A dynamic single particle tracking method and apparatus

Abstract

The application relates to the technical field of single-particle tracking, and discloses a dynamic single-particle tracking method and device, which can collect an original image sequence for tracking a single-particle motion trajectory, select corresponding target space-time characteristic elements according to single-particle performance characteristics in the original image sequence, track the single particle based on the target space-time characteristic elements in the original image sequence, obtain the single-particle motion trajectory, extract single-particle motion data along the single-particle motion trajectory, and visually output the single-particle motion data, so that the single-particle tracking accuracy is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of single-particle tracking technology, and more particularly to a dynamic single-particle tracking method and apparatus. Background Technology

[0002] Single-particle tracking has wide applications in various imaging techniques. However, accurate tracking of single particles presents certain challenges, particularly in situations involving complex backgrounds within living organisms or where single-molecule signals are weak.

[0003] The key to single-particle tracking in related technologies lies in particle identification and localization, and the commonly used technique is label-free imaging. Label-free imaging does not require external labels, can significantly extend the tracking time of single particles, and provides the possibility for studying the behavior of biomolecules in their natural state.

[0004] Specifically, in label-free imaging, related techniques can achieve particle recognition in a single frame of static image by calibrating or adjusting fitting parameters. However, the tracking consistency between consecutive frames decreases, and the particle trajectory connection error rate increases significantly. This shortcoming greatly limits the reliable application of label-free imaging in long-term biological dynamic analysis.

[0005] Therefore, there is an urgent need to develop a way to effectively improve the accuracy of single-particle tracking. Summary of the Invention

[0006] This invention provides a dynamic single-particle tracking method and apparatus to address the shortcomings of low single-particle tracking accuracy in related technologies and improve the accuracy of dynamic single-particle tracking.

[0007] In a first aspect, the present invention provides a dynamic single-event tracking method, comprising: Acquire raw image sequences for tracking the trajectory of a single particle; Based on the single-particle performance characteristics in the original image sequence, select the corresponding target spatiotemporal feature elements; Based on the target spatiotemporal features, single-particle tracking is performed on the original image sequence to obtain the single-particle motion trajectory. Single-particle motion data is extracted along the single-particle trajectory, and the single-particle motion data is visualized and output.

[0008] Optionally, selecting the corresponding target spatiotemporal feature elements based on the single-particle performance features in the original image sequence includes: If the single particles in the original image sequence are detected to be single particles or single molecules with sparse spatial distribution, then the correlation between the created single particle performance features and spatiotemporal feature elements is obtained. Based on the sparse spatial distribution characteristics, the corresponding spatiotemporal feature elements are found in the association relationship as the binding strength, point spread function (PSF), binding event location in two-dimensional space, pure background frame image before binding, and image of slow change of local background over time, and are used as the target spatiotemporal feature elements.

[0009] Optionally, the step of performing single-particle tracking based on the target spatiotemporal feature elements in the original image sequence to obtain the single-particle motion trajectory includes: The original image sequence is preprocessed to suppress noise, resulting in a preprocessed image sequence. Based on the binding strength features and PSF features of each frame image in the preprocessed image sequence, the effective binding positions in each frame image are determined; For any of the effective combination positions, the effective combination position is used as the center point of the local image block to be cropped. Based on the center point and the set first local size, local image blocks are cropped in the target frame image including the effective combination position, the corresponding first N frame images, and the last M frame images to obtain a local image block sequence. The first few local image blocks in the local image block sequence are used as sample image blocks and the model is trained to obtain a trained center background prediction model. The center background prediction model is used to predict the background of the central region pixels of each non-sample image block. Based on the background of the central region pixels of each non-sample image block, the central region pixels of each non-sample image block are purified to obtain the corresponding purified image block sequence. Single-particle tracking is performed on the purified image block sequence to obtain the single-particle motion trajectory corresponding to the effective combination position. The local image block cropped from the target frame image is the non-sample image block.

[0010] Optionally, determining the effective binding position in each frame image based on the binding strength feature and PSF feature in each frame image of the preprocessed image sequence includes: For any frame image in the preprocessed image sequence, the frame image is input into a trained U-shaped convolutional neural network, so that the U-shaped convolutional neural network: extracts the binding strength and PSF features in the frame image, generates and outputs a binding event probability map based on the proposed binding strength and PSF features, and the binding event probability map is marked with candidate binding positions whose confidence exceeds a set confidence threshold; For any candidate combination position, determine the target frame image corresponding to the candidate combination position. If the spatial distribution feature of the candidate combination position in the target frame image is sparse and isolated, then perform a difference operation between the target frame image and the corresponding previous frame image to determine the grayscale change value corresponding to the candidate combination position. If the grayscale change value is greater than a set grayscale threshold, then determine that the candidate combination position has a rapid instantaneous intensity jump, and determine that the candidate combination position is a valid combination position.

[0011] Optionally, the step of using multiple local image patches located at the top of the local image patch sequence as sample image patches and training the model to obtain a trained center background prediction model includes: For any of the sample image blocks, a corresponding target center region is determined in the sample image block according to the set second local size, and the target center region is removed from the sample image block to obtain a sample image block after removal. A set number of consecutively removed sample image blocks are used as a sequence of sample image blocks and input into the model to be trained, so that the model to be trained can: predict the pixel background of the target center region of the next image block in the sample image block sequence based on the pixels in the sample image block sequence, and generate and output the prediction result. The parameters of the model to be trained are updated based on the prediction results until a well-trained center background prediction model is obtained.

[0012] Optionally, the step of performing single-particle tracking on the purified image patch sequence to obtain the single-particle motion trajectory corresponding to the effective binding position includes: For any purified frame image in the purified image block sequence, two-dimensional centroid localization is performed in the PSF of the purified frame image to obtain the sub-pixel level single particle position coordinates of the purified frame image. The single-particle position coordinates of each purified frame image in the purified image block sequence are sequentially associated to obtain the single-particle motion trajectory corresponding to the effective binding position.

