A blood coagulation process dynamic image recognition inspection method and device fusing timing characteristics

Abstract

The application discloses a blood coagulation process dynamic image recognition inspection method and device fusing time sequence features. The method comprises the following steps: acquiring a continuous microscopic dynamic gray image sequence of a blood coagulation process; performing image preprocessing on the image sequence through full-time sequence space-time anchoring, so as to acquire a denoised space-time constrained binary image sequence; performing time sequence irreversible local connected feature coding on the binary image sequence, so as to acquire a globally ordered time sequence topological feature sequence; performing fixed rule recognition on the feature sequence to obtain a coagulation phase change inflection point, so as to acquire a coagulation key time node; and performing quantitative determination of a coagulation function inspection, so as to acquire a coagulation function quantitative index. The method is closely related to the physical nature of the one-way irreversible phase change of blood coagulation, directly encodes full-time sequence dynamic features into irreversible topological features, realizes deep fusion of the coagulation physiological process and image recognition, naturally filters reverse fluctuation interference through irreversible feature coding, and avoids pseudo features and pseudo inflection points from the root.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, specifically to a method and apparatus for dynamic image recognition and testing of blood coagulation process that integrates temporal features. Background Technology

[0002] Blood coagulation function testing is a core component of preoperative clinical screening, diagnosis of hemorrhagic / thrombotic diseases, and monitoring of anticoagulant medications. Currently, two main technical approaches exist in clinical practice and research, both of which have significant limitations:

[0003] Coagulation methods (such as manual methods and semi-automatic coagulation analyzers): rely on manual observation or mechanical probes to detect the formation of fibrin clots. The operation is cumbersome, has poor repeatability, and can only determine a single coagulation endpoint, and cannot obtain dynamic information of the entire coagulation process.

[0004] Optical turbidimetry / scattering turbidimetry: indirectly reflects the coagulation process by detecting changes in plasma turbidity, but it is easily affected by background interference from samples with jaundice, lipemia, hemolysis, etc., and has insufficient sensitivity for samples with low fibrinogen.

[0005] Magnetic bead method: The solidification point is determined by the change in the vibration amplitude of magnetic beads. Although it can partially overcome optical interference, it has extremely high requirements for the quality of magnetic beads and the accuracy of the equipment vibration module, the equipment maintenance cost is high, and it can only obtain limited time point information. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic image recognition and verification method for blood coagulation process that integrates temporal features to at least solve one of the above-mentioned technical problems.

[0007] One aspect of the present invention provides a method for dynamic image recognition and verification of blood coagulation process by fusing temporal features, the method comprising:

[0008] Acquire a continuous microscopic dynamic grayscale image sequence of the blood coagulation process;

[0009] Full-time spatiotemporal anchoring image preprocessing is performed on the continuous microscopic dynamic grayscale image sequence of the blood coagulation process to obtain a denoised spatiotemporally constrained binarized image sequence;

[0010] Temporally irreversible local connectivity feature encoding is performed on the denoised spatiotemporally constrained binarized image sequence to obtain single-grid irreversible connectivity feature sequence and globally sorted temporal topological feature sequence;

[0011] Fixed rule identification of coagulation phase transition inflection points is performed on the global sorted temporal topological feature sequence to obtain key coagulation time nodes;

[0012] Quantitative assessment of coagulation function at key time points is conducted to obtain quantitative indicators of coagulation function.

[0013] Optionally, the image preprocessing of the continuous microscopic dynamic grayscale image sequence of the blood coagulation process with full-time spatiotemporal anchoring to obtain a denoised spatiotemporally constrained binarized image sequence includes:

[0014] Background anchoring based on temporal gray-level accumulation peaks is performed on a continuous microscopic dynamic gray-level image sequence of the blood coagulation process to obtain a global fixed background threshold;

[0015] Binarization of a continuous microscopic dynamic grayscale image sequence of the blood coagulation process is performed based on a globally fixed background threshold and neighborhood temporal consistency constraint to obtain a binarized image sequence.

[0016] Denoising of the binarized image sequence using spatiotemporal three-dimensional fixed structuring elements is performed to obtain a denoised spatiotemporally constrained binarized image sequence.

[0017] Optionally, the step of encoding temporally irreversible local connectivity features on the denoised spatiotemporally constrained binarized image sequence to obtain a single-grid irreversible connectivity feature sequence and a globally ordered temporal topological feature sequence includes:

[0018] Each frame of the denoised spatiotemporally constrained binarized image sequence is divided into spatiotemporally bound grids to obtain a full-time spatiotemporal grid cell set.

[0019] The temporal differential cumulative duty cycle is calculated for the full-time spatiotemporal grid cell set to obtain the temporal differential cumulative duty cycle sequence for each spatiotemporal grid cell;

[0020] Based on the temporal differential cumulative duty cycle sequence of each spatiotemporal grid cell, the final irreversible features are constructed through a one-way jump locking mechanism, thereby obtaining the single-grid irreversible connectivity feature sequence and the globally sorted temporal topology feature sequence.

[0021] Optionally, the step of identifying the fixed rule of the coagulation phase transition inflection point in the globally sorted temporal topological feature sequence to obtain key coagulation time nodes includes:

[0022] The topological entropy slope is linearized by fixed segmentation on the single-grid irreversible connected feature sequence and the globally sorted temporal topological feature sequence to obtain the full-time normalized topological entropy curve and the topological entropy fixed segmentation slope sequence.

[0023] The coagulation initiation point is obtained based on a fixed segmented slope sequence of topological entropy.

[0024] The coagulation completion point and fibrinolysis initiation point are generated based on the first coagulation initiation point and the fixed segmented slope sequence of topological entropy; the coagulation initiation point, coagulation completion point and fibrinolysis initiation point constitute the key time nodes of coagulation.

[0025] Optionally, the quantitative determination of coagulation function at key coagulation time points to obtain quantitative indicators of coagulation function includes:

[0026] For each spatial grid, the coagulation rate is calculated based on the initial features at the start time to obtain the final coagulation rate index.

[0027] For each spatial grid, the agglomeration uniformity is calculated by a fixed combination of basic spatial uniformity and topological distribution entropy, thereby obtaining the final agglomeration uniformity index.

[0028] The fibrinolytic activity is calculated and determined by fixed logic based on the combination of reverse irreversible characteristics and topological entropy at the fibrinolysis initiation point, thereby obtaining the final fibrinolytic activity index.

[0029] Optionally, the step of performing background anchoring based on the temporal gray-level accumulation peak of the continuous microscopic dynamic gray-level image sequence of the blood coagulation process to obtain a globally fixed background threshold includes:

[0030] For the input sequence of continuous microscopic dynamic grayscale images of blood coagulation, the number of pixels corresponding to each grayscale level g is counted frame by frame to obtain the grayscale histogram of a single frame. ;

[0031] To construct a time-series cumulative gray-level histogram, sum the histograms of all single frames bit by bit. ;

[0032] Traversal Find all gray levels g that make The grayscale value that yields the maximum value is the unique global fixed background threshold. .

[0033] Optionally, the binarization of the continuous microscopic dynamic grayscale image sequence of the blood coagulation process according to a globally fixed background threshold with neighborhood temporal consistency constraints to obtain a binarized image sequence includes:

[0034] The following operations were performed on each frame of the continuous microscopic dynamic grayscale image sequence of the blood coagulation process:

[0035] Check the grayscale value I(x,y,t) of the current pixel (x,y) in frame t. If I(x,y,t) ≥ If I(x,y,t) < Proceed to the next step of temporal neighborhood determination;

[0036] Spatial neighborhood extraction between previous and next frames:

[0037] Extract the 3×3 spatial neighborhood grayscale set centered at (x,y) in the previous frame. ;

[0038] Extract the 3×3 spatial neighborhood grayscale set N(x,y,t+1) centered at (x,y) in the next frame;

[0039] Neighborhood consistency determination and binarization:

[0040] statistics Medium grayscale value less than Number of pixels ;

[0041] Statistically count the gray values ​​less than in N(x,y,t+1) Number of pixels ;

[0042] If both conditions are met and If the value is true, then mark the current pixel (x,y,t) as a foreground pixel; otherwise, mark it as a background pixel.

