Standardized equivalent automatic calculation method and system for iPSC-derived pancreatic islet cells

Abstract

The application relates to the technical field of regenerative medicine, and discloses a standardized equivalent automatic calculation method and system for iPSC-derived pancreatic islet cells, which comprises the following steps: step 1, collecting a microscopic image of an iPSC-derived pancreatic islet cell ball suspension; step 2, inputting the microscopic image into a pre-trained deep learning segmentation model to obtain a binary mask representing a candidate region of the pancreatic islet cell ball; step 3, performing geometric demodulation processing based on parameter space voting on the adherent connected region in the binary mask to separate independent cell ball instances; and step 4, calculating the equivalent diameter of each cell ball based on the separated independent cell ball instances, calculating the pancreatic islet equivalent of each cell ball according to a continuous volume integral formula, and accumulating to obtain the total pancreatic islet equivalent of a field of view. The application can objectively, accurately and repeatedly evaluate the cell dose and transplantation efficiency of the iPSC-derived pancreatic islet cells, and provides a reliable quality control basis for clinical conversion of stem cell treatment products.

Description

Technical Field

[0001] This invention relates to the field of regenerative medicine technology, specifically to a standardized equivalent automatic calculation method and system for iPSC-derived pancreatic islet cells. Background Technology

[0002] Diabetes mellitus is a metabolic disease caused by impaired pancreatic function, leading to insufficient insulin secretion or impaired insulin action. Islet transplantation, as an effective treatment for certain types of diabetes, relies heavily on the accurate dosage of transplanted islets. In islet transplantation, cell clusters (i.e., "cell spheres") are typically used instead of single cells. This is primarily because islet cells naturally exist in a three-dimensional cluster structure in vivo. This structure facilitates intercellular communication, paracrine interactions, and cell polarity, thereby supporting the physiological regulation of insulin secretion. Furthermore, the cell sphere structure enhances post-transplant cell survival, promotes vascularization, and improves metabolic response. Therefore, in clinical transplantation and related research, islets are usually evaluated and used in cell sphere form.

[0003] Since its introduction in 1990, the traditional concept of islet equivalent (IEQ) has become a standardized tool for assessing islet volume and function, especially in human islet transplantation and related research. However, the IEQ method relies on the measurement of islet diameter and the assumption of sphericity, while the actual shape of islets is often irregular, leading to subjective errors and systematic biases in volume estimation.

[0004] With the development of stem cell technology, induced pluripotent stem cells (iPSCs) have developed rapidly in the cell therapy industry due to their advantages such as large-scale expansion and strong immune matching potential, providing a new cell source for the treatment of diabetes.

[0005] However, islet cells differentiated from iPSCs differ significantly from traditionally isolated human islets in morphology, size, structure, and function, making it difficult to directly and accurately apply traditional IEQ assessment systems to these cell products. While existing non-IEQ methods based on ATP content, DNA quantification, or oxygen consumption can partially avoid morphology-dependent problems, they still cannot specifically distinguish between islet cells and non-islet cells, do not fully consider the characteristics of iPSC-derived cells, and still contain systematic errors during sampling and detection.

[0006] Currently, the equivalence assessment of iPSC-derived pancreatic islet cells still largely relies on manual microscopic examination, manual measurement of cell spheroid diameter, and conversion based on Ricordi standards. This manual assessment method is cumbersome, time-consuming, and easily affected by operator experience, cell spheroid adhesion, and size grading boundary effects, leading to subjective results and insufficient repeatability. In contrast, automated assessment methods can complete target identification, overlap decoupling, equivalence conversion, and result output under unified imaging conditions and algorithmic frameworks. This helps reduce human error, improve detection efficiency, and enhance the standardization and comparability of results between different batches of samples.

[0007] However, existing conventional image processing techniques (such as general cell classification or particle size detection algorithms) face two major technical bottlenecks when dealing with pancreatic islet cells derived from iPSCs.

[0008] First, high-density three-dimensional adhesion leads to segmentation failure. Unlike free blood cells or rigid industrial particles, iPSC pancreatic islet cells exhibit extremely high spontaneous aggregation and three-dimensional adhesion during culture and preparation, with cell spheres stacked on top of each other and blurred boundaries. Traditional segmentation methods based on thresholds or standard deep learning models are prone to misclassifying overlapping cell clusters as individual cells, resulting in serious miscalculations of the number and equivalent of cell spheres.

[0009] Secondly, traditional clinical islet equivalent conversion suffers from "discrete truncation error." Existing general image detection patents often only output the physical volume or area of ​​cells, while clinical islet transplantation strictly relies on the unique indicator of "islet equivalent (IEQ)." The traditional Ricordi standard uses a discrete coefficient lookup table method, forcibly classifying continuously distributed diameters into a finite number of size intervals and multiplying them by a fixed conversion factor. This step-like grading results in significant volume truncation loss when crossing size thresholds. Furthermore, current technologies have failed to perform a deep mathematical reconstruction of high-precision image segmentation and clinical IEQ conversion. In particular, the diameter distribution of islet cells derived from iPSCs is extremely wide, and a large number of cell spheroids fall precisely near the grading boundaries. The error generated by the discrete lookup table is drastically amplified, making the "step-like truncation error" rise from a tolerable level to a quality control problem that must be solved.