[0013] Optionally, extracting single-particle motion data along the single-particle trajectory includes: For any purified frame image in the purified image block sequence, identify the Airy disk region in the purified frame image, and calculate the pixel value of each pixel in the Airy disk region by integration and average to obtain the instantaneous image intensity of the purified frame image. Based on the instantaneous image intensity of each purified frame image, the dwell time from the first appearance of a single particle to its disappearance or the end of the acquisition is determined; The instantaneous image intensity, single-particle position coordinates, and dwell time of each purified frame image are used as the single-particle motion data extracted along the single-particle motion trajectory.

[0014] Optionally, selecting the corresponding target spatiotemporal feature elements based on the single-particle performance features in the original image sequence includes: If the single particles detected in the original image sequence are cell adhesion specks, then the target spatiotemporal feature elements are selected as the spatiotemporal feature elements extracted from the original image, Gaussian operator image, and Hessian operator image; The step of performing single-particle tracking based on the target spatiotemporal feature elements in the original image sequence to obtain the single-particle motion trajectory includes: For any frame image in the original image sequence, based on the frame image, a first Gaussian operator image and a second Gaussian operator image at a first set scale and a second set scale are generated respectively. A first Hessian operator image and a second Hessian operator image along the X direction at the first set scale and the second set scale are generated respectively. A third Hessian operator image and a fourth Hessian operator image along the Y direction at the first set scale and the second set scale are generated respectively. The first set scale is smaller than the second set scale. The frame image and the generated operator images are combined according to channels to obtain seven-channel data. The seven-channel data is input into a trained U-shaped convolutional neural network for processing to obtain a speckle probability map and a star distance StarDist map. The speckle probability map and the StarDist map are synthesized to obtain a segmented speckle image. Multi-target data association is performed on the adhesion plaque images corresponding to the consecutive frame images to obtain the motion trajectory of each cell adhesion plaque.

[0015] Optionally, extracting single-particle motion data along the single-particle trajectory includes: For any of the cell adhesion spots' motion trajectories, the position coordinates and image intensity are recorded along the motion trajectory of the cell adhesion spot. The intensity jump event, random motion speed, and radial velocity are determined based on the recorded position coordinates and image intensity, and are used as the overall motion data of the cell adhesion spot.

[0016] In a second aspect, the present invention provides a dynamic single-particle tracking device, applied to the dynamic single-particle tracking method of the first aspect or any corresponding embodiment thereof, the device comprising: The acquisition unit is used to acquire raw image sequences for tracking the trajectory of a single particle. The selection unit is used to select the corresponding target spatiotemporal feature elements based on the single-particle performance characteristics in the original image sequence. The tracking unit is used to perform single-particle tracking in the original image sequence based on the spatiotemporal features of the target to obtain the single-particle motion trajectory. The extraction unit is used to extract single-particle motion data along the single-particle motion trajectory; The output unit is used to visualize the single-particle motion data.

[0017] The dynamic single-particle tracking method and apparatus provided by this invention can acquire an original image sequence for tracking the trajectory of a single particle, select corresponding target spatiotemporal feature elements based on the single particle performance characteristics in the original image sequence, perform single-particle tracking based on the target spatiotemporal feature elements in the original image sequence to obtain the single-particle trajectory, extract single-particle motion data along the single-particle trajectory, and visualize the single-particle motion data, thereby effectively improving the accuracy of single-particle tracking.

[0018] Compared with the single-particle tracking method in related technologies, the present invention has the following advantages: (1) This invention integrates the spatial and temporal features of a single particle. This reduces the reliance on fixed thresholds, single-frame image intensity, or single morphological features, and improves the recognition stability of single particles under complex backgrounds and weak signal conditions.

[0019] (2) For single particles or molecules that are sparsely distributed in space, the present invention utilizes the difference between the rapid change of single particle signal and the slow change of local background to suppress background drift and preserve the complete signal during the continuous existence of the object.

[0020] (3) The present invention predicts the local background for each combined event, and combines background subtraction and point spread function subpixel localization to continuously extract the intensity, position and dwell time of the object.

[0021] (4) For cell adhesion plaques with complex morphology and dense distribution, the present invention integrates original strength, multi-scale spot structure and multi-directional ridge structure, which can adapt to adhesion plaques of different scales, morphologies and extension directions, and reduce missed detection.

[0022] (5) Adhesion speckle probability map and StarDist map are used to distinguish single particles that are close to each other, while multi-target data association method is used to establish the correspondence between single particles in consecutive frames, thereby reducing object merging, trajectory breakage and error linking.

[0023] (6) The present invention selects appropriate feature elements and processing methods for different single particles, and then realizes the identification and tracking from the sub-structural scale of single molecules to single cells under a unified processing logic. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in this invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A flowchart of a dynamic single-particle tracking method provided in an embodiment of the present invention; Figure 2 An image processing flowchart for single-particle and single-molecule tracking provided in an embodiment of the present invention; Figure 3 This is a single-particle and single-molecule tracking result diagram provided in an embodiment of the present invention; Figure 4 An image processing flowchart for cell adhesion speckle tracking provided in an embodiment of the present invention; Figure 5 This is a cell adhesion speckle tracking result diagram provided in an embodiment of the present invention; Figure 6 A flowchart of another dynamic single-particle tracking method provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of a dynamic single-particle tracking device provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The following is combined with Figures 1-6 The present invention describes the dynamic single-particle tracking method.

[0028] like Figure 1 As shown, this embodiment proposes a first dynamic single-event tracking method, which may include the following steps: S101. Acquire the original image sequence for tracking the trajectory of a single particle.

[0029] Among them, a single particle refers to a spatial entity that can be located, distinguished, and established as an independent object in a continuous chemical image and has a corresponding relationship between consecutive frames.

[0030] Optionally, a single particle can be a single grain or a single molecule.

[0031] The original image sequence is a time-series image capable of recording single-particle binding, positional changes, or local environmental changes. Specifically, this implementation can use a plasmonic scattering microscopy system to acquire images of the target area where the single particle is located, thus obtaining the original image sequence.