[0043] For boundary frames, directly let Finally, a binarized image sequence is obtained. .

[0044] Optionally, the step of performing spatiotemporal three-dimensional fixed structuring element denoising on the binarized image sequence to obtain a denoised spatiotemporally constrained binarized image sequence includes:

[0045] Construct a 3×3×3 spatiotemporal three-dimensional fixed structural element;

[0046] Spatiotemporal morphological opening is performed on the binarized image sequence B(t) to obtain the denoised spatiotemporally constrained binarized image sequence.

[0047] Optionally, performing a spatiotemporal morphological opening operation on the binarized image sequence B(t) to obtain a denoised spatiotemporally constrained binarized image sequence includes:

[0048] For each pixel (x,y,t) in the binary sequence B(t), extract all 27 pixel values ​​within a 3×3×3 spatiotemporal neighborhood centered on that pixel;

[0049] If all 27 pixel values ​​are 1, then the eroded pixel values ​​will be... Assign a value of 1; otherwise, assign a value of 0.

[0050] For each etched pixel value The etched sequence For each pixel (x,y,t) in the dataset, extract all 27 pixel values ​​within a 3×3×3 spatiotemporal neighborhood centered on that pixel.

[0051] If at least one of these 27 pixel values ​​is 1, then the dilated pixel value will be... The value is assigned to 1; otherwise, it is assigned to 0. The dilated pixel values ​​form the denoised binary image sequence.

[0052] This application also provides a dynamic image recognition and testing device for blood coagulation process that integrates temporal features, the dynamic image recognition and testing device for blood coagulation process that integrates temporal features includes:

[0053] A continuous microscopic dynamic grayscale image sequence acquisition module for the blood coagulation process, wherein the continuous microscopic dynamic grayscale image sequence acquisition module for the blood coagulation process is used to acquire a continuous microscopic dynamic grayscale image sequence for the blood coagulation process;

[0054] A denoised spatiotemporally constrained binarized image sequence acquisition module is used to perform full-time spatiotemporally anchored image preprocessing on a continuous microscopic dynamic grayscale image sequence of the blood coagulation process, thereby obtaining a denoised spatiotemporally constrained binarized image sequence.

[0055] A temporally irreversible local connectivity feature encoding module is used to encode temporally irreversible local connectivity features on a denoised spatiotemporally constrained binarized image sequence, thereby obtaining a single-grid irreversible connectivity feature sequence and a globally sorted temporal topological feature sequence.

[0056] The coagulation key time node acquisition module is used to identify the fixed rule of coagulation phase transition inflection point in the global sorted temporal topological feature sequence, thereby acquiring the coagulation key time node.

[0057] The coagulation function quantitative index acquisition module is used to quantitatively determine the coagulation function at key time points of coagulation, thereby acquiring the coagulation function quantitative index.

[0058] The dynamic image recognition and verification method for blood coagulation process based on the fusion of temporal features in this application closely adheres to the physical nature of the unidirectional irreversible phase transition of blood coagulation. It directly encodes the full-time dynamic features into irreversible topological features, achieving deep integration of the coagulation physiological process and image recognition. The irreversible feature encoding naturally filters out reverse fluctuation interference, avoiding false features and false inflection points from the root. Through the design of full-process spatiotemporal constraints, the anti-interference ability is greatly improved, and it has extremely strong tolerance to common clinical interference samples such as jaundice and lipemia, resulting in significantly improved result stability.

[0059] This application provides accurate and delay-free inflection point identification with completely unified judgment criteria, solving the industry problem of incomparable test results in different scenarios; the quantitative indicators comprehensively cover the core dimensions of coagulation and can identify occult coagulation abnormalities; at the same time, it has low hardware barriers, strong interpretability, and outstanding compliance and industrialization value. Attached Figure Description

[0060] Figure 1 This is a flowchart illustrating a method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features, according to an embodiment of this application. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0062] like Figure 1 The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features, as shown, includes:

[0063] Acquiring a sequence of continuous microscopic dynamic grayscale images of the blood coagulation process T is the total test duration, the sampling interval is fixed Δt, the total number of frames N = T / Δt, and the resolution of a single frame is fixed M×M;

[0064] Full-time spatiotemporal anchoring image preprocessing is performed on the continuous microscopic dynamic grayscale image sequence of the blood coagulation process to obtain a denoised spatiotemporally constrained binarized image sequence;

[0065] Temporally irreversible local connectivity feature encoding is performed on the denoised spatiotemporally constrained binarized image sequence to obtain single-grid irreversible connectivity feature sequence and globally sorted temporal topological feature sequence;

[0066] Fixed rule identification of coagulation phase transition inflection points is performed on the global sorted temporal topological feature sequence to obtain key coagulation time nodes;

[0067] Quantitative assessment of coagulation function at key time points is conducted to obtain quantitative indicators of coagulation function.

[0068] In this embodiment, the image preprocessing of the continuous microscopic dynamic grayscale image sequence of the blood coagulation process with full-time spatiotemporal anchoring to obtain a denoised spatiotemporally constrained binarized image sequence includes:

[0069] Background anchoring based on temporal gray-level accumulation peaks is performed on a continuous microscopic dynamic gray-level image sequence of the blood coagulation process to obtain a global fixed background threshold;

[0070] In this embodiment, background anchoring based on temporal gray-level accumulation peaks is performed on a continuous microscopic dynamic gray-level image sequence of the blood coagulation process to obtain a globally fixed background threshold, including:

[0071] Single-frame grayscale histogram statistics: For the input blood coagulation continuous microscopic dynamic grayscale image sequence I(t) (t=1,2,...,N, where N is the total number of frames, and each frame is an 8-bit grayscale image with a grayscale range of [0,255]), the number of pixels corresponding to each grayscale level g is counted frame by frame to obtain the single-frame grayscale histogram. ( This represents the total number of pixels with a grayscale value of g in frame t.

[0072] Temporal cumulative gray-level histogram construction: Accumulate the histograms of all single frames bit by bit according to their gray levels to construct the temporal cumulative gray-level histogram. The formula is:

[0073] ;

[0074] That is, for each gray level g, the number of pixels of that gray level in all frames of the entire time sequence is added together to obtain the total number of pixels.

[0075] Determining a globally fixed background threshold: Traversal Find all gray levels g that make The grayscale value that yields the maximum value is the unique global fixed background threshold. (Stable grayscale features corresponding to the plasma background).

[0076] In this embodiment, the application uses a time-series cumulative grayscale histogram instead of a single-frame histogram and a multi-frame average histogram for background threshold anchoring.

[0077] Traditional single-frame histograms are susceptible to interference from instantaneous blood cell movement, sampling noise, and slight sample shaking, which can cause background peaks to become blurred, shift, or even disappear. While multi-frame average histograms can alleviate this to some extent, they are still affected by a few abnormal frames (such as a sample suddenly shaking in a certain frame), resulting in insufficient threshold stability.

