and lack structured recognition and graphical expression of heterogeneous motion patterns in local myocardial regions, which limits the clinical ability to conduct detailed analysis of lesion localization and functional zoning. Summary of the Invention

[0004] The present invention provides an automatic analysis method for the pulsation trajectory of engineered cardiac tissue based on an image recognition algorithm. It not only realizes the full-process automation processing of the structure-motion-assessment trinity, but also significantly improves the accuracy and visualization capability of identifying myocardial motion abnormalities, providing a high-resolution structure-function fusion reference for clinical diagnosis and treatment decisions.

[0005] The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm comprises the following steps: S1, multimodal image fusion: Perform spatiotemporal registration and feature fusion on acquired multimodal cardiac images to generate a fused image sequence. Multimodal cardiac images include echocardiography, cardiac MRI, and ECG-gated CT. S2, myocardial tissue segmentation: the fused image sequence is input into the U-Net dynamic segmentation network, and the myocardial tissue segmentation result with a timestamp is output; S3, motion trajectory modeling: Based on the results of myocardial tissue segmentation, a three-dimensional cardiac motion model is constructed. The motion vector of the myocardial tissue between adjacent frames is calculated using an improved optical flow algorithm to generate three-dimensional trajectory point cloud data within the beating cycle. S4, feature parameter extraction: perform spatiotemporal feature analysis on the trajectory point cloud data to extract multi-dimensional motion parameters including local displacement gradient, motion coordination index, and phase lag angle; S5, heterogeneity map generation: Generate myocardial pulsation heterogeneity map based on multidimensional motion parameters.

[0006] Optionally, the multimodal image fusion in S1 includes: S11, spatiotemporal alignment preprocessing: standardize the acquired echocardiography, cardiac MRI, and ECG-gated CT images in terms of format, image size, and coordinate system. S12, time axis synchronization: Time axis calibration of multimodal cardiac images based on ECG gating signals, mapping different imaging time points to a unified cardiac cycle reference frame; S13, spatial registration: A joint spatial alignment method based on feature point matching and affine deformation is used to achieve geometric alignment of MRI, CT, and echocardiography in three-dimensional space; S14, Feature Fusion Construction: By introducing a multi-channel feature fusion network with an attention mechanism, the structural features, density features, and dynamic features of multimodal cardiac images are fused to generate a fused image sequence. .

[0007] Optionally, the myocardial tissue segmentation in S2 includes: S21, fusion image time sequence input reorganization: the fusion image sequence Reconstructed into a four-dimensional tensor , to match the input structure of the dynamic U-Net network; S22, dynamic U-Net structure processing: Input into the dynamic U-Net network with attention gating and multi-scale feature fusion module to obtain the segmentation result tensor at each moment ; S23, segmentation output and timestamp mapping: segmentation results for each frame , appending its original timestamp , forming an output sequence with time labels .

[0008] Optionally, the motion trajectory modeling in S3 includes: S31, 3D myocardial tissue model construction: Based on the myocardial tissue segmentation results obtained for each frame, the spatial boundary and central skeleton information of the myocardial region are extracted, and the 3D myocardial tissue model of each frame is reconstructed using the voxel connectivity algorithm and morphological reconstruction technology; S32, temporal motion vector estimation: Based on the changes in myocardial structure between adjacent frames, an improved optical flow algorithm that introduces deformation priors and boundary preservation constraints is used to calculate the motion vector field; S33, generating a three-dimensional trajectory point cloud: performing a time-series accumulation of the displacement vectors of the myocardial voxels between all frames to obtain the spatial motion trajectory of each voxel within a complete beating cycle, and generating three-dimensional trajectory point cloud data.

[0009] Optionally, the three-dimensional myocardial tissue model construction in S31 includes: S311, Spatial Boundary Extraction: Based on Output Sequence with Time Labels , the segmentation mask for each frame Perform spatial gradient calculation to extract the myocardial region boundary voxel set ; S312, Central Skeleton Extraction: Constructing a 3D distance transform map inside the myocardial region , using local maximum points to form the central skeleton point set ; S313, 3D structure reconstruction: by defining a structurally consistent connectivity graph ,in, , in the connected graph Up through the skeleton point set As the structural backbone, combined with voxel interpolation operations, a topologically continuous 3D myocardial tissue model is generated. .

