Stem cell quality prediction system and method based on multi-modal data

Abstract

The invention belongs to the technical field of image processing, and discloses a stem cell quality prediction system and method based on multi-modal data, and the method comprises the steps: collecting to-be-evaluated stem cell images according to a standard time interval, and extracting a stem cell region to generate an image sequence; dividing each image into a plurality of blocks, and marking a tracking target in the first image; screening out a regional reference image sequence and a single cell reference image sequence matched with the to-be-evaluated stem cell culture condition from a stem cell reference library; calculating the overall similarity between the stem cell region image sequence and the region reference image sequence; calculating an overall tracking rate according to the tracking target; and finally, evaluating the quality of the stem cells by integrating the overall similarity and the overall tracking rate. By comparing the complete development process of the stem cells to be evaluated with the complete development process of the high-quality stem cells in the reference library, the group characteristics and the individual performance are considered at the same time, the limitation of a traditional static evaluation method is overcome, and the comprehensiveness and the accuracy of evaluation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, more particularly, the present application relates to a stem cell quality prediction system and method based on multi-modal data. BACKGROUND

[0002] Stem cells, with their self-renewal ability and multi-directional differentiation potential, have shown great application prospects in the fields of regenerative medicine, tissue engineering, drug screening and disease modeling. With the rapid development of stem cell research and application, stem cell quality evaluation is of great significance.

[0003] In traditional stem cell quality evaluation methods, biological chemistry analysis and molecular biology techniques are mainly relied on, such as flow cytometry, immunohistochemistry, gene expression analysis, etc. Although these methods can provide molecular characteristic information of stem cells, they usually require fixation or lysis treatment of cells, which makes the cells unable to continue to be used for subsequent experiments or clinical applications. In addition, these methods are complex, time-consuming and costly, which makes it difficult to meet the quality control requirements in large-scale stem cell production and application. In addition, the stem cell quality evaluation methods based on image analysis mostly use static image analysis technology, only focusing on the morphological characteristics of stem cells at a certain time point. This static evaluation method cannot capture the dynamic changes of stem cells in the culture process, and the quality of stem cells is not only reflected in the state at a certain moment, but also in the morphological evolution and functional performance in the whole growth and development process. The lack of consideration of the dynamic characteristics of stem cells limits the accuracy and predictability of existing evaluation methods. SUMMARY

[0004] In order to overcome the problem that the accuracy and predictability of the prior art are limited, the present application provides a stem cell quality prediction system and method based on multi-modal data to solve the above problems.

[0005] The present application provides the following technical solutions: A stem cell quality prediction method based on multi-modal data, comprising: The present application also provides a stem cell quality prediction system based on multi-modal data, which is used to implement a stem cell quality prediction method based on multi-modal data, comprising: Obtaining stem cell culture conditions and collecting images of stem cells in culture dishes at predetermined time intervals to obtain a predetermined number of collected images, using image segmentation technology to extract stem cell region images from the collected images, and generating stem cell region image sequences after standardization processing; Dividing each stem cell region image in the stem cell region image sequence into several blocks, and marking a stem cell individual in each block of the first stem cell region image as a tracking target; From the pre-established stem cell reference library, according to the first stem cell area image and the culture condition, a region reference image sequence is screened out, and according to the tracking target and the culture condition, a single cell reference image sequence corresponding to each tracking target is selected; According to the stem cell area image sequence and the region reference image sequence, an overall similarity is obtained; In the stem cell area image sequence, according to the block where each tracking target is located and the corresponding single cell reference image sequence, the tracking rate of each tracking target is calculated, and the mean value of the tracking rates of all tracking targets is used as an overall tracking rate; The overall similarity and the overall tracking rate are integrated to finally evaluate the quality of the stem cell to be evaluated.

[0006] Preferably, the division of each stem cell area image in the stem cell area image sequence into blocks comprises: dividing each image in the stem cell area image sequence into blocks through equally spaced horizontal and vertical division lines; The marking of a stem cell individual in each block in the first stem cell area image as a tracking target comprises: Applying a cell contour recognition technology to detect a stem cell individual in each block; Calculating the average edge gradient of each detected stem cell individual; Selecting the stem cell individual with the highest average edge gradient in each block as the tracking target of the corresponding block.

[0007] Preferably, the stem cell reference library comprises a region reference image sequence set and a single cell reference image sequence set; The establishment of the region reference image sequence set comprises: From the stem cell culture samples historically marked as high quality, stem cell area image sequences under different culture conditions are collected at standard time intervals; the collected image sequences are standardized, the corresponding culture conditions are marked for each region reference image sequence, and a region reference image sequence set is stored; The establishment of the single cell reference image sequence set comprises: From the stem cell samples historically marked as high quality, a single stem cell is selected for continuous tracking and shooting at standard time intervals; the single cell image at each time point is extracted, the corresponding culture condition is marked for each single cell image sequence, and a single cell reference image sequence set is stored.