[0078] This application takes into account that the plasma background persists and accounts for the largest proportion throughout the entire time-series inspection. Through full-time accumulation, the number of pixels corresponding to its grayscale level will be significantly enhanced. It forms an extremely sharp main peak, unaffected by instantaneous disturbances;

[0079] The grayscale features of moving blood cells and transient noise are effectively diluted by averaging during accumulation because their positions and grayscale values ​​are random across different frames. Interference peaks are formed in the middle;

[0080] The entire process only requires calculating the cumulative histogram once and taking the main peak. The rules are completely fixed, with no dynamic adjustments and no threshold. Its stability is far superior to that of traditional methods.

[0081] Binarization of a continuous microscopic dynamic grayscale image sequence of the blood coagulation process is performed based on a globally fixed background threshold and neighborhood temporal consistency constraint to obtain a binarized image sequence.

[0082] In this embodiment, the binarization of a continuous microscopic dynamic grayscale image sequence of the blood coagulation process based on a globally fixed background threshold and neighborhood temporal consistency constraints is performed to obtain a binarized image sequence, including:

[0083] For each pixel (x,y) of each frame I(t) (x,y are pixel coordinates, ranging from [1,M], where M is the resolution of a single frame), binarization is performed according to the following fixed rules:

[0084] Current frame pixel grayscale determination: Check the grayscale value I(x,y,t) of the current pixel (x,y) in frame t. If I(x,y,t) ≥ Directly mark it as a background pixel (assign value 0); if I(x,y,t)< Then proceed to the next step of temporal neighborhood determination.

[0085] Spatial neighborhood extraction between previous and next frames:

[0086] Extract the 3×3 spatial neighborhood grayscale set N(x,y,t−1)={I(x−1,y−1,t−1),I(x,y−1,t−1),...,I(x+1,y+1,t−1)} centered at (x,y) in the previous frame (frame t−1) (a total of 9 pixels; if the boundary pixels' neighborhood exceeds the image range, the excess part is treated as background grayscale by default). filling);

[0087] Extract the 3×3 spatial neighborhood grayscale set N(x,y,t+1) centered at (x,y) in the next frame (frame t+1), using the same extraction rules as the previous frame.

[0088] Neighborhood consistency determination and binarization:

[0089] Statistically, in N(x,y,t−1), the gray values ​​less than Number of pixels ;

[0090] Statistically count the gray values ​​less than in N(x,y,t+1) Number of pixels ;

[0091] If both conditions are met ≥5 and If the value is ≥5 (fixed neighborhood consistency threshold), then the current pixel (x,y,t) is marked as a foreground pixel (assigned a value of 1); otherwise, it is marked as a background pixel (assigned a value of 0).

[0092] Boundary frame processing: For the first frame (t=1), there is no previous frame, so directly set... = (That is, only the neighborhood of the next frame is referenced); for the Nth frame (t=N), there is no next frame, so let = (That is, only the neighborhood of the previous frame is referenced), and finally the binarized image sequence B(t) is obtained.

[0093] This application combines spatial neighborhood constraints with temporal neighborhood constraints through a dual fixed rule of initial grayscale screening of the current frame and secondary screening of 3×3 spatial neighborhood consistency of the preceding and following frames.

[0094] Traditional single-frame binarization relies solely on the grayscale of the current pixel or its neighborhood, which can easily lead to misjudgments in three categories: instantaneous noise points (low grayscale in a single frame but no corresponding features in preceding and following frames); moving single blood cells (low grayscale in a single frame but positional shifts in preceding and following frames, with no consistency in the neighborhood); and weak fibrin clots formed in the early stages, which are easily missed (unobvious neighborhood features in a single frame, but a continuous trend exists in preceding and following frames).

[0095] This application ensures the local spatial continuity of the foreground through spatial neighborhood constraints (3×3), avoiding misjudgment of isolated noise points; it ensures the temporal continuity of the foreground through temporal neighborhood constraints (previous and subsequent frames). Only areas with low grayscale in the current frame and sufficient low grayscale pixels in the neighboring frames before and after are judged as true clots or blood cell aggregations, filtering artifacts from both spatiotemporal dimensions; a fixed threshold of 5 (majority of the 3×3 neighborhood) ensures the strictness of neighborhood consistency.

[0096] Denoising of the binarized image sequence using spatiotemporal three-dimensional fixed structuring elements is performed to obtain a denoised spatiotemporally constrained binarized image sequence.

[0097] In this embodiment, performing spatiotemporal three-dimensional fixed structuring element denoising on the binarized image sequence to obtain a denoised spatiotemporally constrained binarized image sequence includes:

[0098] Construct a 3×3×3 spatiotemporal three-dimensional fixed structuring element (corresponding to 3 pixels each in spatial dimensions x and y, and 1 frame before and after time dimension t, covering a total of 27 pixels). Perform a spatiotemporal morphological opening operation (erosion followed by dilation) on the binary sequence B(t). Specific operations are as follows:

[0099] Spatiotemporal three-dimensional erosion operation:

[0100] For each pixel (x,y,t) in the binary sequence B(t) (t ranges from 2 to N−1, and the original values ​​are directly preserved at boundary frames t=1 and t=N), extract all 27 pixel values ​​in the 3×3×3 spatiotemporal neighborhood centered on that pixel (spatial: x−1 to x+1, y−1 to y+1; temporal: t−1 to t+1).

[0101] If all 27 pixel values ​​are 1, then the eroded pixel values ​​will be... Assign a value of 1; otherwise, assign a value of 0.

[0102] Spatiotemporal three-dimensional dilation operation:

[0103] The etched sequence Each pixel (x, y, t) in the data (t ranges from 2 to N−1, and boundary frames are directly preserved) (Original value), and then extract all 27 pixel values ​​within the 3×3×3 spatiotemporal neighborhood centered on that pixel;

[0104] If at least one of these 27 pixel values ​​is 1, then the dilated pixel value will be... Assign a value of 1; otherwise, assign a value of 0.

[0105] Finally, the denoised binarized image sequence is obtained. .

[0106] This application uses a 3×3×3 spatiotemporal three-dimensional fixed structural element to replace the traditional single-frame two-dimensional structural element, extending morphological operations from the spatial dimension to the three-dimensional dimensions of space and time.

[0107] Traditional two-dimensional morphological opening operations (3×3 structuring elements per frame) can only remove isolated single-pixel noise in the spatial dimension, but cannot handle two types of noise: small noise blocks within a single frame (such as noise blocks smaller than 3×3 that appear in a frame due to sampling interference, which cannot be completely removed by two-dimensional erosion); and instantaneous noise frames in the temporal dimension (such as artifacts that suddenly appear in a frame, which will still be retained after two-dimensional denoising).

[0108] This application requires that three consecutive frames + 3×3 spatial neighborhoods be all 1 to retain pixels through three-dimensional erosion operation. It can simultaneously remove spatially isolated noise (single-frame single-pixel) and temporally isolated noise (single-frame small-area artifacts). Only truly spatiotemporally continuous clot regions will be preserved.

[0109] The three-dimensional expansion operation of this application can restore the real clot area that has been slightly reduced by the corrosion operation, ensuring the integrity of the clot morphology; the size of the structural element is fixed at 3×3×3, the corrosion expansion rule is completely fixed, there is no adaptive structural element adjustment, and the noise reduction effect is stable and reproducible.

[0110] In this embodiment, the step of performing temporally irreversible local connectivity feature encoding on the denoised spatiotemporally constrained binarized image sequence to obtain a single-grid irreversible connectivity feature sequence and a globally ordered temporal topological feature sequence includes:

[0111] Each frame of the denoised spatiotemporally constrained binarized image sequence is divided into spatiotemporally bound grids to obtain a full-time spatiotemporal grid cell set.