[0032] Specifically, when the single particle is a single particle or a single molecule, the original image sequence can be obtained using a label-free dual-objective plasmon scattering microscopy system. A 660 nm laser is used to irradiate the gold sensing film at the surface plasmon resonance angle through a high numerical aperture (1.50) oil immersion objective, exciting surface plasmon waves. The surface plasmon scattering signals of polystyrene nanoparticles or protein molecules flowing across the surface of the sensing film are collected by the objective located on the sample side, and the original image sequence is obtained by continuous acquisition by the camera.

[0033] S102. Select the corresponding target spatiotemporal feature elements based on the single-particle performance characteristics in the original image sequence.

[0034] It should be noted that this embodiment can integrate spatial and temporal feature elements for dynamic single-particle tracking. The spatial feature elements can be used to describe the intensity, position, morphology, and distribution of a single particle in a single-frame image or local spatial region, as well as its relationship with the neighboring background. Specifically, this can include the intensity and point spread function of the single particle in the original single-frame image or local spatial region, its position, morphology, and density distribution in two-dimensional space, and Gaussian operator filtering and Hessian operator filtering feature elements used to characterize the morphology of the single particle in two-dimensional space.

[0035] Temporal features can be used to describe the continuous process of the appearance, disappearance, intensity, and / or motion changes of a single particle and its local background in consecutive frame images. Specifically, they can include the intensity and point spread function of the single particle, the jumps and / or slow changes of spatial features such as position, shape, and density distribution in two-dimensional space between adjacent frame images, and can also include the time of appearance and disappearance of the available single particle combination event, the pure background frame image before combination, and the slow changes of the local background over time.

[0036] Specifically, this embodiment can achieve dynamic single-particle identification and tracking through the comprehensive identification of multiple spatial and temporal features. Not all spatial and temporal features are used simultaneously for every single particle. This embodiment can select some or all spatial and temporal features based on the actual characteristics of the single particle to be identified. The spatial and temporal features selected in this embodiment are collectively referred to as target spatiotemporal features.

[0037] Optionally, step S102 includes: If the single particles in the original image sequence are detected to be single particles or single molecules with sparse spatial distribution, then the correlation between the established single particle performance features and spatiotemporal feature elements is obtained. Based on the sparse spatial distribution characteristics, the corresponding spatiotemporal feature elements are found in the association relationship as the binding strength, point spread function (PSF), binding event location in two-dimensional space, pure background frame image before binding, and image of slow change of local background over time, and are taken as the target spatiotemporal feature elements as a whole.

[0038] Specifically, for sparsely distributed single particles or molecules in space, rapid temporal intensity jumps occur during binding or dissociation, while the local background changes slowly over time. Selectable feature elements include the binding strength and point spread function of single particles, the location of the binding event in two-dimensional space, the pure background frame image before binding, and the original image of the slow change of the local background over time.

[0039] S103. Based on the spatiotemporal features of the target, perform single-particle tracking in the original image sequence to obtain the single-particle motion trajectory.

[0040] Specifically, for single particles or molecules that are sparsely distributed in space and whose signals undergo rapid intensity jumps during binding or dissociation while the local background changes slowly, differential event localization, gated cyclic unit background prediction, background subtraction, and point spread function subpixel localization can be used sequentially to achieve single particle identification and tracking.

[0041] Optionally, when the single particles detected in the original image sequence are single particles or single molecules exhibiting a sparse spatial distribution, step S103 includes: The original image sequence is preprocessed to suppress noise, resulting in a preprocessed image sequence. Based on the binding strength features and PSF features of each frame in the preprocessed image sequence, the effective binding positions in each frame are determined. For any valid combination position, the valid combination position is used as the center point of the local image patch to be cropped. Based on the center point and the set first local size, local image patches are cropped in the target frame image including the valid combination position, the corresponding first N frame images, and the last M frame images to obtain a local image patch sequence. The first few local image patches in the local image patch sequence are used as sample image patches and the model is trained to obtain a trained center background prediction model. Using the center background prediction model, the background of the central region pixels of each non-sample image patch is predicted. Based on the background of the central region pixels of each non-sample image patch, the central region pixels of each non-sample image patch are purified to obtain the corresponding purified image patch sequence. Single-particle tracking is performed on the purified image patch sequence to obtain the single-particle motion trajectory corresponding to the valid combination position. Among them, the local image patch cropped in the target frame image is the non-sample image patch.

[0042] Optionally, the above determination of the effective binding location in each frame image based on the binding strength features and PSF features in each frame image of the preprocessed image sequence includes: For any frame image in the preprocessed image sequence, the frame image is input into the trained U-shaped convolutional neural network so that the U-shaped convolutional neural network: extracts the binding strength and PSF features in the frame image, generates and outputs a binding event probability map based on the proposed binding strength and PSF features, and marks the candidate binding positions with confidence levels exceeding the set confidence threshold in the binding event probability map. For any candidate combination position, determine the target frame image corresponding to the candidate combination position. If the spatial distribution characteristics of the candidate combination position in the target frame image are sparse and isolated, then perform a difference operation between the target frame image and the corresponding previous frame image to determine the gray level change value corresponding to the candidate combination position. If the gray level change value is greater than the set gray level threshold, then it is determined that the candidate combination position has a rapid instantaneous intensity jump, and the candidate combination position is determined to be a valid combination position.

[0043] Optionally, the above-mentioned method uses multiple local image patches located at the top of the local image patch sequence as sample image patches for model training to obtain a trained center background prediction model, including: For any sample image block, the corresponding target center region is determined in the sample image block according to the set second local size, and the target center region is removed from the sample image block to obtain the sample image block after removal. A set number of consecutively removed sample image blocks are used as a sequence of sample image blocks and input into the model to be trained. The model then predicts the pixel background of the target center region of the next image block in the sample image block sequence based on the pixels in the sequence, and generates and outputs the prediction result. The parameters of the model to be trained are updated based on the prediction results until a well-trained center background prediction model is obtained.

