A Stem Cell Classification Method and System Based on Image Segmentation

Abstract

This invention relates to the field of image recognition technology, and more particularly to a stem cell classification method and system based on image segmentation. The method includes the following steps: acquiring raw stem cell image data; performing Wasm processing on the raw stem cell image data to obtain stem cell image quality parameters; performing adaptive preprocessing on the raw stem cell image data based on the stem cell image quality parameters to obtain standardized stem cell image data; and constructing a multi-resolution pyramid on the standardized stem cell image data to obtain multi-level stem cell image representation data. This invention effectively improves the accuracy, reliability, adaptability, and scalability of stem cell image classification through multi-resolution pyramid construction, cell region hierarchical processing, precise morphological feature extraction and hierarchical classification, and distributed storage processing.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a stem cell classification method and system based on image segmentation. Background Technology

[0002] Various interference factors during the microscopic imaging process can affect the accurate extraction and identification of cellular regions. Although existing image processing methods can improve image quality to some extent, these methods often lack targeted adaptability and cannot automatically adjust parameters according to the specific circumstances of different samples, resulting in unsatisfactory preprocessing results. In practical applications, stem cell images often exhibit cell overlap and aggregation, making image processing even more complex. Traditional image segmentation methods typically employ simple methods based on thresholding or edge detection when dealing with overlapping cell areas. While these methods can perform preliminary segmentation, the similarity in morphology, color, and texture information between cells makes it impossible to effectively distinguish overlapping cells. The morphological characteristics of stem cells are crucial for their classification; however, traditional morphological feature extraction methods often rely on coarse edge detection and morphological filtering techniques, resulting in insufficient accuracy in identifying cell structure and morphology. Summary of the Invention

[0003] Therefore, the present invention needs to provide a stem cell classification method and system based on image segmentation to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a stem cell classification method based on image segmentation includes the following steps:

[0005] Step S1: Obtain raw stem cell image data, perform Wasm processing on the raw stem cell image data to obtain stem cell image quality parameters; perform adaptive preprocessing on the raw stem cell image data based on the stem cell image quality parameters to obtain standardized stem cell image data.

[0006] Step S2: Construct a multi-resolution pyramid from the standardized stem cell image data to obtain multi-level stem cell image representation data;

[0007] Step S3: Perform multi-resolution layering processing on the multi-level stem cell image representation data to obtain multi-level cell feature data, which includes potential cell region data, cell aggregation region data and cell overlap region data.

[0008] Step S4: Perform stem cell adaptive algorithm segmentation based on multi-level cell feature data to obtain segmented stem cell image data;

[0009] Step S5: Extract morphological features from the segmented stem cell image data, and perform hierarchical classification based on the morphological feature extraction results to obtain stem cell classification result data;

[0010] Step S6: Perform distributed storage processing on the stem cell classification results data to generate a persistent stem cell classification database.

[0011] This invention effectively improves the accuracy and reliability of stem cell image classification through its various processing steps. First, Wasm processing is used to obtain stem cell image quality parameters, enabling the evaluation of the original image quality. Adaptive preprocessing is then performed based on these parameters, overcoming the limitation of traditional methods that cannot automatically adjust to different sample conditions during image preprocessing. This ensures the standardization of image data and improves the accuracy of subsequent processing. Multi-resolution pyramid construction obtains stem cell image representation data at different levels, allowing for better capture of multi-scale features of cell images during processing and improving the ability to identify cell regions and their complex structures. Further, multi-resolution hierarchical processing divides cell region data into potential cell regions, cell aggregation regions, and cell overlap regions, enabling more detailed region division and avoiding the inaccurate handling of cell overlap and aggregation phenomena in traditional methods, thus improving cell segmentation results. The use of an adaptive stem cell segmentation algorithm automatically adjusts the segmentation strategy according to the characteristics of different samples, overcoming the limitations of traditional image segmentation methods in insufficient handling of cell overlap and morphological similarity, ensuring the accuracy of the segmentation results. Morphological feature extraction combined with hierarchical classification enables cell classification to be more accurate based on cell morphological features, avoiding the influence of overly coarse edge detection and morphological filtering in traditional methods, thus improving the reliability of classification results. Finally, a persistent stem cell classification database is generated through distributed storage processing, ensuring efficient data storage, management, and retrieval, and supporting the processing needs of large-scale datasets, further improving the adaptability and scalability of this method in practical applications.

[0012] Preferably, the present invention also provides an image segmentation-based stem cell classification system for performing the above-described image segmentation-based stem cell classification method, wherein the image segmentation-based stem cell classification system includes:

[0013] The image preprocessing and standardization module is used to acquire raw stem cell image data, perform Wasm processing on the raw stem cell image data to obtain stem cell image quality parameters, and perform adaptive preprocessing on the raw stem cell image data based on the stem cell image quality parameters to obtain standardized stem cell image data.

[0014] The multi-resolution pyramid construction module is used to construct multi-resolution pyramids from standardized stem cell image data to obtain multi-level stem cell image representation data.

[0015] The multi-resolution hierarchical feature extraction module is used to perform multi-resolution hierarchical processing on multi-level stem cell image representation data to obtain multi-level cell feature data, which includes potential cell region data, cell aggregation region data and cell overlap region data.

[0016] The stem cell adaptive segmentation module is used to perform stem cell adaptive algorithm segmentation based on multi-level cell feature data to obtain segmented stem cell image data.

[0017] The morphological feature extraction and classification module is used to extract morphological features from segmented stem cell image data and perform hierarchical classification based on the morphological feature extraction results to obtain stem cell classification result data.

[0018] The distributed storage and persistence module is used to process stem cell classification results data in a distributed manner and generate a persistent stem cell classification database.

[0019] This invention improves the quality and standardizes raw stem cell image data through adaptive preprocessing and Wasm processing, ensuring the accuracy of subsequent analysis. Multi-resolution pyramid construction generates multi-level stem cell image data, providing a multi-scale perspective for cell feature extraction and segmentation, enhancing processing capabilities. Multi-level cell feature extraction accurately identifies potential cell regions, cell aggregation areas, and cell overlap areas, providing rich feature data for subsequent stem cell segmentation and classification. Adaptive algorithms are used for stem cell image segmentation, precisely separating cell regions and providing clear stem cell image data for classification. Morphological feature extraction and hierarchical classification optimize cell classification results, achieving efficient and accurate stem cell classification. Distributed storage and persistent processing ensure efficient storage and management of stem cell classification results data, facilitating subsequent querying and analysis. Attached Figure Description

[0020] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0021] Figure 1 This is a schematic diagram of the steps in the image segmentation-based stem cell classification method of the present invention;

[0022] Figure 2 for Figure 1 A detailed flowchart of step S1;

[0023] Figure 3 for Figure 1A detailed flowchart of step S2. Detailed Implementation

[0024] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0026] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0027] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a stem cell classification method based on image segmentation, the method comprising the following steps:

[0028] Step S1: Obtain raw stem cell image data, perform Wasm processing on the raw stem cell image data to obtain stem cell image quality parameters; perform adaptive preprocessing on the raw stem cell image data based on the stem cell image quality parameters to obtain standardized stem cell image data.

[0029] Step S2: Construct a multi-resolution pyramid from the standardized stem cell image data to obtain multi-level stem cell image representation data;

[0030] Step S3: Perform multi-resolution layering processing on the multi-level stem cell image representation data to obtain multi-level cell feature data, which includes potential cell region data, cell aggregation region data and cell overlap region data.

[0031] Step S4: Perform stem cell adaptive algorithm segmentation based on multi-level cell feature data to obtain segmented stem cell image data;

[0032] Step S5: Extract morphological features from the segmented stem cell image data, and perform hierarchical classification based on the morphological feature extraction results to obtain stem cell classification result data;

[0033] Step S6: Perform distributed storage processing on the stem cell classification results data to generate a persistent stem cell classification database.

[0034] In this embodiment of the invention, reference is made to Figure 1 The diagram shown illustrates the steps of a stem cell classification method based on image segmentation according to the present invention. In this example, the stem cell classification method based on image segmentation includes the following steps:

[0035] Step S1: Obtain raw stem cell image data, perform Wasm processing on the raw stem cell image data to obtain stem cell image quality parameters; perform adaptive preprocessing on the raw stem cell image data based on the stem cell image quality parameters to obtain standardized stem cell image data.

[0036] This invention employs a confocal microscope with a 10X objective to acquire raw stem cell image data in a stem cell culture dish, with an image resolution of 1024×1024 pixels. The acquired raw stem cell image data is processed using an image analysis module compiled with WebAssembly. Specifically, the image data is separated into RGB channels, and the mean brightness, standard deviation, signal-to-noise ratio, contrast, and sharpness of each channel are calculated to form a comprehensive stem cell image quality parameter. Subsequently, based on the stem cell image quality parameter, when the mean image brightness is below 30 or above 220, a histogram equalization algorithm is applied; when the standard deviation is greater than 50, an additional algorithm is applied. Gaussian filtering is used for noise reduction, with a filter kernel size of 5×5. When the signal-to-noise ratio is below 15dB, medium-range filtering is applied to remove salt-and-pepper noise, with a filter window size of 3×3. When the sharpness index is below 0.65, a non-local mean denoising algorithm is applied, with a search window radius of 7. Through the above series of processing operations, image color space standardization, illumination non-uniformity correction, and noise elimination are achieved, ultimately outputting standardized stem cell image data. The data size remains unchanged at 1024×1024 pixels, but the dynamic range is adjusted to between 0 and 255, the RGB channel mean is set to 128±5, and the standard deviation is controlled within the range of 40-60.