[0008] Preferably, the screening of the region reference image sequence according to the first stem cell area image and the culture condition comprises: From the region reference image sequence set, a region reference image sequence subset with the same culture condition as the stem cell to be evaluated is screened out; Calculating the regional morphological similarity between the first stem cell region image and the first image of the reference image sequence of each region in the subset; Select the regional reference image sequence with the highest regional morphological similarity as the screening result; The selecting of a single cell reference image sequence corresponding to each tracking target according to the tracking target and the culture conditions comprises: Screening out a subset of single-cell reference image sequences with the same culture conditions as the stem cells to be evaluated from the single-cell reference image sequence set; Calculate the individual cell similarity between each tracking target and the first image of each single-cell reference image sequence in the subset; For each tracking target, the single-cell reference image sequence with the highest individual cell similarity is selected as its corresponding single-cell reference sequence.

[0009] Preferably, the calculation of the regional morphological similarity includes: Extracting morphological parameter features of the two images, wherein the morphological parameter features include area, contour perimeter and roundness; Calculate the corresponding similarity index based on the extracted morphological parameter features; Perform weighted averaging of each similarity index according to preset weights to obtain the regional morphological similarity; The calculation of the individual cell similarity includes: Extract the morphological parameter features, grayscale distribution features and edge features of the two images; Calculate the morphological parameter similarity, grayscale distribution similarity and edge feature similarity respectively; The three similarities are weighted and calculated according to the preset weights to obtain the individual cell similarity.

[0010] Preferably, obtaining the overall similarity based on the stem cell region image sequence and the region reference image sequence includes: Calculate the regional morphological similarity between each image in the stem cell region image sequence and the image at the corresponding time point in the regional reference image sequence; The average of all calculated regional morphological similarities is calculated to obtain the overall similarity.

[0011] Preferably, calculating the tracking rate of each tracking target includes: For each image except the first image in the stem cell region image sequence, all individual cells in the region are extracted using cell outline recognition technology within the block corresponding to each tracking target; Calculate the similarity between each individual cell in the block and the individual cell in the corresponding image in the single-cell reference image sequence; determining whether the maximum cell individual similarity of all cell individuals in the block is greater than a preset threshold value, and if greater, marking the image at the time point as tracked to; calculating the ratio of the number of images marked as tracked to and the total number of images in the sequence of stem cell region images minus one as the tracking rate of the tracking target.

[0012] Preferably, the comprehensive overall similarity and overall tracking rate are used for final evaluation of the quality of the stem cells to be evaluated, including: weighting the overall similarity and the overall tracking rate according to a preset weight to obtain a comprehensive quality score; determining the quality grade of the stem cells to be evaluated according to a pre-established mapping relationship between the comprehensive quality score and the stem cell quality grade.

[0013] The data acquisition and preprocessing module is configured to obtain stem cell culture conditions, acquire images of the stem cells to be evaluated in a culture dish at a predetermined time interval to obtain a predetermined number of acquisition images, extract stem cell region images from the acquisition images using image segmentation technology, and generate a sequence of stem cell region images after standardization processing. The target marking module is configured to divide each stem cell region image in the sequence of stem cell region images into blocks, and mark a stem cell individual in each block in the first stem cell region image as a tracking target. The reference selection module is configured to select a sequence of region reference images from a pre-established stem cell reference library according to the first stem cell region image and the culture conditions, and select a sequence of single cell reference images corresponding to each tracking target according to the tracking target and the culture conditions. The overall similarity calculation module is configured to obtain the overall similarity according to the sequence of stem cell region images and the sequence of region reference images. The overall tracking rate calculation module is configured to calculate the tracking rate of each tracking target according to the block where each tracking target is located and the corresponding sequence of single cell reference images in the sequence of stem cell region images, and use the average of the tracking rates of all tracking targets as the overall tracking rate. The quality evaluation module is configured to comprehensively evaluate the overall similarity and the overall tracking rate to finally evaluate the quality of the stem cells to be evaluated.