[0010] Optionally, the temporal motion vector estimation in S32 includes: S321, adjacent frame pairing and mask extraction: for any adjacent frame in the frame sequence and , generated from the three-dimensional myocardial tissue model Extract the voxel mask set of the corresponding frame and , establish the analysis domain for optical flow calculation between paired frames; S322, Constructing an improved optical flow energy function: Introducing an optical flow energy function based on 3D voxel deformation prior and boundary structure preservation constraints , for each point Motion vector Optimize S323, Optical flow field optimization and motion vector output: Based on the constructed optical flow energy function , by parameterizing the deep network Approximately model the motion vector field and minimize Training parameterized deep networks for the target , obtain the optical flow field output through forward reasoning and obtain the three-dimensional motion vector field .

[0011] Optionally, the generating of the three-dimensional trajectory point cloud in S33 includes: S331, Initial voxel coordinate set extraction: from the first time frame constructed Three-dimensional myocardial tissue model Extract voxel coordinates from ; S332, displacement vector time series accumulation: for each voxel point , based on the three-dimensional motion vector field estimated between frames , recursively calculate at each time frame spatial location ; S333, trajectory point cloud data generation: each voxel point in the entire beating cycle The position sequence combination within constitutes a three-dimensional trajectory point cloud set .

[0012] Optionally, the feature parameter extraction in S4 includes: S41, Local displacement gradient calculation: For a 3D trajectory point cloud , calculate the local displacement gradient tensor of each voxel point in its spatial neighborhood ; S42, calculation of the movement coordination index: on the complete trajectory set, the overall coordination of the entire myocardial tissue is evaluated and the movement coordination index is calculated ; S43, phase lag angle estimation: The trajectory of each voxel is represented as a three-dimensional vector time series, and principal axis projection and Fourier analysis are performed to extract the phase difference relative to the overall reference signal.

[0013] Optionally, the generation of the heterogeneity map in S5 includes: S51, Multi-dimensional motion parameter standardization and fusion: The extracted local displacement gradient, motion coordination index, and phase lag angle are linearly normalized and the fusion feature vector is constructed. ; S52, Abnormal Motion Region Identification and Classification: Based on the fused feature vectors, K-means cluster analysis is used to identify voxels with abnormal motion patterns in the myocardial region, including voxels with strong local differences, coordination mutations, or phase lags. Voxels with abnormal motion patterns are classified as abnormal motion regions and assigned category labels (strong delay, high variation, low consistency). S53, Construction and visualization of myocardial pulsation heterogeneity maps: Map voxel positions, category labels, and raw motion parameters to a 3D myocardial model to generate a myocardial pulsation heterogeneity map, and use color coding to identify different types of abnormal areas.

[0014] Beneficial effects of the present invention: The present invention, by constructing an automatic analysis method for the entire process from multimodal image fusion, automatic myocardial tissue segmentation, three-dimensional motion modeling to spatiotemporal feature extraction and heterogeneity map visualization, has opened up the technical link between structural imaging, functional motion and anatomical assessment. Compared with traditional myocardial function analysis methods that rely on single-modal imaging and manual assessment, this solution has significantly improved data fusion accuracy, segmentation robustness, time series tracking integrity and motion feature recognition capabilities, and realized the systematization, engineering and automation of cardiac tissue pulsation trajectory modeling.

[0015] This invention improves cardiac image quality by introducing multimodal registration and attention fusion mechanisms, and introduces a dynamic U-Net network to improve myocardial segmentation accuracy. At the same time, through a hybrid motion estimation strategy combining optical flow field and deep learning, the continuity and accuracy of three-dimensional myocardial voxel motion vectors are ensured. The trajectory point cloud-driven local gradient, motion coordination and phase lag angle multi-dimensional indicator extraction mechanism can carefully reveal the subtle differences in the motion state of myocardial tissue between different regions, thereby having higher sensitivity and resolution.