[0037] Step S2: Construct a multi-resolution pyramid from the standardized stem cell image data to obtain multi-level stem cell image representation data;

[0038] In this embodiment of the invention, standardized stem cell image data is used as the initial layer L0 with a resolution of 1024×1024 pixels. A Gaussian kernel function is used to downsample L0, with a kernel size of 5×5, a standard deviation σ = 1.6, and a downsampling factor of 2, generating a layer L1 with a resolution of 512×512 pixels. Then, the same Gaussian kernel parameters are applied to the L1 layer for further downsampling, generating a layer L2 with a resolution of 256×256 pixels. This process is repeated to gradually generate layers L3 (128×128 pixels) and L4 (64×64 pixels), thus constructing a 5-layer multilayer structure. Resolution pyramid; for each level Li (i = 0, 1, 2, 3, 4), calculate the Laplacian difference image Di = Li - upsampled(Li+1), where the upsampling operation uses a bicubic interpolation algorithm to enlarge the size of Li+1 to be the same as Li; for the bottom level L4, directly use it as D4; combine the difference images Di of each level with the original level Li to form multi-level stem cell image representation data {(L0,D0),(L1,D1),(L2,D2),(L3,D3),(L4,D4)}, each level containing the original resolution image and the corresponding Laplacian difference image.

[0039] Step S3: Perform multi-resolution layering processing on the multi-level stem cell image representation data to obtain multi-level cell feature data, which includes potential cell region data, cell aggregation region data and cell overlap region data.

[0040] In this embodiment of the invention, the multi-level stem cell image representation data {(L0,D0),(L1,D1),(L2,D2),(L3,D3),(L4,D4)} obtained in step S2 are subjected to feature extraction processing. First, local contrast enhancement is performed on the L0 level (1024×1024 pixels) using an adaptive histogram equalization algorithm. The image is divided into 16×16 tiles, each 64×64 pixels in size, with a clip limit of 3.0. Potential cell regions are identified by calculating the difference between the 90th and 10th percentiles of pixel intensity in each tile. Then, on the L2 level (256×256 pixels), the Otsu thresholding method is applied to determine the global threshold, which is set to the minimum value between the two peaks in the bimodal distribution of the gray-level histogram, to obtain cell aggregation region data. Next, a zero-crossing detection algorithm is used on the D1 difference image to identify cell boundaries. The detection used the Laplacian operator with a kernel size of 3×3, a smoothing factor α = 0.2, and a boundary extraction threshold of 10 to obtain cell boundary information. Morphological operations were then performed at the L3 level (128×128 pixels), including opening operations using circular structuring elements with a radius of 3 pixels to remove noise, and closing operations using circular structuring elements with a radius of 5 pixels to fill holes and identify cell overlap areas. Finally, the processing results from each level were mapped back to the original resolution L0 using back projection technology. The back projection employed a nearest neighbor interpolation algorithm to maintain boundary sharpness, and the boundary matching threshold was set to 0.85, forming a multi-level cell feature data set M = {P, A, O} containing latent cell region data, cell aggregation region data, and cell overlap area data. Here, P represents the binary mask of the latent cell region, A represents the probability map of the cell aggregation region, and O represents the label map of the cell overlap area. All feature data were stored at the same 1024×1024 resolution as L0.

[0041] Step S4: Perform stem cell adaptive algorithm segmentation based on multi-level cell feature data to obtain segmented stem cell image data;

[0042] This invention takes a multi-level cell feature dataset M = {P, A, O} as input. First, an initial watershed marker map is constructed. A distance transformation is performed on the potential cell region data P, using Euclidean distance and normalized to [0, 255]. Then, the H-maximum suppression algorithm is applied to the transformation result to extract local maxima as cell core region markers, with a suppression radius of 15 pixels. Next, a segmentation weight map W is constructed using the probability values ​​in the cell aggregation region data A. The weight calculation formula is W(x,y) = 255 - A(x,y) × (1 - O(x,y) × 0.5), ensuring that cell aggregation regions have lower weights and boundary regions have higher weights. Subsequently, the marker matrix and weight map W are input into a marker-controlled watershed algorithm. Neighborhood connectivity is set to 8-connectivity, the watershed gradient threshold is set to 25, and the iteration termination condition is... The condition is that the regional change rate is less than 0.01%. For the detected cell overlap data O, the concave point detection algorithm is used to identify the cell boundaries. The concave point detection angle threshold is set to 120 degrees and the edge mutation threshold is set to 0.65. Then, the conditional random field algorithm is applied to segment the overlapping cells. The energy function adopts the Potts model, the smoothing factor λ is set to 0.8, and the number of iterations is 200. Finally, the segmentation results are verified. Noise regions with an area less than 200 pixels or a perimeter / area ratio greater than 0.75 are removed. Regions with an area greater than 2000 pixels and a roundness less than 0.5 are segmented again to obtain the final segmented stem cell image data. This data contains the precise contour boundary and unique identifier of each independent cell, maintains the original 1024×1024 pixel resolution, and is stored in the form of a binary mask array, with each mask corresponding to an independent stem cell.

[0043] Step S5: Extract morphological features from the segmented stem cell image data, and perform hierarchical classification based on the morphological feature extraction results to obtain stem cell classification result data;

[0044] In this embodiment of the invention, the segmented stem cell image data obtained in step S4 is used as input. First, 28 morphological features are extracted for each segmented independent stem cell region, including area (calculated in pixels), perimeter (number of boundary pixels), equivalent diameter (diameter of a circle equal to the cell area), maximum inscribed circle radius, minimum circumscribed rectangle aspect ratio, minimum circumscribed ellipse eccentricity, circularity (4π×area / perimeter²), convex hull area, number of convex defects, total area of ​​convex defects, mean depth of convex defects, elongation (length-to-width ratio), density (area / convex hull area), mean curvature, standard deviation of curvature, seven values ​​of Hu invariant moments, five coefficients of Zernike moments, mean gray level, gray level variance, gray level entropy, texture energy, texture contrast, texture homogeneity, and texture correlation. Then, the extracted features are normalized using the Z-score standardization method, calculated as Z=(x-μ) / σ, where μ is the feature mean and σ is the standard deviation. Subsequently, a hierarchical clustering algorithm was used to classify the standardized features. Euclidean distance was used as the distance metric, Ward's minimum variance method was used as the link criterion, and the hierarchical tree truncation threshold was set to 0.35. Stem cells were divided into 5 main categories: pluripotent stem cells (roundness > 0.85, area < 1000 pixels), progenitor cells (0.65 < roundness < 0.85, 1000 < area < 1500 pixels), differentiating cells (0.4 < roundness < 0.65, elongation > 1.8), fully differentiated cells (roundness < 0.4, elongation > 2.5), and abnormal cells (convex defect area > 25% of total area or texture entropy > 6.0). Finally, the classification results were validated by calculating the ratio of intra-class distance to inter-class distance to ensure classification robustness. The classification robustness threshold was set to 0.75. Stem cell classification result data was generated, with each cell corresponding to a structure containing cell ID, location coordinates, morphological feature vector, and classification label.

[0045] Step S6: Perform distributed storage processing on the stem cell classification results data to generate a persistent stem cell classification database.

[0046] This invention uses stem cell classification results data as input. First, the data is structured and organized, creating a unique identifier for each stem cell classification result. This identifier consists of 16 hexadecimal characters and includes the collection timestamp, sample number, and cell sequence number. Next, the stem cell classification results data is serialized into binary objects using Protocol Buffers format with a compression level of 7 to ensure data integrity. Then, a three-layer distributed storage architecture is constructed. The first layer is an in-memory database using a Redis cluster, configured with 3 master nodes and 3 slave nodes, with 4GB of memory allocated to each node and a data expiration time of 72 hours. The second layer is a relational database using PostgreSQL. The cluster uses a partitioned table with dual partitioning by time and cell type. The primary key index is based on the cell's unique identifier, and the secondary index is based on morphological features and classification results. The third layer is object storage, using a distributed file system with a data block size of 128MB, a replication factor of 3, and a data sharding strategy based on cell type and time range. Finally, a data synchronization mechanism ensures data consistency between the three storage layers, employing a two-phase commit protocol, a transaction timeout of 30 seconds, 3 retries, and a synchronization interval of 5 minutes. This completes the persistent storage of the stem cell classification database, which contains complete stem cell images, segmentation results, morphological features, and classification labels, supporting multi-dimensional queries and retrieval by cell type, morphological feature range, and collection time.

[0047] This invention effectively improves the accuracy and reliability of stem cell image classification through its various processing steps. First, Wasm processing is used to obtain stem cell image quality parameters, enabling the evaluation of the original image quality. Adaptive preprocessing is then performed based on these parameters, overcoming the limitation of traditional methods that cannot automatically adjust to different sample conditions during image preprocessing. This ensures the standardization of image data and improves the accuracy of subsequent processing. Multi-resolution pyramid construction obtains stem cell image representation data at different levels, allowing for better capture of multi-scale features of cell images during processing and improving the ability to identify cell regions and their complex structures. Further, multi-resolution hierarchical processing divides cell region data into potential cell regions, cell aggregation regions, and cell overlap regions, enabling more detailed region division and avoiding the inaccurate handling of cell overlap and aggregation phenomena in traditional methods, thus improving cell segmentation results. The use of an adaptive stem cell segmentation algorithm automatically adjusts the segmentation strategy according to the characteristics of different samples, overcoming the limitations of traditional image segmentation methods in insufficient handling of cell overlap and morphological similarity, ensuring the accuracy of the segmentation results. Morphological feature extraction combined with hierarchical classification enables cell classification to be more accurate based on cell morphological features, avoiding the influence of overly coarse edge detection and morphological filtering in traditional methods, thus improving the reliability of classification results. Finally, a persistent stem cell classification database is generated through distributed storage processing, ensuring efficient data storage, management, and retrieval, and supporting the processing needs of large-scale datasets, further improving the adaptability and scalability of this method in practical applications.