[0014] The present application provides a stem cell quality prediction system and method based on multi-modal data, which has the following advantages: By establishing a stem cell reference library containing regional and single-cell reference image sequences, and based on this library, continuously capturing image sequences at standardized time intervals throughout the stem cell culture process, a dynamic analysis framework for stem cell quality assessment was established. Unlike traditional static image analysis that focuses solely on morphological features at a single time point, this method compares the entire developmental process of the stem cell being evaluated with high-quality stem cells in the reference library, comprehensively capturing the morphological evolution and growth patterns of the stem cells throughout the culture process, thereby improving the accuracy and quality of the assessment results. By calculating the overall similarity between regional stem cell images and regional reference image sequences, the morphological development trends of the stem cell population are assessed to ensure they are consistent with high-quality stem cells. By tracking the morphological evolution of individual stem cells and comparing them with single-cell reference image sequences and calculating the overall tracking rate, the quality performance at the single-cell level is assessed. This approach, which focuses on both population characteristics and individual performance, avoids the potential bias of a single indicator and further improves the comprehensiveness and reliability of the assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of a flow chart of a method for predicting stem cell quality based on multimodal data of the present invention; Figure 2 Schematic diagram of a module of a stem cell quality prediction system based on multimodal data of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Example 1 See also Figure 1 In this embodiment, a method for predicting stem cell quality based on multimodal data includes: S1. Obtaining stem cell culture conditions, and capturing images of the stem cells to be evaluated in the culture dish at preset standard time intervals to obtain a predetermined number of captured images, extracting stem cell region images from the captured images using image segmentation technology, and generating a stem cell region image sequence after standardization processing; In this example, the culture conditions of the stem cells to be evaluated are first obtained. These conditions can include culture medium composition, culture temperature, humidity, and other information. This information is recorded and associated with subsequently acquired images. Continuous image acquisition of the stem cells in the culture dish is performed at a preset standard interval, obtaining a predetermined number of captured images. It should be noted that the standard interval is set to one hour in this example to ensure that the actual acquisition frequency matches that of the images in the reference library, ensuring that the images in the reference library have reference value.

[0018] Image acquisition is performed using a phase contrast microscope equipped with an appropriate objective lens. The microscope field of view is kept consistent during the acquisition process to ensure that the same area is tracked and observed. The acquired original image is processed to extract the stem cell region image. Existing image processing techniques including Gaussian filtering, Otsu adaptive threshold segmentation, morphological operations, etc. can be used to separate the stem cell region from the background. The extracted stem cell region image needs to be standardized, which can usually include denoising, contrast enhancement, etc. Through the above processing steps, a standardized stem cell region image sequence is finally generated. The sequence is arranged in chronological order for subsequent quality assessment and analysis.

[0019] S2, dividing each stem cell region image in the stem cell region image sequence into a plurality of blocks, and marking a stem cell individual in each block in the first stem cell region image as a tracking target; The dividing each stem cell region image in the stem cell region image sequence into a plurality of blocks comprises: dividing each image in the stem cell region image sequence into a plurality of blocks by horizontal and vertical dividing lines distributed at equal intervals; The step of marking a stem cell individual as a tracking target in each block in the first stem cell region image includes: Individual stem cells were detected within each block using cell outline recognition technology; Calculate the average edge gradient of each detected stem cell individual; In each block, the stem cell individual with the highest average edge gradient is selected and marked as the tracking target of the corresponding block.

[0020] In this embodiment, after obtaining a standardized stem cell region image sequence, each stem cell region image in the sequence needs to be divided into several blocks, and the tracking target needs to be marked in the first image.

[0021] First, each image in the stem cell region image sequence is divided using equally spaced horizontal and vertical dividing lines. Because stem cell regions are often irregular in shape, the resulting blocks may not be completely consistent in size and shape. However, each block can be uniquely identified using row and column indices (e.g., row 3, column 4).

[0022] In this embodiment, according to the size and distribution density of the stem cell region, the image can be divided into a grid structure of m rows and n columns. The specific values of m and n can be adjusted according to the actual image size and cell density to ensure that each block contains an appropriate number of cells. The purpose of dividing the blocks is to divide the irregular stem cell region into relatively independent small regions to facilitate subsequent cell tracking.

[0023] After completing the block division, a stem cell individual in each block of the first stem cell region image needs to be marked as a tracking target. The marking process first applies cell contour recognition technology to detect stem cell individuals in each block. Existing edge detection algorithms such as Canny algorithm or active contour model can be used for cell contour recognition. For each detected stem cell individual, the average value of its edge gradient is calculated. The edge gradient reflects the clarity of the cell contour. The cell with a higher average value of edge gradient usually indicates a more complete segmentation and a more explicit contour, and is therefore more suitable as a tracking target. After the calculation is completed, the stem cell individual with the highest average value of edge gradient in each block is selected and marked as the tracking target of that block. This selection method ensures that the selected tracking target has relatively complete contour features.