[0016] The present invention generates a myocardial pulsation heterogeneity map that not only intuitively presents abnormal motion areas in the form of color coding, but also quantifies and outputs multiple evaluation indicators, which helps doctors accurately locate and risk-stratify functional abnormalities such as myocardial ischemia, fibrosis, and motion inconsistency. It is particularly suitable for assisting in the early detection of myocardial diseases, surgical planning evaluation, and postoperative recovery monitoring, and has broad clinical application prospects and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 Schematic diagram of the analysis method according to an embodiment of the present invention; Figure 2 Schematic diagram of motion trajectory modeling according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0020] It should be noted that references in the specification to "one embodiment," "an embodiment," "exemplary embodiments," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not necessarily every embodiment will include such specific features, structures, or characteristics. Furthermore, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0021] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0022] like Figure 1-Figure 2 As shown, the automatic analysis method of the beating trajectory of engineered cardiac tissue based on the image recognition algorithm includes the following steps: S1, multimodal image fusion: Perform spatiotemporal registration and feature fusion on acquired multimodal cardiac images to generate a fused image sequence. Multimodal cardiac images include echocardiography, cardiac MRI, and ECG-gated CT. S2, myocardial tissue segmentation: the fused image sequence is input into the U-Net dynamic segmentation network, and the myocardial tissue segmentation result with a timestamp is output; S3, motion trajectory modeling: Based on the results of myocardial tissue segmentation, a three-dimensional cardiac motion model is constructed. The motion vector of the myocardial tissue between adjacent frames is calculated using an improved optical flow algorithm to generate three-dimensional trajectory point cloud data within the beating cycle. S4, feature parameter extraction: perform spatiotemporal feature analysis on the trajectory point cloud data to extract multi-dimensional motion parameters including local displacement gradient, motion coordination index, and phase lag angle; S5, heterogeneity map generation: Generate myocardial pulsation heterogeneity map based on multi-dimensional motion parameters.

[0023] Multimodal image fusion in S1 includes: S11, spatiotemporal alignment preprocessing: standardize the acquired echocardiography, cardiac MRI, and ECG-gated CT images in terms of format, image size, and coordinate system. S12, time axis synchronization: Time axis calibration of multimodal cardiac images based on ECG gating signals, mapping different imaging time points to a unified cardiac cycle reference frame, expressed as: ; in, is the image sequence of modality at the same time point . is a linear interpolation function is a time sequence of standard cardiac cycle is the cardiac image data sequence of modality . S13, Spatial registration: a joint spatial alignment method based on feature point matching and affine deformation is adopted to realize the geometric alignment of MRI, CT and echocardiogram in three-dimensional space, denoted as: . where, is the spatial coordinates of a voxel in the original image is the coordinates after registration is the linear affine transformation matrix of modality . is the translation vector S14, Feature fusion construction: through the introduction of multi-channel feature fusion network with attention mechanism, the structural features, density features and dynamic features of multi-modal cardiac images are fused to generate a fusion image sequence , denoted as: . . . where, is the fusion feature map at spatial position . is the feature map output of modality at spatial position . is the attention weight of modality at spatial position . is the number of modalities is the attention score function is the linear score weight vector of modality . is the bias term of modality .

[0024] The myocardial tissue segmentation in S2 includes: S21, Fusion image time sequence input reorganization: the fusion image sequence is reconstructed into a four-dimensional tensor to match the input structure of dynamic U-Net network, denoted as: . where, is the time spatial position on the fusion feature vector, is the index of the current channel, , , respectively represent the height, width and depth directions; S22, dynamic U-Net structure processing: input to the dynamic U-Net network with attention gate and multi-scale feature fusion module, obtain the segmentation result tensor of each time , expressed as: ; ; wherein, is an encoder, including a 3D convolution layer, BatchNorm, ReLU and an attention gate module, is an attention gate mechanism, that is, , is the feature map of the layer of the encoding layer, is an attention weight convolution kernel, is an attention activation map (weighted by position in channel dimension), is a Sigmoid activation function, is a decoder, including an up-sampling, multi-scale fusion and residual connection module, is a dynamic U-Net network; S23, segmentation output and timestamp mapping: for each frame of segmentation result , append its original timestamp , constitute the output sequence with time label , expressed as: ; wherein, is the three-dimensional myocardial tissue segmentation result of the frame, is the total number of frames of the fusion image sequence.