[0112] In this embodiment, the spatiotemporal constrained binarized image sequence after denoising is divided into spatiotemporal neighborhood-bound grids to obtain a full-time spatiotemporal grid cell set, including:

[0113] Fixed spatial grid division: Binarize the denoised image of each frame. (A single frame has a fixed resolution of M×M; in this scheme, M is taken as the typical resolution of a microscopic image, 256, and remains fixed throughout.) The image is divided into S×S fixed-size, non-overlapping spatial grids (S is taken as 16 pixels, and remains fixed throughout). The entire image has a total of K=(M / S)×(M / S)=16×16=256 spatial grids. Let the i-th spatial grid be denoted as... (i=1,2,...,256, grid indices are arranged in a fixed order from left to right and from top to bottom).

[0114] Spatiotemporal neighborhood binding: for each spatial grid For each time t, bind it to the corresponding spatial grid of the previous frame t−1 and the next frame t+1 to form a spatiotemporal grid cell C(i,t):

[0115] When t=2,3,...,N−1 (N is the total number of frames), C(i,t) covers 3 frames × S × S pixels, that is... (t−1), (t), (t+1) area;

[0116] When t=1 (first frame), there is no preceding frame, and C(i,1) covers 2 frames × S × S pixels, that is... (1) (2) area;

[0117] When t=N (last frame), there are no subsequent frames, and C(i,N) covers 2 frames × S × S pixels, that is... (N−1) (N) area.

[0118] This application expands the single-frame spatial grid into a spatiotemporal grid unit that binds the spatial grids of the preceding and following frames, breaking the traditional approach of processing the single-frame spatial grid independently.

[0119] Traditional spatial grids only focus on local features within a single frame, ignoring the continuous evolution of clots over time (such as fibrin clots, which grow from points to lines to grids in a continuous spatiotemporal process), resulting in features that cannot reflect the temporal nature of coagulation.

[0120] After spatiotemporal binding, each unit in this application naturally contains local information about the past, present, and future of the block, providing a spatiotemporal correlation basis for subsequent temporal feature encoding;

[0121] A fixed spatial grid size S=16 and a fixed grid indexing rule are used to ensure that the feature extraction benchmarks of all samples are completely consistent, without any adaptive grid division.

[0122] The fixed binding rule of boundary frames (first frame bound to the following frame, last frame bound to the previous frame) ensures the integrity of the algorithm, with no information loss, and ultimately obtains a full-time spatiotemporal grid cell set: .

[0123] The temporal differential cumulative duty cycle is calculated for the full-time spatiotemporal grid cell set to obtain the temporal differential cumulative duty cycle sequence for each spatiotemporal grid cell;

[0124] In this embodiment, the temporal differential cumulative duty cycle is calculated for the full-time spatiotemporal grid cell set to obtain the temporal differential cumulative duty cycle sequence for each spatiotemporal grid cell, including:

[0125] Using the spatiotemporal grid cell C(i,t) as the basic unit, the time-series differential cumulative duty cycle P(i,t) is calculated according to the following fixed rules:

[0126] Current frame spatial foreground duty cycle calculation: For each C(i,t), first extract its corresponding spatial grid of the current frame. exist In the set of pixels (t), count the total number of foreground pixels (assigned a value of 1), and denote it as . ; Calculate the spatial foreground duty cycle of the current frame:

[0127] ;

[0128] Since S=16, therefore S×S=256, which is fixed throughout.

[0129] Obtaining the spatial foreground duty cycle of the previous frame: For each C(i,t), extract the spatial grid corresponding to its previous frame. Foreground duty cycle in B′(t−1) :

[0130] When t=1, there is no previous frame, let (i,0)=0 (fixed initial value);

[0131] When t≥2 (i,t−1) is the value calculated at the previous time step. .

[0132] Forward difference accumulation calculation: Initialize P(i,0)=0 (fixed initial value), and for each t=1,2,...,N, calculate P(i,t) according to the following formula:

[0133] ;

[0134] That is, only the positive difference between the current frame's spatial duty cycle and the previous frame's spatial duty cycle is accumulated. If the difference is negative or zero, it is not accumulated, and P(i,t) remains consistent with P(i,t−1).

[0135] This application proposes a mechanism for accumulating the spatial duty cycle of the current frame versus the previous frame in a positive differential manner, replacing the traditional method of directly accumulating the current frame duty cycle or weighted accumulating the historical duty cycle.

[0136] Traditional accumulation methods are affected by two types of interference: the instantaneous decrease in the single-frame spatial duty cycle caused by Brownian motion of blood cells (reverse fluctuation); and the fluctuation of the duty cycle caused by the random jitter of the initial weak clot. Ultimately, the accumulation characteristics cannot strictly reflect the unidirectional irreversibility of coagulation.

[0137] This application only accumulates the positive difference, naturally filtering out all negative fluctuations, ensuring that P(i,t) is a strictly monotonically non-decreasing sequence, and mathematically directly encoding the unidirectional irreversible nature of coagulation;

[0138] The spatial duty cycle of the current frame is subtracted from the spatial duty cycle of the previous frame (instead of subtracting the cumulative value P(i,t−1) from the previous time step) to ensure that the difference reflects the actual local clot growth and avoids the amplification effect of the cumulative value itself; the final output is the temporal difference cumulative duty cycle sequence for each spatiotemporal grid cell: .

[0139] Based on the temporal differential cumulative duty cycle sequence of each spatiotemporal grid cell, the final irreversible features are constructed through a one-way jump locking mechanism, thereby obtaining the single-grid irreversible connectivity feature sequence and the globally sorted temporal topology feature sequence.

[0140] In this embodiment, based on the temporal differential cumulative duty cycle sequence of each spatiotemporal grid cell, the final irreversible features are constructed through a one-way jump locking mechanism, thereby obtaining the single-grid irreversible connectivity feature sequence and the globally sorted temporal topological feature sequence, including:

[0141] Based on P(i,t), the final irreversible feature D(i,t) is constructed through a one-way jump locking mechanism, and a global topological feature sequence is generated:

[0142] Fixed jump threshold setting: Set a fixed jump threshold =0.05 (duty cycle range is [0,1], 0.05 corresponds to a positive increase of 5% in the local clot duty cycle, which is fixed throughout).

[0143] One-way jump locking feature calculation: initialization =0 (fixed initial value), for each i=1,2,...,256 and t=1,2,...,N, calculate according to the following rules. :

[0144] If the current time satisfies If the positive difference exceeds the transition threshold, a one-way transition lock is triggered.

[0145] ;

[0146] in For a fixed jump range;

[0147] If a transition lock is not triggered (including positive difference ≤ (or growth in subsequent time intervals after triggering), then:

[0148] ;

[0149] That is, directly accumulate the increment of P(i,t);

[0150] Core constraint: Regardless of how P(i,t) changes subsequently, D(i,t) will never be less than D(i,t−1), strictly guaranteeing monotonically non-decreasing.

[0151] Global topological feature sorting: For each time t, sort the irreversible features D(i,t) of all 256 grids in ascending order to obtain the global temporal topological feature sequence. ;

[0152] in .

[0153] The one-way jump locking mechanism designed in this application directly jumps the feature value and locks the irreversibility of subsequent growth when the positive growth of local agglomerates exceeds a fixed threshold, replacing the traditional smooth feature or gradient feature method.

[0154] Traditional features are not sensitive to key phase transition points in the coagulation process (the turning point from slow aggregation of blood cells to rapid cross-linking of fibrin into a network), smooth features can obscure the timing of phase transitions, and gradient features are susceptible to noise interference, resulting in spurious peaks.