[0024] Through the above steps, multiple tracking targets distributed in different blocks in the first stem cell region image are successfully marked. These targets will be continuously tracked in subsequent image sequences to evaluate the quality of stem cells.

[0025] S3, from a pre-established stem cell reference library, according to the first stem cell region image and the culture conditions, a region reference image sequence is selected, and according to the tracking target and the culture conditions, a single cell reference image sequence corresponding to each tracking target is selected; The stem cell reference library includes a region reference image sequence set and a single cell reference image sequence set; The establishment of the region reference image sequence set includes: From the historical stem cell samples marked as high quality, stem cell region image sequences under different culture conditions are collected at standard time intervals. The collected image sequences are standardized, the corresponding culture conditions are marked for each region reference image sequence, and the region reference image sequence set is stored; The establishment of the single cell reference image sequence set includes: From the historical stem cell samples marked as high quality, a single stem cell is selected for continuous tracking and shooting at standard time intervals. The single cell image at each time point is extracted, the corresponding culture conditions are marked for each single cell image sequence, and the single cell reference image sequence set is stored.

[0026] The region reference image sequence is selected according to the first region image of stem cells and the culture condition, and the region reference image sequence includes: A region reference image sequence subset is selected from the region reference image sequence set, which is the same as the culture condition of the stem cells to be evaluated; A region morphological similarity is calculated between the first region image of stem cells and the first image of each region reference image sequence in the subset; A region reference image sequence with the highest region morphological similarity is selected as a screening result; The single cell reference image sequence corresponding to each tracking target is selected according to the tracking target and the culture condition, and the single cell reference image sequence includes: A single cell reference image sequence subset is selected from the single cell reference image sequence set, which is the same as the culture condition of the stem cells to be evaluated; A cell individual similarity is calculated between each tracking target and the first image of each single cell reference image sequence in the subset; A single cell reference image sequence with the highest cell individual similarity is selected as the corresponding single cell reference sequence for each tracking target.

[0027] The region morphological similarity includes: Morphological parameter features of the two images are extracted, and the morphological parameter features include area, contour perimeter and circularity; A corresponding similarity index is calculated according to the extracted morphological parameter features; The region morphological similarity is obtained by weighted average of each similarity index according to a preset weight; The cell individual similarity includes: Morphological parameter features, gray distribution features and edge features of the two images are extracted; Morphological parameter similarity, gray distribution similarity and edge feature similarity are calculated respectively; The cell individual similarity is obtained by weighted calculation of the three similarities according to a preset weight.

[0028] In this embodiment, after the tracking target is labeled, a suitable reference image sequence is selected from the pre-established stem cell reference library as the benchmark for quality assessment. The stem cell reference library includes two parts: the regional reference image sequence set and the single cell reference image sequence set, which are established by systematically collecting and processing historical data. The establishment process of the regional reference image sequence set is to collect regional images of stem cells under different culture conditions at the same standard time interval (1 hour in this example) from historical stem cell culture samples labeled as high quality. Among them, the stem cell samples labeled as high quality refer to those samples confirmed to have good growth state, strong differentiation potential and high purity by traditional biological evaluation methods (such as flow cytometry, immunohistochemical analysis, gene expression analysis, etc.).

[0029] The collected image sequences are subjected to the same standardization process as the sample to be evaluated, and then each regional reference image sequence is labeled with the corresponding culture condition information, such as medium composition, temperature and other key parameters. Finally, these sequences are stored in the regional reference image sequence set. The establishment of the single cell reference image sequence set is to select a single stem cell for continuous tracking and shooting from historical stem cell samples labeled as high quality at the same standard time interval, which requires the use of cell tracking technology to ensure that the same cell is tracked throughout the observation period. Then extract the single cell image at each time point, label the corresponding culture conditions for each single cell image sequence, and store the single cell reference image sequence set.

[0030] The process of screening regional reference image sequences first selects a subset of regional reference image sequences from the regional reference image sequence set that have the same culture conditions as the stem cells to be evaluated, which ensures the comparability of the reference sequence and the sequence to be evaluated in the culture environment. Then calculate the regional morphological similarity of the first stem cell region image and the first image of each regional reference image sequence in the subset, and select the regional reference image sequence with the highest regional morphological similarity as the screening result.

[0031] The calculation of the region morphology similarity degree includes extracting morphology parameter features of two images, including area, contour perimeter and circularity. The area can be obtained by calculating the number of foreground pixels, the contour perimeter is calculated by the cumulative length of the edge pixels, and the circularity can be calculated by the circularity formula. According to the extracted morphology parameter features, the corresponding similarity indicators are calculated, and the specific calculation method is as follows: first, normalize the area, perimeter and circularity to make their numerical ranges consistent; then calculate the similarity values of the corresponding features of the two images, such as the absolute difference between one and two areas divided by the larger value of the two areas to obtain the area similarity; similarly, the perimeter similarity value and the circularity similarity value are obtained, and then the region morphology similarity degree is obtained by weighted average according to the preset weight (such as area weight 0.4, perimeter weight 0.3, and circularity weight 0.3). The greater the similarity value, the more similar the morphology of the two images.