[0025] The motion trajectory modeling in S3 includes: S31, three-dimensional myocardial tissue model construction: based on the obtained myocardial tissue segmentation result of each frame, the spatial boundary and center skeleton information of the myocardial region are extracted, and the three-dimensional myocardial tissue model of each frame is reconstructed by using the voxel connectivity algorithm and morphological reconstruction technology; S32, time sequence motion vector estimation: for the myocardial structure change between adjacent frames, an improved optical flow field algorithm introducing deformation prior and boundary preservation constraint is used to calculate the motion vector field; S33, generating a three-dimensional trajectory point cloud: performing a time-series accumulation of the displacement vectors of the myocardial voxels between all frames to obtain the spatial motion trajectory of each voxel within a complete beating cycle, and generating three-dimensional trajectory point cloud data.

[0026] The construction of the three-dimensional myocardial tissue model in S31 includes: S311, Spatial Boundary Extraction: Based on Output Sequence with Time Labels , the segmentation mask for each frame Perform spatial gradient calculation to extract the myocardial region boundary voxel set , expressed as: ; in, is the myocardial voxel set, Voxel The three-dimensional gradient at is the gradient threshold of the boundary strength; S312, Central Skeleton Extraction: Constructing a 3D distance transform map inside the myocardial region , using local maximum points to form the central skeleton point set , expressed as: ; ; in, For location The shortest distance to the border, is any point in the boundary point set, is the central skeleton point set of the myocardial region, For location The 26-neighborhood voxel set of For location Shortest distance to the border; S313, 3D structure reconstruction: by defining a structurally consistent connectivity graph ,in, , in the connected graph Up through the skeleton point set As the structural backbone, combined with voxel interpolation operations, a topologically continuous 3D myocardial tissue model is generated. , expressed as: ; in, is the voxel connectivity interpolation function based on skeleton guidance; ; ; in, is the number of skeleton points, Skeleton points and The path between is the intermediate voxel point on the path, for point The spherical neighborhood voxel set of is the 3D voxel coordinate, is the interpolation neighborhood radius.

[0027] The temporal motion vector estimation in S32 includes: S321, adjacent frame pairing and mask extraction: for any adjacent frame in the frame sequence and , generated from the three-dimensional myocardial tissue model Extract the voxel mask set of the corresponding frame and , establish the analysis domain for optical flow calculation between paired frames, expressed as: ; ; S322, Constructing an improved optical flow energy function: Introducing an optical flow energy function based on 3D voxel deformation prior and boundary structure preservation constraints , for each point Motion vector For optimization, the energy function is expressed as: ; in, 、 Represents frames and The fused image is located at The grayscale or characteristic intensity at For myocardial voxels in frame Frame The three-dimensional motion vector of is the spatial gradient (tensor) of the optical flow field, is the deformation prior, constraining the rigidity or nonlinearity of the vector change. To maintain the boundary constraint, penalize the blurred optical flow transfer across the boundary, 、 、 is a hyperparameter that controls the weight of each constraint; S323, Optical flow field optimization and motion vector output: Based on the constructed optical flow energy function , by parameterizing the deep network Approximately model the motion vector field and minimize Training parameterized deep networks for the target , obtain the optical flow field output through forward reasoning and obtain the three-dimensional motion vector field , expressed as: ; ; in, is the output three-dimensional motion vector field, The optimal network parameters after training are such that the optical flow output minimizes the energy function .

[0028] The 3D trajectory point cloud generation in S33 includes: S331, Initial voxel coordinate set extraction: from the first time frame constructed Three-dimensional myocardial tissue model Extract voxel coordinates from , which represents the initial myocardial voxel positions involved in trajectory tracking at the beginning of the beating cycle, expressed as: ; in, For the Frame 3D myocardial tissue model, For the The voxels involved in trajectory generation are in the frame spatial location, is the number of initial voxel points involved in tracking; S332, displacement vector time series accumulation: for each voxel point , based on the three-dimensional motion vector field estimated between frames , recursively calculate at each time frame spatial location , expressed as: ; in, For the The voxel The spatial position of the frame, The position at the previous moment 3D displacement vector on the trajectory; S333, trajectory point cloud data generation: each voxel point in the entire beating cycle The position sequence combination within constitutes a three-dimensional trajectory point cloud set , expressed as: .