[0048] Preferably, step S1 includes the following steps:

[0049] Step S11: Obtain raw stem cell image data, compile the raw stem cell image data using the Wasm module to obtain a high-performance image processing module;

[0050] Step S12: The quality of the original stem cell image data is evaluated based on the high-performance image processing module to obtain stem cell image quality parameters;

[0051] Step S13: Perform image quality analysis and preprocessing strategy determination on stem cell image quality parameters to obtain the optimal preprocessing strategy;

[0052] Step S14: Based on the optimal preprocessing strategy, perform image denoising and contrast enhancement on the original stem cell image data to obtain enhanced stem cell image data;

[0053] Step S15: Perform illumination correction and edge optimization on the enhanced stem cell image data to obtain standardized stem cell image data.

[0054] As an embodiment of the present invention, reference Figure 2 As shown, Figure 1 A detailed flowchart of step S1 is shown below. In this embodiment of the invention, step S1 includes the following steps:

[0055] Step S11: Obtain raw stem cell image data, compile the raw stem cell image data using the Wasm module to obtain a high-performance image processing module;

[0056] This invention employs a microscope system to acquire raw stem cell image data. The microscope objective is set to 20X magnification and 0.75 numerical aperture. The image acquisition device is a 12-megapixel CCD camera with a pixel size of 3.75μm × 3.75μm, an exposure time of 25 milliseconds, an acquisition resolution of 2048 × 2048 pixels, and a color depth of 24-bit RGB color. This raw stem cell image data is then compiled using a WebAssembly module. First, the C++ image processing algorithm source code, containing four core functions—Gaussian blur, median filtering, Laplacian edge detection, and threshold segmentation—is converted into LLVM intermediate code using the Emscripten compiler. The compilation command is "emcc-O3-s". WASM=1, with optimization level set to 3; next, the LLVM intermediate code is compiled into a WebAssembly binary module, generating a .wasm file of 256KB; then, JavaScript glue code is built to implement memory management and function calls, with a glue code size of 32KB; subsequently, the module is instantiated, with an initial memory size of 16MB, expandable to a maximum of 64MB, and exported function interfaces including four main functions: processImage, detectEdges, enhanceContrast, and correctIllumination; finally, module performance is optimized by applying the SIMD instruction set for acceleration, parallelizing data processing to 8 channels, and optimizing memory access mode to linear continuous access, thus completing the construction of the high-performance image processing module.

[0057] Step S12: The quality of the original stem cell image data is evaluated based on the high-performance image processing module to obtain stem cell image quality parameters;

[0058] This invention applies a high-performance image processing module to the raw stem cell image data. By calling the `processImage` function in the module, the 2048×2048 pixel raw image is loaded into WebAssembly linear memory, with the memory starting address set to 0x1000 and 16MB of contiguous space allocated. Next, multi-parameter quantization evaluation is performed. First, global brightness statistics parameters are calculated, including the average brightness values ​​μR, μG, and μB of the RGB three channels, calculated using the formula μc=∑I(x,y,c) / (W×H), where c∈{R,G,B}, W=H=2048; then... Contrast parameters were calculated using the Michelson contrast method, with the formula C = (Imax - Imin) / (Imax + Imin), yielding contrast values ​​for the CR, CG, and CB channels. Next, image noise levels were measured using the mean square error method, calculating the local variance σ²local for each 7×7 pixel block, and taking the mean of all local variances as the noise estimate σnoise. Then, image sharpness was evaluated using the Tenengrad algorithm, employing a 3×3 Sobel operator to calculate the horizontal and vertical gradients Gx and Gy, and then calculating the gradient magnitude G(x,y) = √(Gx / Gy). 2 +Gy2), the sharpness index S=∑G(x,y)2, with the value range normalized to [0,1]; then the color balance is measured, and the RGB three-channel ratio deviation δRG=|μR / μG-1|, δRB=|μR / μB-1|, δGB=|μG / μB-1| is calculated; finally, Fourier transform is used to analyze the frequency characteristics, the image is converted to the frequency domain, and the ratio of high frequency to low frequency components ρHF / LF is calculated. All calculation results are combined to form the stem cell image quality parameter set Q={μR,μG,μB,CR,CG,CB,σnoise,S,δRG,δRB,δGB,ρHF / LF}, and the parameter set is stored in the form of a floating-point array.

[0059] Step S13: Perform image quality analysis and preprocessing strategy determination on stem cell image quality parameters to obtain the optimal preprocessing strategy;

[0060] This invention first extracts four indicators—signal-to-noise ratio (SNR), contrast, luminance variance, and edge sharpness—from stem cell image quality parameters. A preprocessing strategy is determined using threshold rules. When the SNR is below 30 dB, a nonlocal means denoising algorithm is triggered, employing OpenCV. The fastNlMeansDenoisingColored function in version 4.5 is used with parameters h=7, templateWindowSize=7, and searchWindowSize=21. When the contrast ratio is below 50, contrast-limited adaptive histogram equalization is performed using the CLAHE algorithm with clipLimit=2.0 and tileGridSize=(8,8). When the luminance variance exceeds 100, multi-scale Retinex illumination correction is applied with a three-layer decomposition structure using Gaussian kernel scales σ=15, 30, and 60. When the edge sharpness is below 0.5, Sobel operator edge optimization is initiated with a fixed convolution kernel size of 3×3. The gradient calculation uses bidirectional difference between dx=1, dy=0 and dx=0, dy=1. Finally, the priority of each strategy is sorted through the parameter weight matrix [0.4,0.3,0.2,0.1] to generate an optimal preprocessing strategy configuration file containing algorithm parameters and execution order.

[0061] Step S14: Based on the optimal preprocessing strategy, perform image denoising and contrast enhancement on the original stem cell image data to obtain enhanced stem cell image data;

[0062] This invention applies an optimal preprocessing strategy to raw stem cell image data. First, adaptive noise reduction is performed based on the noise type. When σnoise < 10, Gaussian filtering is applied for smoothing, with a Gaussian kernel size of 5×5 and a standard deviation σ = 1.2. When 10 ≤ σnoise < 25, bilateral filtering is applied to preserve edge details, with a spatial domain standard deviation of 6.0 and a value domain standard deviation of 25.0. When σnoise ≥ 25, a nonlocal mean filtering algorithm is applied to remove high-intensity noise, with a search window size of 21×21, a comparison window size of 7×7, and a filter intensity parameter set to 1.5 times the noise standard deviation. Next, contrast enhancement is performed. When image contrast C < 0.3, global histogram equalization is applied, using a cumulative distribution function for the 256-level grayscale mapping. When 0.3 ≤ C < 0.5, contrast-limited adaptive histogram equalization (CLAHE) is applied, segmenting the image into... An 8×8 grid of 256×256 pixels is used, with a contrast threshold of 3.5 and bilinear interpolation coefficients α = 0.75. When C ≥ 0.5 and local areas with insufficient contrast exist, a multi-scale Retinex algorithm is applied, using three Gaussian filter kernels σ1 = 15, σ2 = 80, and σ3 = 250, with weights w1 = 0.3, w2 = 0.5, and w3 = 0.2, and a color restoration factor β = 1.2. After implementing the above adaptive noise reduction and contrast enhancement operations, the enhanceContrast function in the high-performance image processing module is called to complete the processing. The function takes the original stem cell image data and the preprocessing parameter set as input and outputs the enhanced stem cell image data. The processed image resolution remains 2048×2048 pixels, and the bit depth remains 24 bits, but the noise level is reduced to less than 30% of the original value, and the contrast is increased to 1.5-2.5 times that of the original image.

[0063] Step S15: Perform illumination correction and edge optimization on the enhanced stem cell image data to obtain standardized stem cell image data.

[0064] This invention uses enhanced stem cell image data as input. First, non-uniform illumination correction is performed. A polynomial-fitted background illumination model is used, dividing the image into 256 sub-regions of 16×16 grid size, each sub-region being 128×128 pixels. The average brightness value of each sub-region is calculated, and a brightness distribution matrix M is constructed. Cubic B-spline interpolation is applied to generate a continuous illumination model function B(x,y), and the correction coefficient K(x,y) = μtarget / B(x,y) is calculated, where μtarget is set to 128. Each pixel value I(x,y) of the original image is multiplied by the corresponding correction coefficient K(x,y) to obtain an illumination-balanced image. Next, edge optimization processing is performed. First, the Canny edge detection algorithm is applied, with a low threshold set to 10% of the image intensity and a high threshold set to 25% of the image intensity, with a Gaussian smoothing factor σ = 1.0. Cells are detected. The initial boundary position is determined; then, edge enhancement is performed using morphological operations. A linear structuring element with a length of 5 and a width of 1 is used for directional dilation, with an angular interval of 45 degrees. The union of the dilation values ​​in eight directions is then calculated. Subpixel edge localization technology is then applied, calculating the second moment of the local grayscale gradient at the edge points and fitting a quadratic curve to obtain the precise boundary position, achieving an edge localization accuracy of 0.1 pixels. Finally, color space standardization is performed, converting the image from RGB to Lab color space. The mean and standard deviation of the L channel are adjusted to 50, the mean and standard deviation of the a channel to 0, and the mean and standard deviation of the b channel to 0 and 5. The image is then converted back to RGB space, outputting standardized stem cell image data. This data remains 2048×2048 pixels with a color depth of 24 bits, but features uniform illumination, clear cell boundaries, and standardized color representation.