[0155] Jump locking can highlight key phase transition points: when a local agglomerate enters a rapid growth phase, the positive differential exceeds... D(i,t) will show a significant jump, forming a clear inflection point marker in the feature sequence;

[0156] The strictly monotonically non-decreasing D(i,t) further reinforces the temporal irreversibility, providing a stable, unidirectional, and fluctuation-free input for the subsequent topological entropy calculation. The final output is a single-grid irreversible connected feature sequence: D(i,t), i=1,2,...,256, t=1,2,...,N;

[0157] Globally sorted temporal topological feature sequence: , t=1,2,...,N.

[0158] In this embodiment, the step of identifying the fixed rule of the coagulation phase transition inflection point in the globally sorted temporal topological feature sequence to obtain key coagulation time nodes includes:

[0159] The topological entropy slope is linearized by fixed segmentation on the single-grid irreversible connected feature sequence and the globally sorted temporal topological feature sequence to obtain the full-time normalized topological entropy curve and the topological entropy fixed segmentation slope sequence.

[0160] In this embodiment, the topological entropy slope of the single-grid irreversible connected feature sequence and the globally sorted temporal topological feature sequence is linearized by fixed segmentation, thereby obtaining the full-time normalized topological entropy curve and the topological entropy fixed segmentation slope sequence, including:

[0161] Global topological feature maximum normalization: global sorted topological feature sequence for each time t (A total of 256 mesh features, arranged in ascending order), first calculate the maximum value at that moment. ;like (If the initial time may be checked), then let the normalized features Otherwise, perform maximum value normalization for each feature:

[0162] ,i=1,2,...,256;

[0163] After normalization ∈[0,1].

[0164] Fixed-width binning and frequency statistics: The interval [0,1] is divided into L=10 fixed-width bins (fixed throughout), each bin having a width of 0.1, and the bin intervals are [0,0.1), [0.1,0.2),...,[0.9,1]; for each time t, the frequency statistics are... The number of grid cells falling into the j-th bin is denoted as the frequency. ,satisfy .

[0165] Normalized topological entropy calculation: Set a fixed minimum value (To avoid the logarithm being meaningless, the value is fixed throughout), calculate the normalized topological entropy E(t) for each time t:

[0166] ;

[0167] Physical meaning: In the initial liquid stage of coagulation, the grid feature distribution is uniform, and E(t) is close to the maximum value ln(10)≈2.3; after coagulation starts, the clot gradually forms, the feature distribution is uneven, and E(t) decreases monotonically; after the clot is fully formed, E(t) drops to the minimum value and remains stable; after fibrinolysis starts, the clot dissolves, the feature distribution tends to be uniform again, and E(t) rises.

[0168] Fixed time interval segmentation and slope calculation: Set a fixed segmentation time interval ΔT = 10 frames (corresponding to a sampling interval Δt = 0.1 seconds, ΔT = 1 second, fixed throughout), and divide the full time series t = 1, 2, ..., N into segments. The k-th segment starts at the following time: (⌊⌋ represents floor function).

[0169] For each segment k, calculate the fixed slope of that segment. :

[0170] ;

[0171] If the last segment is less than ΔT frames, it is discarded and not included in the slope calculation.

[0172] This application adopts a dual fixed rule of fixed equal-width bin topology entropy and fixed time interval piecewise linearization slope to replace the traditional continuous frame differential entropy slope or adaptive bin entropy method.

[0173] The slope of traditional continuous frame differential entropy is easily affected by the fluctuation of single frame features (such as the small jump of individual grid D(i,t), resulting in a large number of spurious peaks; adaptive binning will lead to inconsistent entropy calculation benchmarks due to the differences in feature distribution of different samples, making it impossible to uniformly determine the inflection point.

[0174] This application uses fixed equal-width bins (L=10) to ensure that the entropy calculation benchmark of all samples is completely consistent, eliminating the error caused by bin differences; fixed time interval segmentation (ΔT=10 frames) is used to perform smoothing linear fitting on the entropy curve, naturally filtering out single-frame fluctuations, and the slope only reflects the overall entropy change trend within a time period, with extremely strong noise resistance.

[0175] The final output is the full-time normalized topological entropy curve: E(t), t=1,2,...,N; the topological entropy fixed segmented slope sequence is: Corresponding to the start time .

[0176] The coagulation initiation point is obtained based on a fixed segmented slope sequence of topological entropy.

[0177] In this embodiment, obtaining the coagulation initiation point based on the fixed segmented slope sequence of topological entropy includes:

[0178] Fixed threshold setting: Set a fixed threshold for coagulation activation. (The unit of entropy change rate, since E(t) ranges from approximately 0 to 2.3, when ΔT = 10 frames, =0.005 corresponds to an entropy decrease of 0.05 over time ΔT, which remains constant throughout.

[0179] Slope sequence traversal and starting point locking: Starting from the first segment, traverse the topological entropy fixed segment slope sequence in sequence. :

[0180] Find the first satisfied Segmentation ;

[0181] The start time of this segment It is directly marked as the coagulation initiation point T1.

[0182] This application uses the starting time of the first fixed descent threshold slope segment to directly determine the coagulation initiation point, replacing the traditional multi-frame continuous descent verification or adaptive threshold search method.

[0183] Traditional multi-frame continuous verification delays the start point determination time (requiring confirmation from multiple frames), while adaptive threshold search can lead to inconsistent start point determination benchmarks due to sample differences, and may even result in misjudgments.

[0184] This application directly takes the first segment start time that meets the threshold, with no delay in judgment, and can accurately capture the earliest time point of coagulation initiation; the output of this step is the coagulation initiation point: T1 (corresponding to the segment start time). ).

[0185] The coagulation completion point and fibrinolysis initiation point are generated based on the first coagulation initiation point and the fixed segmented slope sequence of topological entropy; the coagulation initiation point, coagulation completion point and fibrinolysis initiation point constitute the key time nodes of coagulation.

[0186] In this embodiment, generating the coagulation completion point and fibrinolysis initiation point based on the first coagulation initiation point and the topological entropy fixed segmented slope sequence includes:

[0187] Fixed plateau threshold setting: Sets a fixed plateau period plateau threshold. (A unit of entropy change rate, which is fixed throughout the process) is used to determine the entropy stabilization stage after the clot has fully formed.

[0188] Clotting completion point determination: From Corresponding segments Then continue traversing the slope sequence. :

[0189] Find the first two consecutive segments that satisfy The segmented intervals, i.e., exist. , so that:

[0190] ;

[0191] The start time of the first segment of the interval It is directly marked as the coagulation completion point T2.

[0192] Fibrinolysis initiation point determination: from the segment corresponding to T2 Then continue traversing the slope sequence. :

[0193] Find the first satisfied Segmentation ;

[0194] The start time of this segment Marked as fibrinolysis initiation point T3;

[0195] If no segment satisfying the condition is found after traversing to the end of the slope sequence, it is determined that there is no fibrinolysis initiation point and T3 is not marked.

[0196] This application uses two consecutive segmented fixed stable thresholds to determine the coagulation completion point and the first fixed rebound threshold slope to segmentally determine the fibrinolysis initiation point.

[0197] Traditional single-frame stability determination is easily affected by small fluctuations in entropy, which may lead to misjudgment of the coagulation completion time; fibrinolysis initiation point determination often requires the combination of additional indicators (such as the reverse change of clot area), and the rules are complex and easily interfered with.

[0198] This application uses two consecutive segmented stable thresholds. To ensure that entropy truly enters a stable plateau period, avoid misjudgment of single-frame fluctuations, and accurately reflect the state of complete clot formation;

[0199] The initiation of fibrinolysis can be determined directly by the reverse rise slope of the topological entropy, without the need for additional calculation of indicators such as clot area. The rule is simple and closely follows the phase transition nature of coagulation-fibrinolysis.