[0029] Feature parameter extraction in S4 includes: S41, Local displacement gradient calculation: For a 3D trajectory point cloud , calculate the local displacement gradient tensor of each voxel point in its spatial neighborhood , expressed as: ; ; in, Voxel The spatial neighborhood index set of is the relative displacement difference, is the stabilization term; S42, calculation of the movement coordination index: on the complete trajectory set, the overall coordination of the entire myocardial tissue is evaluated and the movement coordination index is calculated , expressed as: ; ; ; in, For the The total displacement vector of the voxel, is the average total displacement vector of all voxels, is the total number of voxels; S43, phase lag angle estimation: The trajectory of each voxel is represented as a three-dimensional vector time series, and principal axis projection and Fourier analysis are performed to extract the phase difference relative to the overall reference signal, which is expressed as: ; ; in, For the The phase lag angle of the voxel, For the Voxel in time The displacement in the main axis direction, For the The position of the center of mass of the voxel within the period (average position), The main direction of movement, is the fundamental cycle angular frequency.

[0030] Heterogeneity map generation in S5 includes: S51, Multi-dimensional motion parameter standardization and fusion: The extracted local displacement gradient, motion coordination index, and phase lag angle are linearly normalized so that all parameters are distributed in a uniform numerical range, and a fusion feature vector is constructed. ,in, For the Normalized local displacement gradient of a voxel, For the Normalized motor coordination score of the voxel, For the Normalized phase lag angle of the voxel; S52, Abnormal Motion Region Identification and Classification: Based on the fused feature vectors, the K-means cluster analysis method is used to identify voxels with abnormal motion patterns in the myocardial region, including voxel regions with strong local differences, coordinated mutations, or phase lags. Voxels with abnormal motion patterns are classified as abnormal motion regions and assigned category labels (strong delay, high variation, low consistency), expressed as: ; in, For the The cluster class label to which the voxel belongs, For the The eigenvectors of cluster centers, For the The eigenvector of the cluster center to which the voxel belongs; S53, Construction and visualization of myocardial pulsation heterogeneity maps: Map voxel positions, category labels, and raw motion parameters to a 3D myocardial model to generate a myocardial pulsation heterogeneity map, and use color coding to identify different types of abnormal areas.

[0031] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0032] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. An automatic analysis method for the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm, characterized in that: The following steps are involved: S1, multimodal image fusion: Perform spatiotemporal registration and feature fusion on acquired multimodal cardiac images to generate a fused image sequence. Multimodal cardiac images include echocardiography, cardiac MRI, and ECG-gated CT. S2, myocardial tissue segmentation: the fused image sequence is input into the U-Net dynamic segmentation network, and the myocardial tissue segmentation result with a timestamp is output; S3, motion trajectory modeling: Based on the results of myocardial tissue segmentation, a three-dimensional cardiac motion model is constructed. The motion vector of the myocardial tissue between adjacent frames is calculated using an improved optical flow algorithm to generate three-dimensional trajectory point cloud data within the beating cycle. S4, feature parameter extraction: perform spatiotemporal feature analysis on the trajectory point cloud data to extract multi-dimensional motion parameters including local displacement gradient, motion coordination index, and phase lag angle; S5, heterogeneity map generation: Generate myocardial pulsation heterogeneity map based on multi-dimensional motion parameters.

2. The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm according to claim 1, characterized in that: The multimodal image fusion in S1 includes: S11, spatiotemporal alignment preprocessing: standardize the acquired echocardiography, cardiac MRI, and ECG-gated CT images in terms of format, image size, and coordinate system. S12, time axis synchronization: Time axis calibration of multimodal cardiac images based on ECG gating signals, mapping different imaging time points to a unified cardiac cycle reference frame; S13, spatial registration: A joint spatial alignment method based on feature point matching and affine deformation is used to achieve geometric alignment of MRI, CT, and echocardiography in three-dimensional space; S14, Feature Fusion Construction: By introducing a multi-channel feature fusion network with an attention mechanism, the structural features, density features, and dynamic features of multimodal cardiac images are fused to generate a fused image sequence. .

3. The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm according to claim 2, characterized in that: The myocardial tissue segmentation in S2 includes: S21, fusion image time sequence input reorganization: the fusion image sequence Reconstructed into a four-dimensional tensor , to match the input structure of the dynamic U-Net network; S22, dynamic U-Net structure processing: Input into the dynamic U-Net network with attention gating and multi-scale feature fusion module to obtain the segmentation result tensor at each moment ; S23, segmentation output and timestamp mapping: segmentation results for each frame , appending its original timestamp , forming an output sequence with time labels .