[0044] Specifically, such as Figure 2 As shown, in this embodiment, the original image sequence is first preprocessed (spatial / temporal averaging) to suppress noise, resulting in a preprocessed image sequence. Then, the preprocessed image sequence is input into a U-Net model (i.e., the U-shaped convolutional neural network mentioned above) that includes an encoder, a fully connected layer, and a decoder. After extracting feature vectors from the preprocessed image sequence, the probability map of the binding event is decoded to locate high-confidence events generated by particle or molecule binding, i.e., high-confidence candidate binding positions.

[0045] For candidate binding sites, differential processing is performed on adjacent frame images, and high-confidence signals generated by particle or molecular binding are selected from the differential images. The effective binding sites are determined by utilizing the spatial sparseness of the objects and the rapid temporal intensity jumps at the event time.

[0046] Subsequently, this embodiment can perform dynamic background prediction. For any effective binding position, this embodiment can extract the target frame image including the effective binding position, the corresponding first 50 frames, and the last 50 frames from the preprocessed image sequence. Using the effective binding position as the center point of the local image patch to be cropped, and setting the first local size to 45×45 pixels, 45×45 pixel local image patches are cropped from the extracted target frame image and the 100 frames respectively, resulting in a local image patch sequence. The first 40 frames of local image patches in the local image patch sequence are used as sample image patches to train the Gated Recurrent Unit (GRU) network, resulting in a trained GRU network.

[0047] Specifically, in this embodiment, during the training of the GRU network, the second local size is set to 15×15. The central 15×15 pixel region of five consecutive frame sample image blocks is removed, resulting in five consecutive frame sample image blocks after removal. These five consecutive frame sample image blocks are then input into the GRU network to be trained to predict the background of the central 15×15 pixel region of the next frame local image block. This allows the network to learn local background changes based on peripheral pixels and consecutive frames. This event-specific training method ensures that the background model corresponds to the local microenvironment where the event occurs. After obtaining the trained GRU network, this embodiment can use the trained GRU network to predict the background of the central pixel region of the last 61 frames of local image blocks (i.e., non-sample image blocks) in the local image block sequence. The corresponding predicted background is subtracted from the 15×15 central pixel region of each non-sample image block to obtain the cleaned image block sequence.

[0048] Optionally, the above-mentioned single-particle tracking of the purified image patch sequence to obtain the single-particle motion trajectory corresponding to the effective binding position includes: For any cleaned frame image in the cleaned image block sequence, perform two-dimensional centroid localization in the PSF of the cleaned frame image to obtain the sub-pixel level single particle position coordinates of the cleaned frame image. The single-particle position coordinates of each purified frame image in the purified image block sequence are sequentially correlated to obtain the single-particle motion trajectory corresponding to the effective combination position.

[0049] S104. Extract single-particle motion data along the single-particle trajectory.

[0050] Specifically, this embodiment can extract the motion data of a single particle along its trajectory.

[0051] Optionally, step S104 above includes: For any cleaned frame image in the cleaned image block sequence, identify the Airy disk region in the cleaned frame image, and calculate the instantaneous image intensity of the cleaned frame image by integrating and averaging the pixel value of each pixel in the Airy disk region. Based on the instantaneous image intensity of each purified frame image, the dwell time from the first appearance of a single particle to its disappearance or the end of the acquisition is determined; The instantaneous image intensity, single-particle position coordinates, and dwell time of each purified frame image are used as a whole to extract single-particle motion data along the single-particle motion trajectory.

[0052] Specifically, in this embodiment, in each purified image block included in the purified image block sequence, pixel integration is performed on the Airy disk region centered on a single particle or a single molecule and the average is taken to obtain the image intensity; the centroid of the point spread function is located to obtain the continuous position; and the dwell time is obtained based on the time of object appearance and disappearance.

[0053] S105. Visualize and output the motion data of a single particle.

[0054] Specifically, this embodiment can output the extracted single-particle motion data in the form of time curves or statistical distributions, and the output results can form... Figure 3 The image intensity temporal curve, dwell time distribution, image intensity distribution, and lateral position results are shown. Figure 3In the figure, subfigures a and b show the image intensity-time curves and z-direction displacement step size distributions obtained after background prediction and first-frame subtraction of the same 28 nm polystyrene microsphere binding event by a gated cyclic unit; subfigure c shows the residence time distribution of stable binding events (>1 s) and transient binding events (<1 s); subfigure d shows the image intensity distribution of total binding events and unstable transient binding events; subfigure e shows the relationship between the image intensity values ​​and molecular diameters of three immunoglobulin molecules (IgG, IgA, and IgM) after background subtraction by a gated cyclic unit; and subfigure f shows the time trajectory curve of the lateral position of the IgA molecule. This figure illustrates that local background prediction can preserve continuous object signals under slowly changing background conditions and supports residence time, image intensity, and position analysis.

[0055] Specifically, this embodiment can achieve comprehensive judgment and tracking by combining differential event location, local spatiotemporal image patch, pure background frame, gated cyclic unit predicted background, continuous cleaned image, point spread function, and continuous position. In result verification, the gated cyclic unit background prediction is compared with the first frame. Through imaging verification with larger particles, the difference between the gated cyclic unit background prediction and the first frame yields a similar image intensity, and the centroid localization result can be used to confirm lateral dynamic information.

[0056] Depend on Figure 3 The residence time distribution in the image can distinguish between short-duration transient binding events and long-duration stable binding events; the image intensity distribution can compare different particle sizes or different protein objects; and the lateral position-time curve and lateral displacement distribution can describe the dynamic changes of the object near the sensing surface.

[0057] It should be noted that the above results collectively demonstrate that this embodiment does not rely on a single intensity, but rather makes a comprehensive judgment based on the event location, background time changes, purification intensity, point diffusion function, and continuous location.

[0058] The dynamic single-particle tracking method proposed in this embodiment can acquire an original image sequence for tracking the trajectory of a single particle, select corresponding target spatiotemporal feature elements based on the single particle performance characteristics in the original image sequence, perform single-particle tracking based on the target spatiotemporal feature elements in the original image sequence to obtain the single particle trajectory, extract single-particle motion data along the single particle trajectory, and visualize the single particle motion data, effectively improving the accuracy of single-particle tracking.