[0032] Next, for each tracking target, a corresponding single-cell reference image sequence is selected. First, a subset of single-cell reference image sequences that are the same as the stem cell culture conditions to be evaluated is selected from the single-cell reference image sequence set. Then, the cell individual similarity of each tracking target with the first image of each single-cell reference image sequence in the subset is calculated, and the single-cell reference sequence with the highest cell individual similarity is selected as the corresponding single-cell reference sequence for each tracking target.

[0033] The calculation of the cell individual similarity requires extracting the morphology parameter features (area, perimeter, circularity), gray distribution features (average gray value, standard deviation, gray histogram), and edge features (edge gradient intensity, direction distribution) of two images. The calculation method of the morphology parameter similarity is the same as that of the region morphology similarity; the gray distribution similarity can be measured by calculating the correlation coefficient or histogram intersection of the gray histograms of the two images; and the edge feature similarity can be measured by calculating the cosine similarity of the edge gradient direction histogram of the two images. Finally, the morphology parameter similarity, the gray distribution similarity and the edge feature similarity are weighted calculated according to the preset weight (such as morphology parameter weight 0.3, gray distribution weight 0.3, and edge feature weight 0.4) to obtain the cell individual similarity. The higher the similarity value, the more similar the two cell individuals.

[0034] Through the above steps, the region reference image sequence most similar to the stem cell to be evaluated and the single-cell reference image sequence corresponding to each tracking target are successfully selected, which will serve as the benchmark data for subsequent quality evaluation.

[0035] S4, obtaining an overall similarity degree according to the stem cell region image sequence and the region reference image sequence; The obtaining of the overall similarity degree according to the stem cell region image sequence and the region reference image sequence includes: Calculate the regional morphological similarity between each image in the stem cell region image sequence and the image at the corresponding time point in the regional reference image sequence; The average of all calculated regional morphological similarities is calculated to obtain the overall similarity.

[0036] In this embodiment, after the reference image sequence is screened, the overall similarity between the stem cell region image sequence and the regional reference image sequence is calculated. This is a key indicator for assessing stem cell quality. Unlike traditional static image analysis methods, this method dynamically compares the stem cell growth process with the growth patterns of known high-quality stem cells, enabling a more comprehensive assessment of stem cell developmental potential and growth characteristics.

[0037] First, each image in the stem cell region image sequence is compared with the image at the corresponding time point in the regional reference image sequence. Specifically, for each time point, the regional morphological similarity between the image of the stem cell region to be evaluated and the reference image is calculated. This time-series comparative analysis can reflect whether the morphological change trend of the stem cell population throughout the culture process is consistent with that of the high-quality reference sample. Stem cell development is a dynamic process, and high-quality stem cells typically exhibit specific morphological evolution patterns and growth rates. By calculating the regional morphological similarity at each time point, the consistency of the developmental trajectory of the stem cell to be evaluated and the reference stem cell can be quantified.

[0038] Regional morphological similarity was calculated using the same method as previously described. For each time point in the sequence, a regional morphological similarity value was obtained. The regional morphological similarities calculated across all time points were then averaged to obtain the overall similarity. The overall similarity reflects the overall similarity of the evaluated stem cells to the high-quality reference stem cells over the entire observation period.

[0039] The overall similarity reflects the consistency of the developmental pattern and direction of the evaluated stem cell population with those of high-quality stem cells. A higher similarity indicates that the morphological changes, proliferation patterns, and spatial distribution characteristics of the evaluated stem cells are closer to those of known high-quality stem cells, which generally indicates that the evaluated stem cells have better developmental potential and functional properties. By calculating the overall similarity, the quality performance of stem cells during culture can be comprehensively assessed, providing a reliable basis for quality judgment in subsequent cell applications.

[0040] S5. Calculate the tracking rate of each tracking target in the stem cell region image sequence based on the block where each tracking target is located and the corresponding single-cell reference image sequence, and use the average tracking rate of all tracking targets as the overall tracking rate; Calculating the tracking rate of each tracking target includes: For each image except the first image in the stem cell region image sequence, all individual cells in the region are extracted using cell outline recognition technology within the block corresponding to each tracking target; Calculate the similarity between each individual cell in the block and the individual cell in the corresponding image in the single-cell reference image sequence; Determine whether the maximum similarity of all individual cells in the block is greater than a preset threshold. If so, mark the image at that time point as tracked. The ratio of the number of images marked as tracked to the total number of images in the stem cell region image sequence minus one was calculated as the tracking rate of the tracking target.