4. The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm according to claim 3, characterized in that: The motion trajectory modeling in S3 includes: S31, 3D myocardial tissue model construction: Based on the myocardial tissue segmentation results obtained for each frame, the spatial boundary and central skeleton information of the myocardial region are extracted, and the 3D myocardial tissue model of each frame is reconstructed using the voxel connectivity algorithm and morphological reconstruction technology; S32, temporal motion vector estimation: Based on the changes in myocardial structure between adjacent frames, an improved optical flow algorithm that introduces deformation priors and boundary preservation constraints is used to calculate the motion vector field; S33, generating a three-dimensional trajectory point cloud: performing a time-series accumulation of the displacement vectors of the myocardial voxels between all frames to obtain the spatial motion trajectory of each voxel within a complete beating cycle, and generating three-dimensional trajectory point cloud data.

5. The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm according to claim 4, characterized in that: The three-dimensional myocardial tissue model construction in S31 includes: S311, Spatial Boundary Extraction: Based on Output Sequence with Time Labels , the segmentation mask for each frame Perform spatial gradient calculation to extract the myocardial region boundary voxel set ; S312, Central Skeleton Extraction: Constructing a 3D distance transform map inside the myocardial region , using local maximum points to form the central skeleton point set ; S313, 3D structure reconstruction: by defining a structurally consistent connectivity graph ,in, , in the connected graph Up through the skeleton point set As the structural backbone, combined with voxel interpolation operations, a topologically continuous 3D myocardial tissue model is generated. .

6. The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm according to claim 5, characterized in that: The temporal motion vector estimation in S32 includes: S321, adjacent frame pairing and mask extraction: for any adjacent frame in the frame sequence and , generated from the three-dimensional myocardial tissue model Extract the voxel mask set of the corresponding frame and , establish the analysis domain for optical flow calculation between paired frames; S322, Constructing an improved optical flow energy function: Introducing an optical flow energy function based on 3D voxel deformation prior and boundary structure preservation constraints , for each point Motion vector Optimize S323, Optical flow field optimization and motion vector output: Based on the constructed optical flow energy function , by parameterizing the deep network Approximately model the motion vector field and minimize Training parameterized deep networks for the target , obtain the optical flow field output through forward reasoning and obtain the three-dimensional motion vector field .

7. The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm according to claim 6, characterized in that: The three-dimensional trajectory point cloud generation in S33 includes: S331, Initial voxel coordinate set extraction: from the first time frame constructed Three-dimensional myocardial tissue model Extract voxel coordinates from ; S332, displacement vector time series accumulation: for each voxel point , based on the three-dimensional motion vector field estimated between frames , recursively calculate at each time frame spatial location ; S333, trajectory point cloud data generation: each voxel point in the entire beating cycle The position sequence combination within constitutes a three-dimensional trajectory point cloud set .

8. The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm according to claim 7, characterized in that: The feature parameter extraction in S4 includes: S41, Local displacement gradient calculation: For a 3D trajectory point cloud , calculate the local displacement gradient tensor of each voxel point in its spatial neighborhood ; S42, calculation of the movement coordination index: on the complete trajectory set, the overall coordination of the entire myocardial tissue is evaluated and the movement coordination index is calculated ; S43, phase lag angle estimation: The trajectory of each voxel is represented as a three-dimensional vector time series, and principal axis projection and Fourier analysis are performed to extract the phase difference relative to the overall reference signal.

9. The method for automatically analyzing the pulse trajectory of engineered cardiac tissue based on an image recognition algorithm according to claim 8, characterized in that: The generation of the heterogeneity map in S5 includes: S51, Multi-dimensional motion parameter standardization and fusion: The extracted local displacement gradient, motion coordination index, and phase lag angle are linearly normalized and the fusion feature vector is constructed. ; S52, Abnormal Motion Region Identification and Classification: Based on the fused feature vectors, K-means cluster analysis is used to identify voxels with abnormal motion patterns in the myocardial region, including voxel regions with strong local differences, coordination mutations, or phase lags. Voxels with abnormal motion patterns are classified as abnormal motion regions and assigned category labels. S53, Construction and visualization of myocardial pulsation heterogeneity maps: Map voxel positions, category labels, and raw motion parameters to a 3D myocardial model to generate a myocardial pulsation heterogeneity map, and use color coding to identify different types of abnormal areas.

Images