[0065] This invention utilizes the Wasm module for high-performance image processing, enhancing both processing efficiency and computational performance. Based on this high-performance module, the raw stem cell image data undergoes quality assessment, accurately acquiring image quality parameters to provide a reliable basis for subsequent processing. Image quality analysis and preprocessing strategies are used to optimize the preprocessing workflow, ensuring image processing is more adaptable to the needs of different samples. Image denoising and contrast enhancement improve image quality and enhance the visualization of cell structures and regions, providing clearer image data for subsequent segmentation. Illumination correction and edge optimization eliminate illumination interference and improve edge recognition accuracy, ensuring image data standardization, which is beneficial for subsequent cell segmentation and classification.

[0066] Preferably, step S2 includes the following steps:

[0067] Step S21: Perform Gaussian filtering and stepwise downsampling on the standardized stem cell image data to obtain a multi-scale stem cell image sequence;

[0068] Step S22: Based on multi-scale stem cell image sequences, different resolution levels are organized and arranged to obtain a hierarchical stem cell image pyramid structure;

[0069] Step S23: Perform adjacent-level difference operations on the hierarchical stem cell image pyramid structure, and normalize the results of the adjacent-level difference operations to obtain standardized feature representation data;

[0070] Step S24: Perform information fusion processing at each level based on the standardized feature representation data to obtain multi-level stem cell image representation data.

[0071] As an embodiment of the present invention, reference Figure 3 As shown, Figure 1 A detailed flowchart of step S2 is shown below. In this embodiment of the invention, step S2 includes the following steps:

[0072] Step S21: Perform Gaussian filtering and stepwise downsampling on the standardized stem cell image data to obtain a multi-scale stem cell image sequence;

[0073] In this embodiment of the invention, standardized stem cell image data is first input into a Gaussian filter with a kernel size of 5×5 and a standard deviation of 1.6 for smoothing to reduce image noise and retain key features. Then, a progressive downsampling operation is performed on the filtered stem cell image. The downsampling uses a bilinear interpolation algorithm to successively reduce the image size to 1 / 2, 1 / 4, 1 / 8, and 1 / 16 of the original size. Specifically, the original image I0(x,y) (resolution W×H) is Gaussian filtered to obtain I0'(x,y), and then I0'(x,y) is downsampled to obtain the first-layer image I1(x,y) (resolution W / 2×H / 2). The same Gaussian filter with the same parameters is applied to I1(x,y) to obtain I1'(x,y), and then downsampled to obtain the second layer image I2(x,y) (resolution W / 4×H / 4). This process is repeated to obtain a total of 5 layers of stem cell images with different resolutions. During the downsampling process, a Gaussian filter with the same parameters is applied for preprocessing before each downsampling. Each layer of image maintains consistent pixel depth. For RGB color stem cell images, the above Gaussian filtering and downsampling operations are performed on the R, G, and B channels respectively to finally obtain a multi-scale stem cell image sequence. This sequence preserves the morphological features and cell boundary information of stem cells at different resolutions.

[0074] Step S22: Based on multi-scale stem cell image sequences, different resolution levels are organized and arranged to obtain a hierarchical stem cell image pyramid structure;

[0075] This invention employs a hierarchical stem cell image pyramid structure, meticulously arranging multi-scale stem cell image sequences from high resolution to low resolution. First, the original resolution image I0(x,y)(W×H) is used as the base layer of the pyramid. Downsampled images I1(x,y)(W / 2×H / 2), I2(x,y)(W / 4×H / 4), I3(x,y)(W / 8×H / 8), and I4(x,y)(W / 16×H / 16) are used as the second, third, fourth, and fifth layers, respectively. Next, an index relationship is established between the layers. For any pixel (x,y) in any layer of the pyramid, a mapping relationship is established with the pixel (x / 2,y / 2) in the next layer using a formula, ensuring spatial correspondence between layers. Finally, memory space is allocated for each layer of the image, and pointers are created to it. The system employs a pointer structure for adjacent layers, containing the address of the current layer's image data, image size information, and address references to adjacent upper and lower layers. For color stem cell images, independent pyramid structures are established for the R, G, and B channels to maintain hierarchical consistency between channels. To ensure the integrity of the pyramid structure, a quadtree data structure is used during storage, with each parent node connecting four child nodes. The parent node stores low-resolution image information, and the child nodes store high-resolution image information for the corresponding region. After the pyramid structure is created, the continuity between layers is verified by calculating the mean and standard deviation of each layer's image, ensuring that images at different layers in the pyramid structure can correctly represent the characteristics of stem cells at different scales. Finally, a hierarchical stem cell image pyramid structure with five layers and strict spatial correspondence is output.

[0076] Step S23: Perform adjacent-level difference operations on the hierarchical stem cell image pyramid structure, and normalize the results of the adjacent-level difference operations to obtain standardized feature representation data;

[0077] In this embodiment of the invention, the lower-resolution image Il(x,y) is first upsampled to the same size as the adjacent higher-resolution image Ih(x,y) using a bilinear interpolation algorithm. The interpolation process uses the formula f(x,y)=(1-a)(1-b)f(i,j)+a(1-b)f(i+1,j)+(1-a)bf(i,j+1)+abf(i+1,j+1) to calculate the pixel value of the interpolation point, where a=xi, b=yj, and (i,j) is the nearest integer coordinate point to the upper left corner of the interpolation point. Then, pixel-level difference operations are performed on the upsampled image Il'(x,y) and the corresponding high-resolution image Ih(x,y), with the calculation formula D(x,y)=Ih(x,y)-Il'(x,y) to obtain the difference image D(x,y). The above difference calculation is repeated for all adjacent levels in the pyramid to obtain four sets of differences. Images D1(x,y), D2(x,y), D3(x,y), and D4(x,y) represent the detailed information between adjacent layers. Next, each set of difference images is normalized using the min-max normalization method, calculated as Dn(x,y) = (D(x,y) - Dmin) / (Dmax - Dmin) × 255, where Dmin and Dmax are the minimum and maximum values ​​of the difference image D(x,y), respectively. For color stem cell images, the same normalization process is performed on the difference images of the R, G, and B channels. To eliminate noise interference in the difference images, a hard threshold of 10 is applied to the normalized difference images, setting difference values ​​with absolute values ​​less than the threshold to 0. Finally, the four sets of normalized difference images are combined with the original five-layer pyramid image to form standardized feature representation data.

[0078] Step S24: Perform information fusion processing at each level based on the standardized feature representation data to obtain multi-level stem cell image representation data.

[0079] In this embodiment of the invention, all original and differential images at all levels are first uniformly scaled to the same resolution W×H as the original stem cell image. The scaling method uses a bicubic interpolation algorithm, with the interpolation kernel function h(x)=-(α+2)|x|3+(α+3)|x|2. When |x|<1, h(x)=α|x|3-5α|x|2+8α|x|-4α. When 1≤|x|<2, the parameter α is set to -0.5. Then, based on the importance of image edge and texture features, different weights are assigned to each level of image. The weight of the original image I0(x,y) is set to 0.35, and the weights of the differential images D1(x,y), D2(x,y), D3(x,y), and D4(x,y) are set to 0.25, 0.20, and 0.20, respectively. The weights are 0.12 and 0.08, with a total weight of 1. Then, for each pixel position (x, y), a weighted summation operation M(x, y) is performed: M(x, y) = 0.35 × I0(x, y) + 0.25 × D1(x, y) + 0.20 × D2(x, y) + 0.12 × D3(x, y) + 0.08 × D4(x, y), generating a fused image M(x, y). For the edge information in the fused image, the Sobel operator is applied for enhancement processing. The horizontal Sobel operator is [-1, -2, -1; 0, 0, 0; 1, 2, 1], and the vertical Sobel operator is [-1, 0, 1; -2, 0, 2; -1, 0, 1]. The gradient magnitude G(x, y) of each pixel is calculated as √(Gx...). 2 +Gy2), where Gx and Gy are the gradient values ​​in the horizontal and vertical directions, respectively; non-maximum suppression is applied to the enhanced image to retain local gradient maximum points and suppress non-maximum points, highlighting the stem cell boundary features; finally, histogram equalization is performed on the processing result to improve image contrast, and the calculation formula is T(rk)=(L-1) / MN×∑i=0 to k ni, where L is the number of gray levels, M×N is the image size, and ni is the number of pixels with gray value i; the multi-level stem cell image representation data obtained through the above operations.

[0080] This invention reduces image noise and achieves multi-scale image representation through Gaussian filtering and progressive downsampling, enhancing the ability to extract information at different scales. By organizing and arranging multi-scale images into a hierarchical pyramid structure, different resolution levels of the image are effectively integrated, providing richer multi-level information for subsequent processing. Through differential operations and normalization between adjacent levels, detailed features of the image are effectively extracted, improving the image's recognizability and representational capabilities. By fusing information from various levels, multi-level stem cell image representation data is formed, providing comprehensive cell image information for subsequent analysis and processing.

[0081] Preferably, the adjacent-level difference operation on the hierarchical stem cell image pyramid structure described in step S23 includes:

[0082] The hierarchical stem cell image pyramid structure is subjected to adjacent-level interpolation and alignment processing to obtain scale-matched stem cell image data.