[0059] based on Figure 1This embodiment proposes a second dynamic single-particle tracking method, in which the single particle can be a focal adhesion of a single cellular substructure, i.e., a cell focal adhesion. For example, this embodiment can be used to track the dynamic changes in focal adhesion induced by 293T cells transfected with transient receptor potential vanillic acid isoform 1 (TRPV1) under thermal stimulation.

[0060] Specifically, when the single particle is a cell adhesion speckle, the original image sequence can be obtained using a label-free locally heated plasmon scattering microscopy system. A 660nm laser is used to excite surface plasmon waves over a wide field, and a smaller 638nm laser beam is used as the local heating light. After the objective lens collects the surface plasmon scattering signal, the imaging light and the heating light are separated by a long-pass dichroic mirror and imaged onto the camera. The original image sequence is obtained by continuous acquisition.

[0061] At this time, step S102 above includes: If the single particles detected in the original image sequence are cell adhesion specks, then the target spatiotemporal feature elements are selected as the spatiotemporal feature elements extracted from the original image, Gaussian operator image, and Hessian operator image.

[0062] Specifically, for cell adhesion spots, which have different shapes such as near-circular or non-circular, they may be adjacent to each other and move in consecutive frame images. The available spatiotemporal feature elements include the original image, Gaussian operator filtered image, and Hessian operator filtered image to extract various feature elements.

[0063] At this point, step S103 includes: For any frame image in the original image sequence, based on the frame image, a first Gaussian operator image and a second Gaussian operator image at a first set scale and a second set scale are generated respectively. A first Hessian operator image and a second Hessian operator image along the X direction at the first set scale and a second set scale are also generated respectively. A third Hessian operator image and a fourth Hessian operator image along the Y direction at the first set scale and a second set scale are also generated respectively. The first set scale is smaller than the second set scale. The frame image and the generated operator images are combined according to channels to obtain seven-channel data. The seven-channel data is input into a trained U-shaped convolutional neural network for processing to obtain a speckle probability map and a star distance StarDist map. The speckle probability map and StarDist map are synthesized to obtain a segmented speckle image. Multi-target data association is performed on the plaque images corresponding to consecutive frames to obtain the motion trajectory of each cell plaque.

[0064] The first set scale is a small scale, and the second set scale is a large scale.

[0065] Specifically, such as Figure 4As shown, in this embodiment, for each frame of plasmon scattering microscopic image in the original image sequence, the corresponding Gaussian operator 1 (i.e., the first Gaussian operator image at a small scale), Gaussian operator 2 (i.e., the second Gaussian operator image at a large scale), Hessian operator x1 (i.e., the first Hessian operator image along the X direction at a small scale), Hessian operator x2 (i.e., the second Hessian operator image along the X direction at a large scale), Hessian operator y1 (i.e., the first Hessian operator image along the Y direction at a small scale), and Hessian operator y2 (i.e., the second Hessian operator image along the Y direction at a large scale) can be generated respectively.

[0066] It is understood that the six operator images generated in this embodiment are all filtered images. This embodiment combines the original image with the six generated filtered image representations by channel to form a seven-channel input. After processing by a seven-channel U-Net, the output is a speckle probability map and a StarDist map, from which the segmented speckle image is obtained. The seven channels correspond to the speckle features: the original image retains the actual image intensity; Gaussian operator 1 and Gaussian operator 2 present smooth representations at different scales; Hessian operators x1, x2, y1, and y2 present structures with different directions and scales; the speckle probability map gives the speckle region; and the StarDist map is used for object detection and instance separation. This combination allows both near-circular and non-circular specks to be tracked subsequently. Hungarian linking is performed on the segmented specks in consecutive frames to form a continuous trajectory for each speckle instance. The x-position, y-position, and image intensity are recorded along the trajectory. The identification of a single adhesion speckle relies simultaneously on the original intensity, Gaussian operator filtering representation, Hessian operator filtering representation, and StarDist object representation, rather than using a single channel as the final judgment criterion. Multi-target data association (Hungarian linking) is performed on segmented adhesion specks in consecutive frames to form a continuous trajectory for each cell adhesion speckle.

[0067] At this point, step S105 above includes: For any cell adhesion speckle's motion trajectory, the position coordinates and image intensity are recorded along the trajectory. The intensity jump event, random motion velocity, and radial velocity are determined based on the recorded position coordinates and image intensity, and are used as the overall motion data of the cell adhesion speckle.

[0068] Specifically, this embodiment records the x-position, y-position, and image intensity along the trajectory, and further obtains intensity jump events, random motion speed, and radial velocity. The comprehensive identification and tracking of adhesion specks is achieved through a combination of seven-channel images, a speckle probability map, a StarDist map, segmentation instances, and continuous frame linking.

[0069] like Figure 5As shown, this embodiment compares the trajectory obtained by the seven-channel U-Net method with the manually labeled trajectory and the TrackMate trajectory. The x-position, y-position, and image intensity changes of the seven-channel U-Net method are closer to the manually labeled results. In dynamic parameter extraction, intensity jump events are recorded based on the rapid changes in the intensity of the adhesion speckle image, and the frequency of intensity jump events is calculated. Random motion velocity and radial velocity relative to the local heating center or cell center are obtained based on continuous positions. The above parameters are statistically analyzed during the stimulus-off and stimulus-on phases to form time curves and histogram distributions, and Gaussian fitting is performed on the random motion velocity and radial velocity distributions. Figure 5 In the figure, subfigure a shows the x-position, y-position, and image intensity temporal curves obtained by manual annotation, seven-channel U-Net, and TrackMate methods; subfigure b shows the spatial heatmap of intensity jump events before and after thermal stimulation; subfigure c shows the spatial distribution heatmap of plaque random motion velocity before and after thermal stimulation; subfigure d shows the spatial distribution heatmap of plaque radial motion velocity before and after thermal stimulation; subfigure e shows the count of intensity jump events before and after stimulation; subfigure f shows the histogram distribution and Gaussian fitting of plaque random motion velocity before and after stimulation; subfigure g shows the histogram distribution and Gaussian fitting of plaque radial motion velocity before and after stimulation; subfigure h shows the time curve of intensity jump event counts during the experiment; subfigure i shows the time curve of random motion velocity during the experiment; and subfigure j shows the time curve of radial motion velocity during the experiment. This figure is used to illustrate that seven-channel instance recognition and Hungarian linking can obtain continuous plaque trajectories and support dynamic parameter comparison.