[0200] Stable threshold , rebound threshold The entire process is fixed, and the judgment criteria for all samples are completely consistent. The output of this step is the coagulation completion point: T2 (corresponding to the start time of the segment). ); fibrinolysis initiation point: T3 (corresponding to the start time of the segment) (if it exists).

[0201] In this embodiment, the quantitative determination of coagulation function at key coagulation time points to obtain quantitative indicators of coagulation function includes:

[0202] For each spatial grid, the coagulation rate is calculated based on the initial features at the start time to obtain the final coagulation rate index.

[0203] In this embodiment, the coagulation rate is calculated for each spatial grid based on the initial feature hierarchy at the start time, thereby obtaining the final coagulation rate index, including:

[0204] Initial feature extraction at the initiation time: For each spatial grid i (i=1,2,...,256), extract its features at the coagulation initiation point. The irreversible nature of time .

[0205] Fixed three-layer division: according to The size is fixed, and the 256 grids are divided into three layers throughout:

[0206] Low-speed layer: (Areas with a low initial clot percentage);

[0207] Medium-speed layer: (Areas with a moderate initial clot percentage);

[0208] High-speed layer: (Areas with a high proportion of initial clots).

[0209] Intra-layer grid rate calculation: For all grids within each layer, calculate the coagulation rate of each individual grid.

[0210] ;

[0211] in The frame difference from coagulation initiation to completion (unit: feature value / frame; if it needs to be converted to seconds, it can be multiplied by the fixed sampling interval Δt; this scheme directly uses the frame difference for calculation).

[0212] Double median for final rate: Calculate the median of v(i) in the low-speed, medium-speed, and high-speed layers respectively, and denot it as... The median of these three medians is taken as the final coagulation rate index V.

[0213] This application adopts a dual median strategy based on the fixed three-layer partitioning and the median of each layer, which is based on the initial irreversible characteristics at the start time, to replace the traditional full-map average rate or single-grid rate weighting method.

[0214] Traditional full-map average rate is easily affected by individual abnormal grids (such as local contamination causing a certain grid to have an excessively fast / slow rate), resulting in insufficient robustness.

[0215] The stratification in this application is based on the initial characteristics at the start time, which can cover regions with different initial clot states and avoid the bias of a single average. The double median (first taking the median within the stratum, then taking the median between stratum) further improves robustness and is completely unaffected by individual abnormal grids. The stratification threshold and median calculation rules are fixed throughout the process, without any adaptive partitioning or weight allocation, and the results are completely reproducible. This step obtains the final coagulation rate index.

[0216] For each spatial grid, the agglomeration uniformity is calculated by a fixed combination of basic spatial uniformity and topological distribution entropy, thereby obtaining the final agglomeration uniformity index.

[0217] In this embodiment, the agglomeration uniformity is calculated for each spatial grid using a fixed combination of basic spatial uniformity and topological distribution entropy, thereby obtaining the final agglomeration uniformity index, including:

[0218] Completion Time Feature Extraction: For each spatial grid i, extract its irreversible features at the coagulation completion point T2. .

[0219] Basic spatial uniformity calculation:

[0220] Calculate all 256 grids Mean:

[0221] ;

[0222] Calculate the standard deviation:

[0223] ;

[0224] Basic spatial uniformity:

[0225] (like =0, an extreme and abnormal situation that will not actually occur under the fixed rules of this solution. =0).

[0226] Topological distribution entropy calculation:

[0227] For time T2 Perform maximum value normalization: , ;

[0228] Divide the [0,1] region into L=10 fixed-width bins and count the frequency of each bin. ;

[0229] Calculate the topological distribution entropy:

[0230] ,in The entire process is fixed.

[0231] Fixed arithmetic mean combination: converts distribution entropy into uniformity components ( ln(10)≈2.3 is The theoretical maximum value (the closer to 1, the more concentrated the distribution), and the final agglomerate uniformity:

[0232] ;

[0233] The value of U ranges from [0,1], and the closer it is to 1, the more uniform the agglomerates are.

[0234] This application proposes a fixed arithmetic mean combination index of basic spatial dispersion uniformity and global topological distribution entropy uniformity to replace the single spatial standard deviation uniformity.

[0235] Traditional single spatial uniformity only considers the characteristic differences between local grids and ignores the concentration of global features, which cannot fully characterize the quality of agglomerates (such as the possibility of small local differences but scattered global distribution).

[0236] This application Reflects the characteristic dispersion between local grids. The degree of concentration of global characteristics is reflected, and the combination of the two is more comprehensive; the output of this step is the final clot uniformity index: U (value range [0,1]).

[0237] The fibrinolytic activity is calculated and determined by fixed logic based on the combination of reverse irreversible characteristics and topological entropy at the fibrinolysis initiation point, thereby obtaining the final fibrinolytic activity index.

[0238] In this embodiment, fibrinolytic activity is calculated and fixed logic is applied based on a combination of reverse irreversible characteristics and topological entropy, according to the fibrinolysis initiation point, to obtain the final fibrinolytic activity index, including:

[0239] Calculation of fibrinolytic activity (two cases):

[0240] Case A: Fibrinolysis initiation point T exists. 3：

[0241] Topological entropy recovery rate calculation: The test ends at time... (i.e., total number of frames N), calculate the topology entropy recovery rate:

[0242] ;

[0243] in This represents the entropy change from the completion of blood clotting to the end of the test. This is the frame difference between the start and end of fibrinolysis.

[0244] Accumulation of reverse irreversible features: Defining reverse irreversible features ,initialization ; for t from T3+1 to Accumulate according to fixed rules:

[0245] ;

[0246] That is, only the reverse decrease of D(i,t) is accumulated, strictly guaranteeing Monotonically non-decreasing; extraction Moment Calculate the mean of all grid cells:

[0247] ;

[0248] Fixed fibrinolytic activity: The final fibrinolytic activity index is a fixed arithmetic mean. ;

[0249] Case B: Since there is no fibrinolysis initiation point T3, simply set F=0.

[0250] Fixed reference interval and final judgment

[0251] Based on clinically pre-defined, fixed normal reference intervals throughout the entire process (in this protocol, the reference interval is a fixed value: V normal interval). The normal interval U is [Ulow_norm, 1], and the normal interval F is [0, Fhigh_norm], which is determined using pure fixed AND / OR gate logic.

[0252] If V∈ And U∈ and The patient's coagulation function was determined to be normal.

[0253] like If it is determined to be delayed clotting; It was diagnosed as hypercoagulability;

[0254] like This was determined to be due to poor clot formation.

[0255] like It was determined to be hyperfibrinolytic activity;

[0256] If multiple abnormalities exist simultaneously, all abnormalities must be listed at the same time (e.g., slow coagulation and poor clot formation).

[0257] This application designs a reverse irreversible feature accumulation index, which, combined with the topological entropy recovery rate, obtains a fibrinolytic activity index through a fixed arithmetic mean combination.

[0258] Traditional fibrinolysis determination often requires the combination of additional indicators (such as the reverse change of clot area), and the rules are complex; the single topological entropy recovery rate is easily affected by local fluctuations and lacks comprehensiveness.

[0259] This application utilizes the reverse irreversible feature. By characterizing the clot dissolution process from a spatiotemporal perspective and combining it with the topological entropy recovery rate (a global feature), the two complement each other, resulting in a more comprehensive fibrinolysis index.