[0041] In this example, the overall tracking rate reflects the stability of individual stem cells during culture and is an important parameter for evaluating stem cell quality at the single-cell level.

[0042] First, the tracking rate is calculated for each tracking target. For each image in the stem cell region image sequence, except for the first, cell outline recognition technology is used within the block corresponding to each tracking target to extract all individual cells within the area. Using the same cell outline recognition technology as previously described, the individual cells within the block can be effectively identified. Because stem cells undergo morphological changes and growth during culture, the original tracking target may have undergone size changes or morphological changes in images at subsequent time points.

[0043] Next, the similarity between each individual cell in the block and the individual cell image at the corresponding time point in the single-cell reference image sequence is calculated. This step is to determine which individual cell is most likely to be the continuation of the original tracking target and to evaluate whether its morphological changes are consistent with the high-quality reference stem cell. The calculation method is the same as the individual cell similarity calculation method described above, taking into account multiple aspects such as morphological parameter characteristics, grayscale distribution characteristics, and edge characteristics. High-quality stem cells will exhibit specific morphological evolution characteristics during their growth, such as regular changes in cell boundaries and orderly development of internal structures. By comparing the similarity between the stem cell to be evaluated and the reference stem cell at each time point, it can be determined whether the stem cell to be evaluated has a morphological evolution pattern similar to that of the high-quality stem cell.

[0044] Then, the maximum cell-to-cell similarity of all individual cells in the block is determined to be greater than a preset threshold. If so, the image at that time point is marked as tracked. The preset threshold here can be set based on experimental data, expert experience, etc. When the similarity is greater than the threshold, it is considered that the continuation of the original tracking target has been found, and this continuation is consistent with the morphological state of high-quality stem cells. If the similarity of all individual cells in the block is lower than the threshold, it is considered that the tracking target has been lost at that time point, indicating that the stem cell to be evaluated may have morphological characteristics or growth patterns that are different from those of high-quality stem cells, indicating that the stem cell quality is poor.

[0045] Finally, the ratio of the number of images marked as tracked to the total number of images in the stem cell area image sequence minus one is calculated as the tracking rate of the tracking target. The tracking rates of all tracking targets are averaged to obtain the overall tracking rate. The overall tracking rate reflects the degree of consistency between the morphological changes of the stem cells to be evaluated and the high-quality reference stem cells during growth, and is an important indicator for evaluating the quality of stem cells. High-quality stem cells usually have a higher tracking rate, indicating that the cells maintain morphological evolution characteristics similar to those of high-quality reference stem cells during culture. In contrast, low-quality stem cells may exhibit abnormal morphological changes, such as irregular boundary deformation, abnormal internal structure, or inconsistent growth patterns, which increase their morphological differences from high-quality reference stem cells, resulting in a lower tracking rate. By calculating the overall tracking rate, the quality of stem cells can be evaluated from the perspective of single-cell morphological evolution, providing an important basis for the comprehensive evaluation of stem cells.

[0046] S6. Comprehensively evaluate the overall similarity and overall tracking rate to make a final assessment of the quality of the evaluated stem cells.

[0047] The comprehensive overall similarity and overall tracking rate are used to make a final assessment of the quality of the stem cells to be evaluated, including: The overall similarity and overall tracking rate are weighted and calculated according to the preset weights to obtain a comprehensive quality score; The quality grade of the stem cells to be evaluated is determined based on the pre-established mapping relationship between the comprehensive quality score and the stem cell quality grade.

[0048] In this embodiment, the overall similarity and the overall tracking rate are combined to make a final assessment of the quality of the stem cells to be evaluated.

[0049] First, the overall similarity and overall tracking rate are weighted according to preset weights to obtain a comprehensive quality score. The weight setting reflects the relative importance of the two indicators in the quality assessment and can be adjusted according to the actual application requirements. In this embodiment, the overall similarity weight can be set to 0.6 and the overall tracking rate weight can be set to 0.4. This weight distribution not only emphasizes the overall morphological development trend of the stem cell population as a whole, but also takes into account the morphological evolution characteristics of individual stem cells.

[0050] The quality grade of the evaluated stem cells is then determined based on a pre-established mapping between the comprehensive quality score and the stem cell quality grade. This mapping can be established using extensive experimental data and expert evaluation, categorizing the continuous comprehensive quality score into discrete quality grades, such as excellent, good, fair, and unqualified. For example, a comprehensive quality score greater than 0.85 could be designated as excellent, 0.7-0.85 as good, 0.5-0.7 as fair, and less than 0.5 as unqualified. This grading facilitates quality management and decision-making in practical applications.