[0083] This invention employs interpolation and alignment processing on adjacent layers of a hierarchical stem cell image pyramid structure to achieve scale-matched stem cell image data acquisition. First, an image pyramid method is used to generate multi-layered image data, where each layer represents a different scale and has a different resolution. Then, considering the size differences between adjacent layers, an interpolation algorithm (such as bilinear interpolation or cubic interpolation) is used to scale-align each layer, ensuring that adjacent layers are size-matched. This interpolation and alignment operation must be performed while ensuring that the feature information of the stem cell image is not distorted as much as possible. During the operation, precise control of the transition between pixels during interpolation calculations is required to avoid excessive smoothing or over-enhancing of image information, ensuring that key information is preserved during the alignment process between layers. After interpolation, scale-matched image data is obtained.

[0084] Pixel-level subtraction between adjacent layers is performed on scale-matched stem cell image data to obtain interlayer difference information data.

[0085] This invention employs scale-matched stem cell image data to perform pixel-level subtraction operations between adjacent layers to extract interlayer difference information. First, scale-matched image data is used as input, and each layer is processed pixel-by-pixel. Within each layer, the corresponding pixel values ​​of adjacent layers are subtracted from the pixel values ​​of the lower-resolution layer to obtain a difference image. This difference image reflects changes in image content at different scales and possesses certain interlayer information. To ensure the accuracy of the calculation process, the difference of each pixel is processed layer by layer to ensure pixel alignment accuracy between the two layers. During the operation, the calculation precision of the subtraction operation can be controlled to ensure that the difference information between adjacent layers is fully reflected in the difference image without losing detailed information. After the subtraction operation is completed, the interlayer difference information data is obtained.

[0086] Nonlinear enhancement processing is applied to the interlayer difference information data to obtain data of salient feature regions;

[0087] This invention first uses inter-layer difference information data as input. Nonlinear enhancement processing transforms pixel values ​​in the image by applying a series of nonlinear functions. These nonlinear functions can enhance subtle details in the image while preserving its overall structure. Especially in stem cell images, subtle structural differences have a significant impact on classification results; therefore, enhancement processing is necessary to make these details more prominent. Parameter settings during enhancement should be adjusted according to image features. Common adjustment methods include controlling the enhancement intensity and processing range to ensure enhancement of details in local areas without affecting the integrity of global information. During enhancement, a layer-by-layer operation is adopted, performing nonlinear transformations on each layer of the image individually. Finally, the processed results from each layer are merged into salient feature region data.

[0088] Pure differential feature data is obtained by performing adaptive threshold filtering of low signal-to-noise ratio regions based on data from salient feature regions.

[0089] This invention first inputs salient feature region data, which includes stem cell image information after nonlinear enhancement processing. Next, an adaptive thresholding method is used to process low signal-to-noise ratio (SNR) regions. Specifically, the noise level of each pixel's region is determined by calculating the local statistical features of each pixel in the image. Then, an adaptive threshold is set; pixels below this threshold are considered noise and filtered out, while pixels above the threshold are retained to ensure that important feature regions in the image are not weakened. The threshold is determined based on the local SNR of each region in the image and dynamically adjusted according to the image content, thus enabling targeted processing of noise in different regions. The entire filtering process is performed pixel-by-pixel, combined with local region features for selection, thereby retaining useful differential information in the stem cell image, filtering out low SNR noise, and ultimately obtaining clean differential feature data.

[0090] The pure differential feature data is integrated at multiple scales to obtain hierarchical feature set data.

[0091] This invention first inputs clean differential feature data, which includes cell image features filtered through low signal-to-noise ratio regions. The integration process is completed by fusing the feature data layer by layer at multiple scales. First, based on the image pyramid structure, the clean differential feature data is divided into multiple scale levels, representing different resolutions and image details. At each level, feature information is extracted according to its resolution and integrated with data from adjacent levels. The integration operation includes pixel-level weighted averaging or interpolation to ensure effective alignment of feature data at different scales during integration and to preserve the multi-level information of the image. Specifically, the influence of image features at each scale is adjusted by weighting coefficients, and the weights can be determined based on the detail complexity or resolution of the image. Finally, features from all scales are summarized and integrated into a unified hierarchical feature set.

[0092] Based on the hierarchical feature set data, the features of each level are weighted and combined to obtain the difference operation results of adjacent levels.

[0093] This invention first inputs a hierarchical feature set, which includes stem cell image features extracted at different scales and resolutions. To achieve weighted feature combination, the feature data of each layer is first weighted, with the weights set based on the resolution, detail richness, and contribution of each layer to the final classification result. Specifically, the contribution of each layer's features is calculated, which can be determined by analyzing the pixel entropy or texture complexity of each layer. Then, weighting coefficients are applied to amplify or reduce the features of each layer, ensuring that important scale features dominate the combination. Next, by performing addition operations or pixel-by-pixel stitching on the weighted features of each layer, the feature data at different scales are integrated into a unified feature vector or feature map, completing the weighted combination of features at each level. Finally, the difference operation results between adjacent layers are obtained.

[0094] This invention improves the alignment accuracy of different image layers through scale matching, ensuring image information consistency. It effectively extracts inter-layer differences, highlighting changes in cell structure at different resolutions. By enhancing salient features, it improves the recognizability of cell regions, making key information more prominent. By filtering out noisy areas, it improves the purity of image data, ensuring the accuracy of subsequent analysis. By integrating information from different scales, it enhances the multidimensional representation of features, improving cell recognition. By assigning different weights to features at each layer, it optimizes feature combinations, making the difference calculation results more accurate.

[0095] Preferably, the multi-resolution hierarchical processing of the multi-level stem cell image representation data in step S3 includes:

[0096] Preliminary cell region identification is performed on the low-resolution layer in the multi-level stem cell image representation to obtain potential cell region data;

[0097] This invention first extracts stem cell image data from low-resolution layers and uses image processing methods, such as thresholding or edge detection algorithms, for preliminary cell region identification. For low-resolution layers, cell regions in the image are often blurry; therefore, potential cell regions are identified by calculating the brightness or color features of local areas. For example, the Otsu method is used to perform global thresholding on the grayscale image to distinguish cell regions from the background. Further screening of potential cell regions involves morphological processing (such as dilation and erosion) to remove noise and fill small holes, resulting in more accurate preliminary cell region outlines. Next, by calculating the connectivity of adjacent regions, potential cell regions in the image are separated and labeled to ensure accurate identification of each potential cell region. Finally, the potential cell region data is obtained.

[0098] Cell aggregation region analysis was performed on the medium-resolution layer in the multi-level stem cell image representation to obtain cell aggregation region data;

[0099] This invention first extracts stem cell image data from a medium-resolution layer and preprocesses it, such as removing noise and enhancing contrast, to improve the recognizability of cell regions. Next, clustering algorithms (such as K-means clustering or peak density clustering) are applied to analyze the cell regions in the image and identify cell clusters. A cluster refers to a region of cells with a high density in the image. The clustering algorithm calculates the pixel value features of each region in the image to group the cells into different populations or clusters. During the cell cluster identification process, morphological operations (such as dilation and erosion) are further applied to smooth the edges of the clusters, remove isolated noise points, and fill in holes caused by unclear cluster boundaries. Furthermore, connectivity analysis is used to ensure that cell regions within a cluster are effectively connected, avoiding misidentification of clusters due to image resolution or loss of detail during processing. Finally, cell cluster data is obtained.

[0100] Fine processing of cell overlap regions is performed on high-resolution layers in multi-level stem cell image representation to obtain cell overlap region data;

[0101] This invention first extracts stem cell image data from a high-resolution layer, which typically displays minute details between cells more clearly. Cell overlap regions are areas where cells contact or overlap, and these areas significantly influence cell classification. To refine the processing of cell overlap regions, precise edge detection is first performed on the image using the Canny edge detection algorithm to extract cell boundary information. Then, morphological operations (such as opening and closing operations) are combined to smooth the boundaries, removing unclear areas caused by noise or resolution limitations. Based on this, connectivity analysis is used to segment the overlap regions, ensuring that cells in overlapping areas are separated. The overlap regions are further refined by calculating the contact area and overlap characteristics of cell boundaries. In the identification of overlap regions, a region growing algorithm is used to segment contacting cells, preventing overlapping areas from being misclassified as single cells. Finally, cell overlap region data is obtained.

[0102] The potential cell region data, cell aggregation region data, and cell overlap region data are recorded as multi-level cell feature data.

[0103] This invention first inputs potential cell region data, cell aggregation region data, and cell overlap region data. Potential cell region data comes from preliminary cell region identification at low resolution layers, cell aggregation region data from aggregation region analysis at medium resolution layers, and cell overlap region data from fine-tuning of overlap regions at high resolution layers. To represent these three types of data uniformly, image processing methods are first used to normalize the data at each level. Then, image overlay technology is used to fuse the three types of data in hierarchical order, forming a multi-level cell feature set. The fusion process uses pixel-level weighted overlay to combine the cell region features of each layer with data from other layers, where the weighting coefficients are adjusted according to the resolution and importance of each layer's data. In this way, multi-level cell feature data is obtained.

[0104] This invention rapidly locates potential cellular regions, providing a preliminary framework for subsequent detailed analysis. It identifies cell aggregation areas, improving the accuracy of cell aggregation detection. It accurately handles overlapping cell regions, resolving identification problems caused by cell overlap. It integrates information from different resolution levels into comprehensive cellular features, ensuring the comprehensiveness and accuracy of subsequent analysis.