[0070] Local heating experiments can be used to observe the adhesion speckle response after TRPV1 activation. By comparing the frequency of intensity jump events, random motion velocity, and radial velocity at different locations on the membrane, the spatial heterogeneity of TRPV1 expression and non-specific thermal effects on the membrane can be distinguished.

[0071] This embodiment can use seven-channel image representation, adhesion speckle probability map, StarDist object detection results, segmentation instances, Hungarian linking, and continuous trajectory parameters for comprehensive judgment in sequence.

[0072] based on Figure 1 ,like Figure 6 As shown, this embodiment proposes a third dynamic single-particle tracking method. The specific tracking process differs depending on the type of single particle, but follows the same main principle: multiple complementary parameters can be selected based on the object's characteristics, and spatial and continuous temporal information are then used to complete identification and tracking. Specifically, this method may include the following steps: S1. Image acquisition, obtaining the original microscopic image sequence.

[0073] S2. Feature element selection: Specifically, based on the type and characteristics of a single particle object, spatial and temporal feature elements are selected.

[0074] S3. Perform single-frame object recognition and use spatial features to identify single particles in each frame.

[0075] S4. Continuous frame association establishes object correspondence and forms single particle trajectory.

[0076] S5. Extract dynamic parameters along the single-particle trajectory, specifically image intensity, dwell time, position coordinates, and motion parameters.

[0077] S6. Visualize and display the extracted dynamic parameters in the form of time curves or statistical distributions.

[0078] The dynamic single-event tracking method proposed in this embodiment has the following advantages compared with the single-event tracking methods in related technologies: (1) This embodiment integrates the spatial and temporal features of a single particle. This reduces the reliance on fixed thresholds, single-frame image intensity, or single morphological features, and improves the recognition stability of single particles under complex backgrounds and weak signal conditions.

[0079] (2) For single particles or molecules that are sparsely distributed in space, this embodiment utilizes the difference between the rapid change of single particle signal and the slow change of local background to suppress background drift and preserve the complete signal during the continuous existence of the object.

[0080] (3) In this embodiment, the local background is predicted for each combined event. By combining background subtraction and point spread function subpixel localization, the intensity, position and dwell time of the object can be continuously extracted.

[0081] (4) For cell adhesion plaques with complex morphology and dense distribution, this embodiment integrates original intensity, multi-scale spot structure and multi-directional ridge structure, which can adapt to adhesion plaques of different scales, morphologies and extension directions, and reduce missed detections.

[0082] (5) Adhesion speckle probability maps and StarDist maps are used to distinguish single particles that are close to each other, while Hungarianlinking is used to establish the correspondence between single particles in consecutive frames, thereby reducing object merging, trajectory breakage and faulty linking.

[0083] (6) This embodiment selects appropriate feature elements and processing methods for different single particles, and then realizes the identification and tracking from the sub-structural scale of single molecules to single cells under a unified processing logic.

[0084] like Figure 7As shown, this embodiment proposes a dynamic single-event tracking device, applicable to any of the above-mentioned dynamic single-event tracking methods. The device may include: Acquisition unit 101 is used to acquire raw image sequences for tracking the trajectory of a single particle; Selection unit 102 is used to select the corresponding target spatiotemporal feature elements based on the single-particle performance characteristics in the original image sequence; Tracking unit 103 is used to perform single-particle tracking in the original image sequence based on the spatiotemporal features of the target to obtain the single-particle motion trajectory; Extraction unit 104 is used to extract single-particle motion data along the single-particle motion trajectory; Output unit 105 is used to visualize the motion data of a single particle.

[0085] It should be noted that the processing procedures and beneficial effects of the acquisition unit 101, selection unit 102, tracking unit 103, extraction unit 104, and output unit 105 can be referred to respectively. Figure 1 Steps S101 to S105 are not described in detail here.

[0086] Optionally, selection unit 102 is also used for: If the single particles in the original image sequence are detected to be single particles or single molecules with sparse spatial distribution, then the correlation between the established single particle performance features and spatiotemporal feature elements is obtained. Based on the sparse spatial distribution characteristics, the corresponding spatiotemporal feature elements are found in the association relationship, namely the binding strength, point spread function (PSF), the location of the binding event in two-dimensional space, the pure background frame image before binding, and the image of the local background slowly changing over time, and are taken as the target spatiotemporal feature elements as a whole.

[0087] Optionally, the tracking unit 103 is also used for: The original image sequence is preprocessed to suppress noise, resulting in a preprocessed image sequence. Based on the binding strength features and PSF features of each frame in the preprocessed image sequence, the effective binding positions in each frame are determined. For any valid combination position, the valid combination position is used as the center point of the local image patch to be cropped. Based on the center point and the set first local size, local image patches are cropped in the target frame image including the valid combination position, the corresponding first N frame images, and the last M frame images to obtain a local image patch sequence. The first few local image patches in the local image patch sequence are used as sample image patches and the model is trained to obtain a trained center background prediction model. Using the center background prediction model, the background of the central region pixels of each non-sample image patch is predicted. Based on the background of the central region pixels of each non-sample image patch, the central region pixels of each non-sample image patch are purified to obtain the corresponding purified image patch sequence. Single-particle tracking is performed on the purified image patch sequence to obtain the single-particle motion trajectory corresponding to the valid combination position. Among them, the local image patch cropped in the target frame image is the non-sample image patch.