[0051] This comprehensive assessment method comprehensively evaluates stem cell quality at both the holistic and individual levels. The overall similarity reflects whether the overall morphological changes of the stem cell population during culture are consistent with those of high-quality reference stem cells, a key indicator for evaluating stem cell quality from a macroscopic perspective. The overall tracking rate reflects whether the morphological evolution characteristics of individual stem cells during growth are similar to those of high-quality reference stem cells, a key indicator for evaluating stem cell quality from a microscopic perspective. This combination of factors, which considers both population characteristics and individual performance, enables a more comprehensive and accurate assessment of stem cell quality.

[0052] Example 2 See also Figure 2 The present invention provides a stem cell quality prediction system based on multimodal data, which is used to implement a stem cell quality prediction method based on multimodal data, including: The data acquisition and preprocessing module is used to obtain stem cell culture conditions and collect images of the stem cells to be evaluated in the culture dish at preset standard time intervals to obtain a predetermined number of collected images. The image segmentation technology is used to extract the stem cell region images from the collected images, and after standardization processing, a stem cell region image sequence is generated. a target marking module, configured to divide each stem cell region image in the stem cell region image sequence into a plurality of blocks, and mark a stem cell individual in each block in the first stem cell region image as a tracking target; The reference selection module is configured to select a region reference image sequence from a pre-established stem cell reference library according to the first stem cell region image and the culture condition, and select a single cell reference image sequence corresponding to each tracking target according to the tracking target and the culture condition; The overall similarity calculation module is configured to obtain an overall similarity according to the stem cell region image sequence and the region reference image sequence. The overall tracking rate calculation module is configured to calculate a tracking rate of each tracking target according to a block where each tracking target is located and the corresponding single cell reference image sequence in the stem cell region image sequence, and use a mean value of the tracking rates of all the tracking targets as an overall tracking rate. The quality evaluation module is configured to comprehensively evaluate the overall similarity and the overall tracking rate to finally evaluate the quality of the stem cell to be evaluated.

[0053] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0054] The above description is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be included in the protection scope of the present application.

[0055] Finally, the above description is only the preferred embodiment of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for predicting stem cell quality based on multimodal data, characterized in that: include: Obtaining stem cell culture conditions, and capturing images of the stem cells to be evaluated in the culture dish at preset standard time intervals to obtain a predetermined number of captured images, extracting stem cell region images from the captured images using image segmentation technology, and generating a stem cell region image sequence after standardized processing; Each stem cell region image in the stem cell region image sequence is divided into a number of blocks, and a stem cell individual is marked in each block in the first stem cell region image as a tracking target; From the pre-established stem cell reference library, regional reference image sequences are screened based on the first stem cell regional image and culture conditions. Single-cell reference image sequences corresponding to each tracking target are selected based on the tracking target and culture conditions. Based on the stem cell region image sequence and the regional reference image sequence, the overall similarity is obtained; In the stem cell region image sequence, the tracking rate of each tracking target is calculated based on the block where each tracking target is located and the corresponding single-cell reference image sequence, and the average tracking rate of all tracking targets is used as the overall tracking rate; The overall similarity and overall tracking rate are combined to make a final assessment of the quality of the evaluated stem cells.

2. The method for predicting stem cell quality based on multimodal data according to claim 1, characterized in that: The dividing each stem cell region image in the stem cell region image sequence into a plurality of blocks comprises: dividing each image in the stem cell region image sequence into a plurality of blocks by horizontal and vertical dividing lines distributed at equal intervals; The step of marking a stem cell individual as a tracking target in each block in the first stem cell region image includes: Individual stem cells were detected within each block using cell outline recognition technology; Calculate the average edge gradient of each detected stem cell individual; In each block, the stem cell individual with the highest average edge gradient is selected and marked as the tracking target of the corresponding block.

3. The method for predicting stem cell quality based on multimodal data according to claim 2, characterized in that: The stem cell reference library includes a regional reference image sequence set and a single cell reference image sequence set; The establishment of the regional reference image sequence set includes: From stem cell culture samples that have historically been labeled as high-quality, regional image sequences of stem cells under different culture conditions are collected at standard time intervals; the collected image sequences are standardized, and the corresponding culture conditions are labeled for each regional reference image sequence, and a set of regional reference image sequences is stored; The establishment of the single cell reference image sequence set includes: From stem cell samples historically labeled as high-quality, single stem cells were selected for continuous tracking and imaging at standard time intervals. Single-cell images at each time point were extracted, and the corresponding culture conditions were labeled for each single-cell image sequence, and a set of single-cell reference image sequences was stored.