[0105] Preferably, step S4 includes the following steps:

[0106] Step S41: Process the potential cell region data using a region growing algorithm to obtain the initial segmentation region data;

[0107] This invention first inputs latent cell region data and extracts the initially identified cell regions from the image. A region growing algorithm processes these regions by selecting seed points, which are typically certain pixels within the latent cell region that reflect the cell's edge or center. Starting from these seed points, the algorithm gradually expands based on the similarity of neighboring pixels (such as grayscale values, color, or texture features) until a set threshold condition is met, such as the similarity metric reaching a predetermined standard or the grown region reaching a maximum preset area. The selection of neighboring pixels during the expansion process is usually based on connectivity principles, ensuring the continuity of region growth by judging the spatial relationships between pixels (such as 8-neighborhood or 4-neighborhood). The region growing algorithm continuously expands the cell region until it covers the entire latent cell region, generating preliminary segmentation region data.

[0108] Step S42: Optimize the cell boundary activity contour model based on the initial segmented region data to obtain accurate cell boundary data;

[0109] This invention first uses initial segmented region data as input, and then optimizes the cell boundary active contour model. By combining the initial segmented region data with the active contour model, the boundary accuracy is optimized. The active contour model is optimized based on image gradient information. By calculating the image gradient field, regions with significant brightness changes in the image are identified as preliminary estimates of the cell boundaries. Through model iteration, the boundaries gradually approach the actual positions of the cell boundaries in the image. During the iteration process, the boundaries are adjusted according to changes in image gradient values, and a mechanical model is used to simulate the contraction and expansion of the boundaries to ensure they conform to the shape characteristics of the cells. This optimization process is achieved by minimizing an energy function, which is calculated based on the difference between the cell boundaries and image features. The boundaries are adjusted in the direction of reducing energy, ultimately obtaining accurate cell boundary data.

[0110] Step S43: Perform watershed transformation on the cell aggregation region data and cell overlap region data to obtain separated cell unit data;

[0111] This invention first utilizes precise cell boundary data and preliminary segmentation region data to perform watershed transform processing on cell aggregation region data and cell overlap region data. Watershed transform is a gradient-based segmentation method used to handle cell regions with different densities. First, gradient calculations are performed on the images of cell aggregation regions and cell overlap regions to obtain a gradient map. Then, this gradient map is transformed into a topographic map model, and segmentation is performed by simulating the process of water flow from low to high points. Each pixel in the image is considered the "starting point of the water flow," gradually expanding from low grayscale areas to the surrounding areas based on gradient information. When the water flow encounters a high grayscale area, the flow is "blocked," forming a watershed line, thus dividing the image into different regions. Finally, the separated cell unit data is obtained.

[0112] Step S44: Verify cell integrity based on accurate cell boundary data and separated cell unit data to obtain valid cell object data;

[0113] This invention first verifies cell integrity based on precise cell boundary data and separated cell unit data. This process ensures the structural integrity of each cell by analyzing its shape, size, and boundary continuity. First, each separated cell unit undergoes shape feature analysis, calculating the smoothness, convex hull, and boundary closure parameters of the cell boundary. If some cells have cracks or obviously irregular shapes at their boundaries, the cells are considered incomplete and require further processing. At this point, the boundaries are compared with those in the precise cell boundary data to check for consistency. If the boundaries do not precisely coincide, there is missegmentation or overlap. Thresholds are set to limit the area and shape of cell units, filtering out those that do not meet the boundary matching requirements. Finally, cells that are morphologically complete and match the precise boundary data are retained, resulting in valid cell object data.

[0114] Step S45: Perform shape constraint filtering on the valid cell object data to obtain pure cell set data;

[0115] This invention, after obtaining valid cell object data, performs shape constraint filtering to further improve the accuracy of the cell data. By analyzing and filtering the shapes of valid cell objects, cells that do not conform to the characteristics of stem cells are excluded. First, shape constraints are set based on the shape characteristics of the cells. By calculating the geometric features of each cell object, it is determined whether it meets the preset cell morphology requirements. For example, for the shape of stem cells, the cell boundaries are required to be relatively smooth and nearly circular, and the aspect ratio of the cell is within a reasonable range. Cell objects that do not meet these conditions are removed or adjusted based on the geometric calculation results. Cell objects that meet the shape constraints are retained, forming a pure cell set data.

[0116] Step S46: Reconstruct the original image region based on the pure cell set data to obtain segmented stem cell image data.

[0117] This invention first relabels the cells in the original image using the boundary information of each cell in the purified cell set data. The position and shape of each cell in the image are reconstructed by matching the cell object's position in the original image with its corresponding labeled region. Cell contour data is extracted from the purified cell set and used as a reference to align the cell contours with the original image at the pixel level. For each valid cell object, its corresponding pixel region is extracted from the original image, and the cell region in the original image is accurately reconstructed based on the cell contour, ultimately forming the labeled segmented stem cell image data. In this process, each cell region is accurately mapped back to the original image, and the pixel values ​​of these regions are updated to the corresponding cell label values. The reconstructed image data is the segmented stem cell image data, which can be used for subsequent analysis or classification.

[0118] This invention effectively extracts potential cellular regions using a region growing algorithm, providing preliminary cell segmentation regions. Optimization through an active contour model accurately determines cell boundaries, improving the accuracy of the segmentation results. Watershed transformation effectively handles cell aggregation and overlapping regions, achieving accurate separation of cell units. Cell integrity verification ensures the validity of the segmentation results, removing incomplete or erroneous cell segmentation regions. Shape constraint filtering excludes regions that do not conform to stem cell morphology, obtaining clean cell ensemble data. Label reconstruction generates high-quality stem cell image segmentation results, providing reliable data for subsequent analysis.

[0119] Preferably, step S43 includes the following steps:

[0120] Step S431: Perform distance transformation calculation on the cell aggregation region data, and extract local maxima of the image region from the distance transformation calculation results to obtain the set of cell seed points;

[0121] This invention first acquires cell cluster region data and performs distance transformation calculations. Distance transformation generates a distance map by calculating the distance from each pixel in the image to its nearest background pixel, where each pixel value represents its distance to the nearest background pixel. For cell cluster region data, the purpose of distance transformation is to highlight the structural information of the cell region. Next, local maxima extraction is performed on the distance transformation results. Local maxima extraction involves comparing neighborhoods in the distance transformation map to find the maximum value point within each local region. These local maxima points represent the seed points of the cells, typically located at the center of the cell nucleus, and can accurately identify the cell's location. To ensure the quality of the seed points, a thresholding process is applied to the local maxima to remove noise and filter out seed points with higher reliability. Finally, a set of cell seed points is obtained, which contains the core location of each cell in the cell cluster region.

[0122] Step S432: Assign unique identifiers to the set of cell seed points to obtain a cell-labeled image;

[0123] In this embodiment of the invention, after obtaining the set of cell seed points, a unique identifier is assigned to each seed point. First, each seed point in the set is assigned a unique identifier based on its position in the image. For this purpose, the pixel coordinate system in the image is used, and a labeling algorithm is employed to assign a unique integer identifier to each seed point. This is done by traversing each seed point in the image and assigning a unique identifier to each one based on its spatial position. The assignment of identifiers is determined based on their position in the image. After assigning identifiers, a cell-labeled image is generated, in which each cell seed point corresponds to a unique identifier, while areas other than seed points remain as background or unlabeled.

[0124] Step S433: Based on the cell-labeled image, the cell overlap area data is processed by the watershed algorithm, and the watershed algorithm processing result is optimized by boundary smoothing to obtain fine cell boundary line data;

[0125] This invention first utilizes cell-labeled images and applies a watershed algorithm to the overlapping cell regions. By calculating the gradient map of the cell-labeled image, intensity variation information in different regions of the image is obtained. Then, the gradient map is transformed into a topographic map model, where low grayscale areas represent depressions and high grayscale areas represent ridgelines. The watershed algorithm simulates the process of water flowing from low grayscale areas to high grayscale areas, finding boundaries and segmenting them within the gradient map. Each cell seed point in the cell-labeled image serves as the starting point for the watershed algorithm, which gradually expands across the topographic map until it encounters the boundaries of other cells. After watershed segmentation, preliminary cell separation regions are obtained. Next, the watershed algorithm processing results are optimized for boundary smoothing. Boundary smoothing is achieved through filtering or curve fitting techniques to reduce jagged boundaries caused by image noise or gradient calculation errors. Finally, refined cell boundary line data is obtained.

[0126] Step S434: Based on the fine cell boundary line data, perform connected component labeling of the separated regions to obtain separated cell unit data.

[0127] In this embodiment of the invention, after obtaining fine cell boundary line data, connected component labeling of separated regions is performed based on this boundary line data. First, using the binarized image result (i.e., boundary line data), each independent cell region is labeled as a connected region. The connected component labeling algorithm traverses each pixel in the image, checking whether it belongs to the same connected region as its neighboring pixels. If neighboring pixels belong to the same region, that pixel is merged with the already labeled region until all cell regions are accurately labeled. The connected component labeling algorithm can be implemented using depth-first search (DFS) or breadth-first search (BFS) methods, both of which label each connected region recursively or in a queue. Finally, separated cell unit data is obtained.

[0128] This invention accurately locates seed points for cells through distance transformation and local maximum extraction, providing an initial basis for cell segmentation. A unique identifier is assigned to each cell seed point to ensure independent identification of each cell in subsequent processing. The watershed algorithm and boundary smoothing optimization effectively separate overlapping cell regions, providing accurate cell boundary data. Connected component labeling effectively separates cell units, laying the foundation for independent analysis of stem cells.