[0260] This application has the following advantages:

[0261] This invention breaks through the inherent approach of existing image-based coagulation test technologies that only extract single-frame spatial features and treat temporal information as an additional supplement. It takes the core physiological and physical essence of blood coagulation—the unidirectional irreversible phase transition from liquid plasma to solid fibrin clots, without spontaneous reverse phase transition—as a core constraint throughout the entire algorithm process.

[0262] In the feature encoding stage, a strictly monotonically non-decreasing irreversible feature is constructed through positive differential accumulation and unidirectional jump locking mechanism, which naturally filters out all reverse fluctuation interference and perfectly fits the unidirectional growth law of coagulation.

[0263] In the inflection point identification stage, based on the monotonic change law of topological entropy with irreversible characteristics, the coagulation phase transition node is directly identified, which perfectly matches the physiological processes of coagulation initiation, clot completion, and fibrinolysis initiation.

[0264] In the quantitative judgment stage, the fibrinolysis process is characterized by reverse irreversible features, achieving full coverage of the physiological process of the entire coagulation-fibrinolysis cycle.

[0265] Compared to existing technologies that use spatial features and weighted cumulative features that are disconnected from the physiological essence of coagulation, the feature system of this invention is completely anchored to the real coagulation process, fundamentally avoiding the occurrence of false features and false inflection points, and the detection results have extremely strong physiological relevance and clinical orientation.

[0266] This invention significantly improves the algorithm's anti-interference ability and result stability through a fixed constraint design across the entire spatiotemporal dimensions:

[0267] In the preprocessing stage, background anchoring is achieved through temporal cumulative grayscale histogram, which significantly enhances the stability features of the plasma background throughout the entire time series and naturally dilutes the interference of transient noise, Brownian motion of blood cells, and slight sample shaking. Compared with single-frame adaptive thresholding, the stability of background anchoring is improved by an order of magnitude. Through spatiotemporal neighborhood consistency binarization and three-dimensional spatiotemporal structuring element denoising, while filtering spatially isolated noise and temporal artifacts, the accuracy of the binarization result is much higher than that of traditional single-frame binarization.

[0268] In the feature encoding stage, only the irreversible features of the positive difference are accumulated, which naturally filters out the instantaneous reverse fluctuations caused by Brownian motion and sampling jitter. The feature sequence is free of spikes and fluctuations, and has extremely strong stability.

[0269] In the inflection point identification process, the slope calculation is linearized in segments at fixed time intervals, which naturally filters out single-frame feature fluctuations and avoids the pseudo-peak problem caused by continuous frame difference.

[0270] Meanwhile, this invention has extremely strong tolerance to common clinical interference samples such as jaundice, lipemia, and hemolysis: the background interference of such samples is a characteristic that exists stably throughout the entire time series. In the time series cumulative grayscale histogram, only the background main peak will be shifted, without changing the position lock of the main peak. Compared with optical turbidimetry and traditional single-frame image method, the ability to resist background interference is greatly improved, and no special preprocessing is required for abnormal samples.

[0271] This invention achieves inflection point identification through purely fixed thresholds and purely deterministic rules, representing a revolutionary improvement over existing technologies.

[0272] Accurate and Delay-Free Identification: The first segment starting time that meets the fixed descent threshold is directly used as the coagulation initiation point. No multi-frame continuous verification is required, and there is no judgment delay. It can accurately capture the earliest time point of coagulation initiation, and its ability to identify early coagulation function abnormalities far exceeds that of existing technologies. By locking the coagulation completion point through the fixed stable threshold of two consecutive segments, it accurately matches the physiological state of complete clot formation, without premature or delayed misjudgment.

[0273] The judgment criteria are completely consistent: all thresholds and rules are fixed throughout the process, without adaptive adjustment, curve fitting, or machine learning models. Regardless of the sample type, equipment differences, or changes in the detection environment, the judgment criteria for all samples are completely consistent, which completely solves the industry pain point of incomparable test results between different equipment, different laboratories, and different samples in existing technologies. The repeatability and reproducibility of the test results are extremely strong.

[0274] Full-cycle coverage: Simultaneously and accurately identifies three core time nodes: coagulation initiation point, coagulation completion point, and fibrinolysis initiation point. Compared with traditional coagulation tests that can only obtain a single coagulation endpoint, it can provide dynamic information on the entire coagulation cycle, providing a more comprehensive basis for the early diagnosis of clinical hemorrhagic and thrombotic diseases.

[0275] This invention constructs a quantitative indicator system covering three core dimensions: "coagulation rate, clot quality, and fibrinolytic activity." The entire process involves no artificial weighting or model fitting, and the robustness and comprehensiveness of the indicators far surpass existing technologies.

[0276] The coagulation rate index adopts a fixed three-layer division + double median strategy, which is completely unaffected by individual abnormal grids or local contamination areas. Its robustness is much higher than that of the traditional whole-map average rate, and it can truly reflect the overall coagulation speed of the sample.

[0277] The clot uniformity index combines the basic spatial dispersion and the global topological distribution entropy to simultaneously characterize the feature differences between local grids and the concentration of global features, comprehensively reflecting the quality of clot formation. It can detect hidden coagulation abnormalities that are normal in clotting time but poor in clot quality, which cannot be identified by traditional indicators.

[0278] The fibrinolytic activity index combines topological entropy recovery rate and reverse irreversible feature accumulation, simultaneously characterizing the fibrinolytic process from both global and local dimensions. No additional detection indicators are required; fibrinolytic activity can be accurately quantified solely through image sequences, filling the gap in existing image-based coagulation tests that cannot accurately quantify fibrinolytic function.

[0279] This application also provides a dynamic image recognition and testing device for blood coagulation process that integrates temporal features, the dynamic image recognition and testing device for blood coagulation process that integrates temporal features includes:

[0280] A continuous microscopic dynamic grayscale image sequence acquisition module for the blood coagulation process, wherein the continuous microscopic dynamic grayscale image sequence acquisition module for the blood coagulation process is used to acquire a continuous microscopic dynamic grayscale image sequence for the blood coagulation process;

[0281] A denoised spatiotemporally constrained binarized image sequence acquisition module is used to perform full-time spatiotemporally anchored image preprocessing on a continuous microscopic dynamic grayscale image sequence of the blood coagulation process, thereby obtaining a denoised spatiotemporally constrained binarized image sequence.

[0282] A temporally irreversible local connectivity feature encoding module is used to encode temporally irreversible local connectivity features on a denoised spatiotemporally constrained binarized image sequence, thereby obtaining a single-grid irreversible connectivity feature sequence and a globally sorted temporal topological feature sequence.

[0283] The coagulation key time node acquisition module is used to identify the fixed rule of coagulation phase transition inflection point in the global sorted temporal topological feature sequence, thereby acquiring the coagulation key time node.

[0284] The coagulation function quantitative index acquisition module is used to quantitatively determine the coagulation function at key time points of coagulation, thereby acquiring the coagulation function quantitative index.

[0285] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for dynamic image recognition and verification of blood coagulation process integrating temporal features, characterized in that, The method for dynamic image recognition and verification of the blood coagulation process, which integrates temporal features, includes: Acquire a continuous microscopic dynamic grayscale image sequence of the blood coagulation process; Full-time spatiotemporal anchoring image preprocessing is performed on the continuous microscopic dynamic grayscale image sequence of the blood coagulation process to obtain a denoised spatiotemporally constrained binarized image sequence; Temporally irreversible local connectivity feature encoding is performed on the denoised spatiotemporally constrained binarized image sequence to obtain single-grid irreversible connectivity feature sequence and globally sorted temporal topological feature sequence; Fixed rule identification of coagulation phase transition inflection points is performed on the global sorted temporal topological feature sequence to obtain key coagulation time nodes; The coagulation function is quantitatively assessed at key time points to obtain quantitative indicators of coagulation function.