[0088] Optionally, the tracking unit 103 is also used for: For any frame image in the preprocessed image sequence, the frame image is input into the trained U-shaped convolutional neural network so that the U-shaped convolutional neural network: extracts the binding strength and PSF features in the frame image, generates and outputs a binding event probability map based on the proposed binding strength and PSF features, and marks the candidate binding positions with confidence levels exceeding the set confidence threshold in the binding event probability map. For any candidate combination position, determine the target frame image corresponding to the candidate combination position. If the spatial distribution characteristics of the candidate combination position in the target frame image are sparse and isolated, then perform a difference operation between the target frame image and the corresponding previous frame image to determine the gray level change value corresponding to the candidate combination position. If the gray level change value is greater than the set gray level threshold, then it is determined that the candidate combination position has a rapid instantaneous intensity jump, and the candidate combination position is determined to be a valid combination position.

[0089] Optionally, the tracking unit 103 is also used for: For any sample image block, the corresponding target center region is determined in the sample image block according to the set second local size, and the target center region is removed from the sample image block to obtain the sample image block after removal. A set number of consecutively removed sample image blocks are used as a sequence of sample image blocks and input into the model to be trained. The model then predicts the pixel background of the target center region of the next image block in the sample image block sequence based on the pixels in the sequence, and generates and outputs the prediction result. The parameters of the model to be trained are updated based on the prediction results until a well-trained center background prediction model is obtained.

[0090] Optionally, the tracking unit 103 is also used for: For any cleaned frame image in the cleaned image block sequence, perform two-dimensional centroid localization in the PSF of the cleaned frame image to obtain the sub-pixel level single particle position coordinates of the cleaned frame image. The single-particle position coordinates of each purified frame image in the purified image block sequence are sequentially correlated to obtain the single-particle motion trajectory corresponding to the effective combination position.

[0091] Optionally, the extraction unit 104 is also used for: For any cleaned frame image in the cleaned image block sequence, identify the Airy disk region in the cleaned frame image, and calculate the instantaneous image intensity of the cleaned frame image by integrating and averaging the pixel value of each pixel in the Airy disk region. Based on the instantaneous image intensity of each purified frame image, the dwell time from the first appearance of a single particle to its disappearance or the end of the acquisition is determined; The instantaneous image intensity, single-particle position coordinates, and dwell time of each purified frame image are used as a whole to extract single-particle motion data along the single-particle motion trajectory.

[0092] Optionally, selection unit 102 is also used for: If the single particles detected in the original image sequence are cell adhesion specks, then the target spatiotemporal feature elements are selected as the spatiotemporal feature elements extracted from the original image, Gaussian operator image, and Hessian operator image; Single-particle tracking is performed on the original image sequence based on the spatiotemporal features of the target to obtain the single-particle motion trajectory, including: For any frame image in the original image sequence, based on the frame image, a first Gaussian operator image and a second Gaussian operator image at a first set scale and a second set scale are generated respectively. A first Hessian operator image and a second Hessian operator image along the X direction at the first set scale and a second set scale are also generated respectively. A third Hessian operator image and a fourth Hessian operator image along the Y direction at the first set scale and a second set scale are also generated respectively. The first set scale is smaller than the second set scale. The frame image and the generated operator images are combined according to channels to obtain seven-channel data. The seven-channel data is input into a trained U-shaped convolutional neural network for processing to obtain a speckle probability map and a star distance StarDist map. The speckle probability map and StarDist map are synthesized to obtain a segmented speckle image. Multi-target data association is performed on the plaque images corresponding to consecutive frames to obtain the motion trajectory of each cell plaque.

[0093] Optionally, the extraction unit 104 is also used for: For any cell adhesion speckle's motion trajectory, the position coordinates and image intensity are recorded along the trajectory. The intensity jump event, random motion velocity, and radial velocity are determined based on the recorded position coordinates and image intensity, and are used as the overall motion data of the cell adhesion speckle.

[0094] The dynamic single-particle tracking device proposed in this embodiment can acquire an original image sequence for tracking the trajectory of a single particle, select corresponding target spatiotemporal feature elements based on the single particle performance characteristics in the original image sequence, perform single-particle tracking based on the target spatiotemporal feature elements in the original image sequence to obtain the single-particle trajectory, extract single-particle motion data along the single-particle trajectory, and visualize the single-particle motion data, effectively improving the accuracy of single-particle tracking.

[0095] In this embodiment, the dynamic single-particle tracking device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0096] This invention also provides a computer device having the above-described features. Figure 7 The dynamic single-particle tracking device shown.

[0097] Please see Figure 8 The present invention provides a schematic diagram of the structure of a computer device according to an optional embodiment. The computer device includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, multiple processors and / or multiple buses can be used with multiple memories, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 8 Take a processor 10 as an example.

[0098] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0099] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0100] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function. The data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0101] Memory 20 may include volatile memory, such as random access memory. Memory may also include non-volatile memory, such as flash memory, hard disk, or solid-state drive. Memory 20 may also include combinations of the above types of memory.

[0102] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0103] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic single-event tracking method, characterized in that, include: Acquire raw image sequences for tracking the trajectory of a single particle; Based on the single-particle performance characteristics in the original image sequence, select the corresponding target spatiotemporal feature elements; Based on the target spatiotemporal features, single-particle tracking is performed on the original image sequence to obtain the single-particle motion trajectory. Single-particle motion data is extracted along the single-particle trajectory, and the single-particle motion data is visualized and output.

2. The method according to claim 1, characterized in that, The step of selecting corresponding target spatiotemporal feature elements based on the single-particle performance characteristics in the original image sequence includes: If the single particles in the original image sequence are detected to be single particles or single molecules with sparse spatial distribution, then the correlation between the created single particle performance features and spatiotemporal feature elements is obtained. Based on the sparse spatial distribution characteristics, the corresponding spatiotemporal feature elements are found in the association relationship as the binding strength, point spread function (PSF), binding event location in two-dimensional space, pure background frame image before binding, and image of slow change of local background over time, and are used as the target spatiotemporal feature elements.