[0129] Preferably, step S5 includes the following steps:

[0130] Step S51: Calculate the geometric morphological parameters of the segmented stem cell image data to obtain basic cell morphological feature data;

[0131] In this embodiment of the invention, after obtaining segmented stem cell image data, the geometric morphological parameters of the image are first calculated. Geometric morphological analysis mainly involves the morphological features of cells in the image. First, by extracting the boundary contour of each cell region, the area and perimeter of each cell are calculated. Based on this, shape factors (such as cell roundness) are calculated, which reflects the degree to which the cell shape approximates a circle. The aspect ratio of the cells is further calculated to assess the proportional characteristics of their shape. Morphological transformation tools, such as erosion, dilation, opening, and closing operations, are used to further remove noise from the image and extract more accurate cell shape information. Through the calculation of these geometric morphological parameters, the basic morphological feature data of each cell can be extracted.

[0132] Step S52: Perform gray-level co-occurrence matrix analysis of cell regions based on basic cell morphology feature data to obtain cell texture feature data;

[0133] This invention, based on fundamental cell morphology feature data, performs gray-level co-occurrence matrix (GLCM) analysis of cell regions. First, the segmented stem cell image data is converted into a grayscale image; this conversion involves converting the RGB channel data of the color image to grayscale levels. Then, the GLCM of the cell regions in the image is calculated, describing the grayscale relationships between pixel pairs. The calculation of the GLCM involves statistically analyzing the frequency of each pixel pair (i,j) in the image at a specified direction and distance. After obtaining the GLCM, several texture feature indices are further calculated, including contrast, correlation, energy, and uniformity. Through these calculations, data describing cell texture features are obtained.

[0134] Step S53: Calculate the spatial density of cell region distribution based on cell texture feature data to obtain cell density feature data;

[0135] This invention first extracts boundary information of cell regions from segmented stem cell image data based on cell texture feature data, and then subdivides the cell regions, labeling the position and size of each cell. Using this cell region information, spatial distribution density is calculated. The calculation process includes dividing the image region into multiple small grid units and counting the number of cells in each unit. By calculating the number of cells within each grid unit, the spatial distribution density of cells is obtained. Simultaneously, to ensure the accuracy of density calculation, a weighted calculation method can be applied to specific cell clustering or sparse regions based on cell boundaries and morphological characteristics to improve the accuracy of density estimation. Finally, based on the density distribution of cells in different regions, density feature data of the cell regions is obtained.

[0136] Step S54: Perform multi-dimensional feature fusion on the basic cell morphology feature data, cell texture feature data, and cell density feature data to obtain a comprehensive cell feature vector;

[0137] This invention integrates multidimensional features from basic cell morphology data, cell texture data, and cell density data. The morphological, texture, and spatial density features of each cell are standardized to ensure consistency in numerical range for each feature type. The standardized data are then concatenated to form a multidimensional vector containing the cell's geometric shape, texture pattern, and spatial density characteristics. To avoid scale differences affecting the final fusion result, a weighted summation of different features is used during the fusion process to ensure each feature has an appropriate weight in the integration. Through this fusion process, the multidimensional features of the cell are transformed into a unified, integrated cell feature vector.

[0138] Step S55: Perform support vector machine classification on the comprehensive cell feature vector to obtain first-level cell classification data;

[0139] This invention applies Support Vector Machine (SVM) classification to a comprehensive cell feature vector to obtain first-level cell classification data. First, an SVM model is constructed using the fused multidimensional feature vector as input. The SVM constructs an optimal hyperplane to divide the cell feature space into different categories, thus assigning each cell to its corresponding category. During training, a linear kernel function is used for feature mapping, mapping low-dimensional data to a high-dimensional feature space to improve classification accuracy. After training, the comprehensive feature vector of a new cell sample is input into the SVM model, and predictions are made based on the hyperplane's position and category boundaries to obtain first-level cell classification data.

[0140] Step S56: Perform deep learning sub-classification of cell subclasses based on primary cell classification data to obtain stem cell classification results data.

[0141] This invention first establishes a deep neural network model using primary cell classification data as input. The input layer of the network model receives feature information from the primary cell classification data and processes it through a multi-layer convolutional neural network (CNN) to extract deeper cell features. The convolutional and pooling layers in the network are iteratively optimized to analyze the details of the cell images and distinguish different cell subclasses. To enhance the model's performance, residual connections (ResNet) are used to address the vanishing gradient problem during deep network training. The output layer of the network classifies according to a predetermined number of categories, ultimately outputting the stem cell classification results. To ensure the accuracy of the deep learning model, a large-scale dataset is used for training, and cross-validation is used to adjust hyperparameters and avoid overfitting. During training, optimization algorithms (such as the Adam optimizer) are used to update the weights.

[0142] This invention extracts the basic morphological features of cells through geometric morphological parameter calculation, providing fundamental morphological information for subsequent analysis. Gray-level co-occurrence matrix analysis reveals the texture features of cellular regions, providing texture information for cell images. The spatial density of cell regions is calculated to reflect the spatial distribution characteristics of cells. Through multi-dimensional feature fusion, cell morphology, texture, and density information are combined to generate a more comprehensive integrated cell feature vector. Support vector machines are used for classification to achieve preliminary classification of stem cells and distinguish different cell types. Deep learning is used to further subdivide cell subclasses, resulting in more refined stem cell classification results.

[0143] Preferably, step S6 includes the following steps:

[0144] Step S61: Optimize and transform the data structure of the stem cell classification results data to obtain standardized storage format data;

[0145] This invention optimizes and transforms the data structure of stem cell classification results. First, a standardized data structure is defined based on the characteristic data of the stem cell classification results. Through structured design, the storage format of all classification results data is unified, for example, using a tabular data structure or JSON format. Next, based on predefined fields, such as cell category, feature values, and identifiers, data fields are standardized to avoid inconsistencies between different data sources. Simultaneously, numerical data in the classification results is standardized. Then, for each cell category and its characteristic data, appropriate data compression and encoding methods (such as Huffman coding or Base64 encoding) are applied to further optimize data storage, reduce data redundancy, and improve storage efficiency. After these processes, standardized storage format data is finally obtained.

[0146] Step S62: Construct a multidimensional index of cell classification information based on standardized storage format data to obtain a cell classification retrieval index;

[0147] This invention first constructs a multi-dimensional index structure for cell classification information based on standardized storage format data. Suitable data fields, such as cell category, cell morphology features, texture features, and density features, are selected as the dimensions of the index. By indexing each dimension, it ensures that each cell classification data can be quickly retrieved from multiple perspectives. During the construction process, B-tree or KD-tree data structures are used to spatially partition the cell data and sort it according to cell category and related features. Then, based on the different categories of data in the cell classification results, an index for cell categories is constructed, and these index information are further associated with the original data. Furthermore, for cell data with high-dimensional features, clustering algorithms are used for dimensionality reduction while preserving the key features of the cell data. Next, incremental updates are performed on each dimension of the index to ensure that the index structure can be updated promptly as new data is added. Through these operations, an efficient, dynamically updatable cell classification retrieval index is ultimately formed.

[0148] Step S63: Perform hash partitioning on the cell classification retrieval index, and based on the hash partitioning results, fragment the standardized storage format data to obtain a multi-node stem cell data cluster;

[0149] This invention first performs hash partitioning based on a cell classification retrieval index. First, relatively stable and highly discriminative feature fields from the cell classification information, such as cell type, morphological parameters, or texture features, are selected as the basis for hash partitioning. For these features, a consistent hash algorithm (such as MD5, SHA-256, or a custom hash function) is applied to calculate the hash value for each cell classification information. Based on the hash value, the data is divided into multiple data partitions. Each partition contains a portion of data, and the distribution of hash values ​​ensures that the data is evenly distributed across the partitions, thus avoiding excessive load on some partitions. After hash partitioning, the cell data in each partition is stored in shards. Each partition's data is stored in a separate storage unit, such as a database table, file system, or distributed storage node, and each storage unit is considered an independent shard. During storage, the storage path is mapped based on the partition's hash value. This ultimately forms a multi-node stem cell data cluster.

[0150] Step S64: Perform consistency verification on the multi-node stem cell data cluster and integrate the metadata to obtain a persistent stem cell classification database.

[0151] This invention first performs a consistency check on a multi-node stem cell data cluster. During the check, a hash algorithm is used to re-hash the content of each data shard. Specifically, using the same hash algorithm as when hashing the partition, the hash value of each data shard is compared with the original value. If inconsistencies are found, a data recovery mechanism is triggered to recover the lost or damaged parts from the original data source or backup data. After the consistency check is completed, metadata integration is performed. Metadata integration involves classifying and organizing the index information, storage path, and data type metadata of each data shard to establish a unified metadata management process. The metadata of each shard is recorded in a dedicated metadata table. At this point, the storage information of all cell classification data has been fully integrated according to a specific storage structure and indexing system, forming a complete persistent stem cell classification database.

[0152] This invention optimizes and transforms data structures to ensure standardized storage formats for stem cell classification results, improving data processing compatibility and operability. Multidimensional indexing enhances the retrieval efficiency of cell classification information, supporting fast and accurate classification data queries. Hash partitioning and sharding distribute data across multiple nodes, optimizing data storage management and improving system scalability and load balancing. Consistency checks and metadata integration ensure data integrity and accuracy, persisting it as a reliable stem cell classification database.

[0153] Preferably, the present invention also provides an image segmentation-based stem cell classification system for performing the above-described image segmentation-based stem cell classification method, wherein the image segmentation-based stem cell classification system includes:

[0154] The image preprocessing and standardization module is used to acquire raw stem cell image data, perform Wasm processing on the raw stem cell image data to obtain stem cell image quality parameters, and perform adaptive preprocessing on the raw stem cell image data based on the stem cell image quality parameters to obtain standardized stem cell image data.

[0155] The multi-resolution pyramid construction module is used to construct multi-resolution pyramids from standardized stem cell image data to obtain multi-level stem cell image representation data.