2. The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features as described in claim 1, characterized in that, The image preprocessing of the continuous microscopic dynamic grayscale image sequence of the blood coagulation process with full-time spatiotemporal anchoring to obtain a denoised spatiotemporally constrained binarized image sequence includes: Background anchoring based on temporal gray-level accumulation peaks is performed on a continuous microscopic dynamic gray-level image sequence of the blood coagulation process to obtain a global fixed background threshold; Binarization of a continuous microscopic dynamic grayscale image sequence of the blood coagulation process is performed based on a globally fixed background threshold and neighborhood temporal consistency constraint to obtain a binarized image sequence. Denoising of the binarized image sequence using spatiotemporal three-dimensional fixed structuring elements is performed to obtain a denoised spatiotemporally constrained binarized image sequence.

3. The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features as described in claim 2, characterized in that, The step of encoding temporally irreversible local connectivity features in the denoised spatiotemporally constrained binarized image sequence to obtain a single-grid irreversible connectivity feature sequence and a globally ordered temporal topological feature sequence includes: Each frame of the denoised spatiotemporally constrained binarized image sequence is divided into spatiotemporally bound grids to obtain a full-time spatiotemporal grid cell set. The temporal differential cumulative duty cycle is calculated for the full-time spatiotemporal grid cell set to obtain the temporal differential cumulative duty cycle sequence for each spatiotemporal grid cell; Based on the temporal differential cumulative duty cycle sequence of each spatiotemporal grid cell, the final irreversible features are constructed through a one-way jump locking mechanism, thereby obtaining the single-grid irreversible connectivity feature sequence and the globally sorted temporal topology feature sequence.

4. The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features as described in claim 3, characterized in that, The fixed rule identification of coagulation phase transition inflection points in the globally sorted temporal topological feature sequence, thereby obtaining key coagulation time nodes, includes: The topological entropy slope is linearized by fixed segmentation on the single-grid irreversible connected feature sequence and the globally sorted temporal topological feature sequence to obtain the full-time normalized topological entropy curve and the topological entropy fixed segmentation slope sequence. The coagulation initiation point is obtained based on a fixed segmented slope sequence of topological entropy. The coagulation completion point and fibrinolysis initiation point are generated based on the first coagulation initiation point and the fixed segmented slope sequence of topological entropy; the coagulation initiation point, coagulation completion point and fibrinolysis initiation point constitute the key time nodes of coagulation.

5. The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features as described in claim 4, characterized in that, The quantitative determination of coagulation function at key coagulation time points, thereby obtaining quantitative indicators of coagulation function, includes: For each spatial grid, the coagulation rate is calculated based on the initial features at the start time to obtain the final coagulation rate index. For each spatial grid, the agglomeration uniformity is calculated by a fixed combination of basic spatial uniformity and topological distribution entropy, thereby obtaining the final agglomeration uniformity index. The fibrinolytic activity is calculated and determined by fixed logic based on the combination of reverse irreversible characteristics and topological entropy at the fibrinolysis initiation point, thereby obtaining the final fibrinolytic activity index.

6. The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features as described in claim 2, characterized in that, The step of performing background anchoring based on temporal gray-level accumulation peaks on a continuous microscopic dynamic gray-level image sequence of the blood coagulation process to obtain a globally fixed background threshold includes: For the input sequence of continuous microscopic dynamic grayscale images of blood coagulation, the number of pixels corresponding to each grayscale level g is counted frame by frame to obtain the grayscale histogram of a single frame. ; To construct a time-series cumulative gray-level histogram, sum the histograms of all single frames bit by bit. ; Traversal Find all gray levels g that make The grayscale value that yields the maximum value is the unique global fixed background threshold. .

7. The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features as described in claim 6, characterized in that, The binarization of the continuous microscopic dynamic grayscale image sequence of the blood coagulation process according to a globally fixed background threshold with neighborhood temporal consistency constraints, thereby obtaining a binarized image sequence, includes: The following operations were performed on each frame of the continuous microscopic dynamic grayscale image sequence of the blood coagulation process: Check the grayscale value I(x,y,t) of the current pixel (x,y) in frame t. If I(x,y,t) ≥ If I(x,y,t) < Proceed to the next step of temporal neighborhood determination; Spatial neighborhood extraction between previous and next frames: Extract the 3×3 spatial neighborhood grayscale set centered at (x,y) in the previous frame. ; Extract the 3×3 spatial neighborhood grayscale set N(x,y,t+1) centered at (x,y) in the next frame; Neighborhood consistency determination and binarization: statistics Medium grayscale value less than Number of pixels ; Statistically count the gray values ​​less than in N(x,y,t+1) Number of pixels ; If both conditions are met and If the value is true, then mark the current pixel (x,y,t) as a foreground pixel; otherwise, mark it as a background pixel. For boundary frames, directly let Finally, a binarized image sequence is obtained. .

8. The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features as described in claim 7, characterized in that, The step of performing spatiotemporal three-dimensional fixed structuring element denoising on the binarized image sequence to obtain a denoised spatiotemporally constrained binarized image sequence includes: Construct a 3×3×3 spatiotemporal three-dimensional fixed structural element; Spatiotemporal morphological opening is performed on the binarized image sequence B(t) to obtain the denoised spatiotemporally constrained binarized image sequence.

9. The method for dynamic image recognition and verification of blood coagulation process based on the fusion of temporal features as described in claim 8, characterized in that, The step of performing a spatiotemporal morphological opening operation on the binarized image sequence B(t) to obtain a denoised spatiotemporally constrained binarized image sequence includes: For each pixel (x,y,t) in the binary sequence B(t), extract all 27 pixel values ​​within a 3×3×3 spatiotemporal neighborhood centered on that pixel; If all 27 pixel values ​​are 1, then the eroded pixel values ​​will be... Assign a value of 1; otherwise, assign a value of 0. For each etched pixel value The etched sequence For each pixel (x,y,t) in the dataset, extract all 27 pixel values ​​within a 3×3×3 spatiotemporal neighborhood centered on that pixel. If at least one of these 27 pixel values ​​is 1, then the dilated pixel value will be... The value is assigned to 1; otherwise, it is assigned to 0. The dilated pixel values ​​form the denoised binary image sequence.

10. A dynamic image recognition and testing device for blood coagulation process integrating temporal features, characterized in that, The blood coagulation process dynamic image recognition and testing device that integrates temporal features includes: A continuous microscopic dynamic grayscale image sequence acquisition module for the blood coagulation process, wherein the continuous microscopic dynamic grayscale image sequence acquisition module for the blood coagulation process is used to acquire a continuous microscopic dynamic grayscale image sequence for the blood coagulation process; A denoised spatiotemporally constrained binarized image sequence acquisition module is used to perform full-time spatiotemporally anchored image preprocessing on a continuous microscopic dynamic grayscale image sequence of the blood coagulation process, thereby obtaining a denoised spatiotemporally constrained binarized image sequence. A temporally irreversible local connectivity feature encoding module is used to encode temporally irreversible local connectivity features on a denoised spatiotemporally constrained binarized image sequence, thereby obtaining a single-grid irreversible connectivity feature sequence and a globally sorted temporal topological feature sequence. The coagulation key time node acquisition module is used to identify the fixed rule of coagulation phase transition inflection point in the global sorted temporal topological feature sequence, thereby acquiring the coagulation key time node. The coagulation function quantitative index acquisition module is used to quantitatively determine the coagulation function at key time points of coagulation, thereby acquiring the coagulation function quantitative index.

Images