3. The method according to claim 2, characterized in that, The step of performing single-particle tracking based on the target spatiotemporal feature elements in the original image sequence to obtain the single-particle motion trajectory includes: The original image sequence is preprocessed to suppress noise, resulting in a preprocessed image sequence. Based on the binding strength features and PSF features of each frame image in the preprocessed image sequence, the effective binding positions in each frame image are determined; For any of the effective combination positions, the effective combination position is used as the center point of the local image block to be cropped. Based on the center point and the set first local size, local image blocks are cropped in the target frame image including the effective combination position, the corresponding first N frame images, and the last M frame images to obtain a local image block sequence. The first few local image blocks in the local image block sequence are used as sample image blocks and the model is trained to obtain a trained center background prediction model. The center background prediction model is used to predict the background of the central region pixels of each non-sample image block. Based on the background of the central region pixels of each non-sample image block, the central region pixels of each non-sample image block are purified to obtain the corresponding purified image block sequence. Single-particle tracking is performed on the purified image block sequence to obtain the single-particle motion trajectory corresponding to the effective combination position. The local image block cropped from the target frame image is the non-sample image block.

4. The method according to claim 3, characterized in that, The step of determining the effective binding position in each frame image based on the binding strength feature and PSF feature in each frame image of the preprocessed image sequence includes: For any frame image in the preprocessed image sequence, the frame image is input into a trained U-shaped convolutional neural network, so that the U-shaped convolutional neural network: extracts the binding strength and PSF features in the frame image, generates and outputs a binding event probability map based on the proposed binding strength and PSF features, and the binding event probability map is marked with candidate binding positions whose confidence exceeds a set confidence threshold; For any candidate combination position, determine the target frame image corresponding to the candidate combination position. If the spatial distribution feature of the candidate combination position in the target frame image is sparse and isolated, then perform a difference operation between the target frame image and the corresponding previous frame image to determine the grayscale change value corresponding to the candidate combination position. If the grayscale change value is greater than a set grayscale threshold, then determine that the candidate combination position has a rapid instantaneous intensity jump, and determine that the candidate combination position is a valid combination position.

5. The method according to claim 3, characterized in that, The step of using multiple local image patches located at the top of the local image patch sequence as sample image patches and training the model to obtain a trained center background prediction model includes: For any of the sample image blocks, a corresponding target center region is determined in the sample image block according to the set second local size, and the target center region is removed from the sample image block to obtain a sample image block after removal. A set number of consecutively removed sample image blocks are used as a sequence of sample image blocks and input into the model to be trained, so that the model to be trained can: predict the pixel background of the target center region of the next image block in the sample image block sequence based on the pixels in the sample image block sequence, and generate and output the prediction result. The parameters of the model to be trained are updated based on the prediction results until a well-trained center background prediction model is obtained.

6. The method according to claim 3, characterized in that, The step of performing single-particle tracking on the purified image patch sequence to obtain the single-particle motion trajectory corresponding to the effective binding position includes: For any purified frame image in the purified image block sequence, two-dimensional centroid localization is performed in the PSF of the purified frame image to obtain the sub-pixel level single particle position coordinates of the purified frame image. The single-particle position coordinates of each purified frame image in the purified image block sequence are sequentially associated to obtain the single-particle motion trajectory corresponding to the effective binding position.

7. The method according to claim 6, characterized in that, The extraction of single-particle motion data along the single-particle trajectory includes: For any purified frame image in the purified image block sequence, identify the Airy disk region in the purified frame image, and calculate the pixel value of each pixel in the Airy disk region by integration and average to obtain the instantaneous image intensity of the purified frame image. Based on the instantaneous image intensity of each purified frame image, the dwell time from the first appearance of a single particle to its disappearance or the end of the acquisition is determined; The instantaneous image intensity, single-particle position coordinates, and dwell time of each purified frame image are used as the single-particle motion data extracted along the single-particle motion trajectory.

8. The method according to claim 1, characterized in that, The step of selecting corresponding target spatiotemporal feature elements based on the single-particle performance characteristics in the original image sequence includes: If the single particles detected in the original image sequence are cell adhesion specks, then the target spatiotemporal feature elements are selected as the spatiotemporal feature elements extracted from the original image, Gaussian operator image, and Hessian operator image; The step of performing single-particle tracking based on the target spatiotemporal feature elements in the original image sequence to obtain the single-particle motion trajectory includes: For any frame image in the original image sequence, based on the frame image, a first Gaussian operator image and a second Gaussian operator image at a first set scale and a second set scale are generated respectively. A first Hessian operator image and a second Hessian operator image along the X direction at the first set scale and the second set scale are generated respectively. A third Hessian operator image and a fourth Hessian operator image along the Y direction at the first set scale and the second set scale are generated respectively. The first set scale is smaller than the second set scale. The frame image and the generated operator images are combined according to channels to obtain seven-channel data. The seven-channel data is input into a trained U-shaped convolutional neural network for processing to obtain a speckle probability map and a star distance StarDist map. The speckle probability map and the StarDist map are synthesized to obtain a segmented speckle image. Multi-target data association is performed on the adhesion plaque images corresponding to the consecutive frame images to obtain the motion trajectory of each cell adhesion plaque.

9. The method according to claim 8, characterized in that, The extraction of single-particle motion data along the single-particle trajectory includes: For any of the cell adhesion spots' motion trajectories, the position coordinates and image intensity are recorded along the motion trajectory of the cell adhesion spot. The intensity jump event, random motion speed, and radial velocity are determined based on the recorded position coordinates and image intensity, and are used as the overall motion data of the cell adhesion spot.

10. A dynamic single-particle tracking device, characterized in that, The apparatus used in the dynamic single-particle tracking method according to any one of claims 1 to 8 includes: The acquisition unit is used to acquire raw image sequences for tracking the trajectory of a single particle. The selection unit is used to select the corresponding target spatiotemporal feature elements based on the single-particle performance characteristics in the original image sequence. The tracking unit is used to perform single-particle tracking in the original image sequence based on the spatiotemporal features of the target to obtain the single-particle motion trajectory. The extraction unit is used to extract single-particle motion data along the single-particle motion trajectory; The output unit is used to visualize the single-particle motion data.

Images