[0156] The multi-resolution hierarchical feature extraction module is used to perform multi-resolution hierarchical processing on multi-level stem cell image representation data to obtain multi-level cell feature data, which includes potential cell region data, cell aggregation region data and cell overlap region data.

[0157] The stem cell adaptive segmentation module is used to perform stem cell adaptive algorithm segmentation based on multi-level cell feature data to obtain segmented stem cell image data.

[0158] The morphological feature extraction and classification module is used to extract morphological features from segmented stem cell image data and perform hierarchical classification based on the morphological feature extraction results to obtain stem cell classification result data.

[0159] The distributed storage and persistence module is used to process stem cell classification results data in a distributed manner and generate a persistent stem cell classification database.

[0160] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is not limited by the foregoing description. Thus, all changes falling within the meaning and scope of the equivalents of the application are intended to be included within the scope of the invention.

[0161] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A stem cell classification method based on image segmentation, characterized in that, Includes the following steps: Step S1: Obtain raw stem cell image data, perform Wasm processing on the raw stem cell image data to obtain stem cell image quality parameters; perform adaptive preprocessing on the raw stem cell image data based on the stem cell image quality parameters to obtain standardized stem cell image data. Step S2: Construct a multi-resolution pyramid from the standardized stem cell image data to obtain multi-level stem cell image representation data; Step S3: Perform multi-resolution layering processing on the multi-level stem cell image representation data to obtain multi-level cell feature data, which includes potential cell region data, cell aggregation region data, and cell overlap region data. The multi-resolution layering processing of the multi-level stem cell image representation data in step S3 includes: Preliminary cell region identification is performed on the low-resolution layer in the multi-level stem cell image representation to obtain potential cell region data; Cell aggregation region analysis was performed on the medium-resolution layer in the multi-level stem cell image representation to obtain cell aggregation region data; Fine processing of cell overlap regions is performed on high-resolution layers in multi-level stem cell image representation to obtain cell overlap region data; The potential cell region data, cell aggregation region data, and cell overlap region data are recorded as multi-level cell feature data; Step S4: Perform stem cell adaptive algorithm segmentation based on multi-level cell feature data to obtain segmented stem cell image data. Step S4 includes the following steps: Step S41: Process the potential cell region data using a region growing algorithm to obtain the initial segmentation region data; Step S42: Optimize the cell boundary activity contour model based on the initial segmented region data to obtain accurate cell boundary data; Step S43: Perform watershed transformation on the cell aggregation region data and cell overlap region data to obtain separated cell unit data; Step S44: Verify cell integrity based on accurate cell boundary data and separated cell unit data to obtain valid cell object data; Step S45: Perform shape constraint filtering on the valid cell object data to obtain pure cell set data; Step S46: Reconstruct the original image region based on the pure cell set data to obtain segmented stem cell image data; Step S5: Extract morphological features from the segmented stem cell image data, and perform hierarchical classification based on the morphological feature extraction results to obtain stem cell classification result data; Step S6: Perform distributed storage processing on the stem cell classification results data to generate a persistent stem cell classification database.

2. The stem cell classification method based on image segmentation according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain raw stem cell image data, compile the raw stem cell image data using the Wasm module to obtain a high-performance image processing module; Step S12: The quality of the original stem cell image data is evaluated based on the high-performance image processing module to obtain stem cell image quality parameters; Step S13: Perform image quality analysis and preprocessing strategy determination on stem cell image quality parameters to obtain the optimal preprocessing strategy; Step S14: Based on the optimal preprocessing strategy, perform image denoising and contrast enhancement on the original stem cell image data to obtain enhanced stem cell image data; Step S15: Perform illumination correction and edge optimization on the enhanced stem cell image data to obtain standardized stem cell image data.

3. The stem cell classification method based on image segmentation according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Perform Gaussian filtering and stepwise downsampling on the standardized stem cell image data to obtain a multi-scale stem cell image sequence; Step S22: Based on multi-scale stem cell image sequences, different resolution levels are organized and arranged to obtain a hierarchical stem cell image pyramid structure; Step S23: Perform adjacent-level difference operations on the hierarchical stem cell image pyramid structure, and normalize the results of the adjacent-level difference operations to obtain standardized feature representation data; Step S24: Perform information fusion processing at each level based on the standardized feature representation data to obtain multi-level stem cell image representation data.

4. The stem cell classification method based on image segmentation according to claim 3, characterized in that, Step S23, which involves performing adjacent-level difference operations on the hierarchical stem cell image pyramid structure, includes: The hierarchical stem cell image pyramid structure is subjected to adjacent-level interpolation and alignment processing to obtain scale-matched stem cell image data. Pixel-level subtraction between adjacent layers is performed on scale-matched stem cell image data to obtain interlayer difference information data. Nonlinear enhancement processing is applied to the interlayer difference information data to obtain data of salient feature regions; Pure differential feature data is obtained by performing adaptive threshold filtering of low signal-to-noise ratio regions based on data from salient feature regions. The pure differential feature data is integrated at multiple scales to obtain hierarchical feature set data. Based on the hierarchical feature set data, the features of each level are weighted and combined to obtain the difference operation results of adjacent levels.

5. The stem cell classification method based on image segmentation according to claim 1, characterized in that, Step S43 includes the following steps: Step S431: Perform distance transformation calculation on the cell aggregation region data, and extract local maxima of the image region from the distance transformation calculation results to obtain the set of cell seed points; Step S432: Assign unique identifiers to the set of cell seed points to obtain a cell-labeled image; Step S433: Based on the cell-labeled image, the cell overlap area data is processed by the watershed algorithm, and the watershed algorithm processing result is optimized by boundary smoothing to obtain fine cell boundary line data; Step S434: Based on the fine cell boundary line data, perform connected component labeling of the separated regions to obtain separated cell unit data.

6. The stem cell classification method based on image segmentation according to claim 1, characterized in that, Step S5 includes the following steps: Step S51: Calculate the geometric morphological parameters of the segmented stem cell image data to obtain basic cell morphological feature data; Step S52: Perform gray-level co-occurrence matrix analysis of cell regions based on basic cell morphology feature data to obtain cell texture feature data; Step S53: Calculate the spatial density of cell region distribution based on cell texture feature data to obtain cell density feature data; Step S54: Perform multi-dimensional feature fusion on the basic cell morphology feature data, cell texture feature data, and cell density feature data to obtain a comprehensive cell feature vector; Step S55: Perform support vector machine classification on the comprehensive cell feature vector to obtain first-level cell classification data; Step S56: Perform deep learning sub-classification of cell subclasses based on primary cell classification data to obtain stem cell classification results data.

7. The stem cell classification method based on image segmentation according to claim 1, characterized in that, Step S6 includes the following steps: Step S61: Optimize and transform the data structure of the stem cell classification results data to obtain standardized storage format data; Step S62: Construct a multidimensional index of cell classification information based on standardized storage format data to obtain a cell classification retrieval index; Step S63: Perform hash partitioning on the cell classification retrieval index, and based on the hash partitioning results, fragment the standardized storage format data to obtain a multi-node stem cell data cluster; Step S64: Perform consistency verification on the multi-node stem cell data cluster and integrate the metadata to obtain a persistent stem cell classification database.

8. A stem cell classification system based on image segmentation, characterized in that, For performing the image segmentation-based stem cell classification method as described in claim 1, the image segmentation-based stem cell classification system comprises: The image preprocessing and standardization module is used to acquire raw stem cell image data, perform Wasm processing on the raw stem cell image data to obtain stem cell image quality parameters, and perform adaptive preprocessing on the raw stem cell image data based on the stem cell image quality parameters to obtain standardized stem cell image data. The multi-resolution pyramid construction module is used to construct multi-resolution pyramids from standardized stem cell image data to obtain multi-level stem cell image representation data. A multi-resolution hierarchical feature extraction module is used to perform multi-resolution hierarchical processing on multi-level stem cell image representation data to obtain multi-level cell feature data, wherein the multi-level cell feature data includes potential cell region data, cell aggregation region data, and cell overlap region data. The multi-resolution hierarchical processing of the multi-level stem cell image representation data includes: Preliminary cell region identification is performed on the low-resolution layer in the multi-level stem cell image representation to obtain potential cell region data; Cell aggregation region analysis was performed on the medium-resolution layer in the multi-level stem cell image representation to obtain cell aggregation region data; Fine processing of cell overlap regions is performed on high-resolution layers in multi-level stem cell image representation to obtain cell overlap region data; The potential cell region data, cell aggregation region data, and cell overlap region data are recorded as multi-level cell feature data; The stem cell adaptive segmentation module is used for stem cell adaptive algorithm segmentation based on multi-level cell feature data to obtain segmented stem cell image data, including: The potential cell region data is processed by a region growing algorithm to obtain the initial segmentation region data; The cell boundary activity contour model is optimized based on the initial segmented region data to obtain accurate cell boundary data. Watershed transformation was performed on the data of cell aggregation regions and cell overlap regions to obtain data of separated cell units. Cell integrity is verified based on precise cell boundary data and separated cell unit data to obtain valid cell object data; Shape constraint filtering is applied to the valid cell object data to obtain a pure cell set data; Original image region labeling and reconstruction were performed based on pure cell set data to obtain segmented stem cell image data; The morphological feature extraction and classification module is used to extract morphological features from segmented stem cell image data and perform hierarchical classification based on the morphological feature extraction results to obtain stem cell classification result data. The distributed storage and persistence module is used to process stem cell classification results data in a distributed manner and generate a persistent stem cell classification database.

Images