4. The method for predicting stem cell quality based on multimodal data according to claim 3, characterized in that: The step of screening the regional reference image sequence according to the first stem cell region image and the culture conditions comprises: Screening out a subset of regional reference image sequences having the same culture conditions as the stem cells to be evaluated from the regional reference image sequence set; Calculating the regional morphological similarity between the first stem cell region image and the first image of the reference image sequence of each region in the subset; Select the regional reference image sequence with the highest regional morphological similarity as the screening result; The selecting of a single cell reference image sequence corresponding to each tracking target according to the tracking target and the culture conditions comprises: Screening out a subset of single-cell reference image sequences with the same culture conditions as the stem cells to be evaluated from the single-cell reference image sequence set; Calculate the individual cell similarity between each tracking target and the first image of each single-cell reference image sequence in the subset; For each tracking target, the single-cell reference image sequence with the highest individual cell similarity is selected as its corresponding single-cell reference sequence.

5. The method for predicting stem cell quality based on multimodal data according to claim 4, characterized in that: The calculation of the regional morphological similarity includes: Extracting morphological parameter features of the two images, wherein the morphological parameter features include area, contour perimeter and roundness; Calculate the corresponding similarity index based on the extracted morphological parameter features; Perform weighted averaging of each similarity index according to preset weights to obtain the regional morphological similarity; The calculation of the individual cell similarity includes: Extract the morphological parameter features, grayscale distribution features and edge features of the two images; Calculate the morphological parameter similarity, grayscale distribution similarity and edge feature similarity respectively; The three similarities are weighted and calculated according to the preset weights to obtain the individual cell similarity.

6. The method for predicting stem cell quality based on multimodal data according to claim 5, characterized in that: The obtaining of the overall similarity based on the stem cell region image sequence and the region reference image sequence includes: Calculate the regional morphological similarity between each image in the stem cell region image sequence and the image at the corresponding time point in the regional reference image sequence; The average of all calculated regional morphological similarities is calculated to obtain the overall similarity.

7. The method for predicting stem cell quality based on multimodal data according to claim 6, characterized in that: Calculating the tracking rate of each tracking target includes: For each image except the first image in the stem cell region image sequence, all individual cells in the region are extracted using cell outline recognition technology within the block corresponding to each tracking target; Calculate the similarity between each individual cell in the block and the individual cell in the corresponding image in the single-cell reference image sequence; Determine whether the maximum similarity of all individual cells in the block is greater than a preset threshold. If so, mark the image at that time point as tracked. The ratio of the number of images marked as tracked to the total number of images in the stem cell region image sequence minus one was calculated as the tracking rate of the tracking target.

8. The method for predicting stem cell quality based on multimodal data according to claim 1, characterized in that: The comprehensive overall similarity and overall tracking rate are used to make a final assessment of the quality of the stem cells to be evaluated, including: The overall similarity and overall tracking rate are weighted and calculated according to the preset weights to obtain a comprehensive quality score; The quality grade of the stem cells to be evaluated is determined based on the pre-established mapping relationship between the comprehensive quality score and the stem cell quality grade.

9. A stem cell quality prediction system based on multimodal data, used to implement the stem cell quality prediction method based on multimodal data according to any one of claims 1 to 8, characterized in that: include: The data acquisition and preprocessing module is used to obtain stem cell culture conditions and collect images of the stem cells to be evaluated in the culture dish at preset standard time intervals to obtain a predetermined number of collected images. The image segmentation technology is used to extract the stem cell region images from the collected images, and after standardization processing, a stem cell region image sequence is generated. a target marking module, configured to divide each stem cell region image in the stem cell region image sequence into a plurality of blocks, and mark a stem cell individual in each block in the first stem cell region image as a tracking target; A reference selection module is used to select a regional reference image sequence from a pre-established stem cell reference library based on the first stem cell regional image and culture conditions, and select a single cell reference image sequence corresponding to each tracking target based on the tracking target and culture conditions; An overall similarity calculation module is used to obtain overall similarity based on the stem cell region image sequence and the region reference image sequence; The overall tracking rate calculation module is used to calculate the tracking rate of each tracking target in the stem cell region image sequence based on the block where each tracking target is located and the corresponding single-cell reference image sequence, and use the average tracking rate of all tracking targets as the overall tracking rate; The quality assessment module is used to comprehensively evaluate the overall similarity and overall tracking rate to make a final assessment of the quality of the evaluated stem cells.

Images