[0010] Therefore, there is an urgent need in this field to develop a standardized equivalent calculation method specifically for iPSC-derived pancreatic islet cells, in order to more objectively, accurately, and reproducibly assess their cell dosage and transplantation efficacy, and provide a reliable basis for clinical translation. Summary of the Invention

[0011] The present invention aims to provide a standardized equivalent automatic calculation method and system for iPSC-derived pancreatic islet cells, which can significantly reduce human subjective error, shorten detection time and reduce the number of samplings required to achieve the target estimation accuracy, and objectively, accurately and repeatably evaluate the cell dose and transplantation efficacy of iPSC-derived pancreatic islet cells, providing a reliable quality control basis for the clinical translation of stem cell therapy products.

[0012] To achieve the above objectives, the present invention provides the following basic solution.

[0013] Option 1 An automated method for calculating standardized equivalents of iPSC-derived pancreatic islet cells includes the following steps: Step 1: Acquire microscopic images of iPSC-derived pancreatic islet cell suspensions; Step 2: Input the microscopic image into a pre-trained deep learning segmentation model to obtain a binary mask representing the candidate region of pancreatic islet spheres; Step 3: Perform geometric demodulation processing based on parameter space voting on the connected regions in the binarized mask to separate independent cell sphere instances; Step 4: Based on the isolated individual cell sphere instances, calculate the equivalent diameter of each cell sphere, and calculate the islet equivalent of each cell sphere according to the continuous volume integral formula, and sum them up to obtain the total islet equivalent of the field of view.

[0014] Option 2 An automated system for calculating the standardized equivalent of iPSC-derived pancreatic islet cells is used to execute the automated calculation method for standardized equivalent of iPSC-derived pancreatic islet cells as described in Scheme 1; including: An optical signal acquisition module is used to acquire microscopic images of iPSC-derived pancreatic islet cell suspensions; The intelligent feature extraction and segmentation module has a built-in pre-trained deep learning segmentation model, which is used to receive the microscopic image and output a binary mask characterizing the candidate region of pancreatic islet spheres. The overlapping signal geometric demodulation module is used to perform geometric demodulation processing based on parameter space voting on the adhesive connected regions in the binarized mask to separate independent cell sphere instances; The adaptive calculation and error control module includes: The continuous volume integral unit is used to calculate the equivalent diameter of each cell sphere based on the isolated individual cell sphere instances, and to calculate the islet equivalent of each cell sphere using the continuous volume integral formula, and to sum them up to obtain the total islet equivalent of the field of view; The sampling number estimation unit is used to calculate the sample variance and sample mean based on the total pancreatic islet equivalent of multiple pre-sampled visual fields, and to estimate the optimal total number of samplings required to achieve qualified estimation accuracy based on the set confidence threshold and allowable relative error, and to prompt supplementary sampling based on the comparison results between the actual number of samplings and the optimal total number of samplings. The concentration back-calculation unit is used to back-calculate the stock solution concentration based on the average total islet equivalent, single sample volume, and dilution factor after the cumulative sampling number reaches the optimal total sampling number, and outputs clinical solution preparation guidance.

[0015] The working principle and advantages of this invention are as follows: This invention relates to an automated method and system for calculating the standardized equivalent of iPSC-derived pancreatic islet cells. It significantly reduces human error, shortens detection time, and decreases the number of samples required to achieve the target estimation accuracy. It objectively, accurately, and repeatedly assesses the cell dosage and transplantation efficacy of iPSC-derived pancreatic islet cells, providing a reliable quality control basis for the clinical translation of stem cell therapy products. It is suitable for large-scale, standardized quality control applications of iPSC-derived pancreatic islet cell products. The key point is: First, compared with traditional manual microscopy and the Ricordi discrete coefficient lookup table method, this invention fundamentally eliminates step cutoff error through a continuous volume integral formula, preventing abrupt changes in estimated values ​​when the diameter crosses size gradation boundaries, thus restoring the true volume contribution of each cell sphere. Simultaneously, this invention does not simply output cell physical dimensions, but directly performs a lossless quantification conversion between deep learning segmentation results and clinically specific islet equivalent indicators. This achieves extremely high consistency with the coefficient of determination of the manual gold standard, making it interchangeable with manual microscopy in clinical quality control and completely eliminating inconsistencies caused by operator experience, visual fatigue, or subjective judgment.

[0016] Secondly, this solution achieves orders-of-magnitude optimization in both detection efficiency and sampling cost. Traditional manual methods typically require tens of minutes to complete the equivalent assessment of a single batch of samples. Furthermore, because the lack of representativeness must be compensated for by increasing the number of samples, the time cost increases linearly with sampling demand. In high-precision clinical release scenarios, it is often difficult to balance efficiency and reliability. This invention achieves second-level image analysis through a deep learning model, compressing the processing time of a single field of view to an extremely short range. Simultaneously, it innovatively introduces a sampling number estimation unit based on statistical variance. This system can dynamically calculate the minimum total number of samples required to meet a given confidence level based on the variance distribution of the initial pre-sampling and the clinically set tolerance threshold. Because the underlying algorithm of the automated system eliminates subjective fluctuations and discrete lookup table errors, its sample variance is significantly lower than that of traditional manual methods. Therefore, it achieves the same qualification estimation accuracy with fewer samplings, which not only greatly reduces the physical loss of precious cell samples but also makes it possible to upgrade from traditional "sampling assessment" to "high-throughput full detection."

[0017] Third, this solution possesses robust instance separation capabilities in highly adhesive scenarios and can provide end-to-end clinical decision support. Addressing the high-density, three-dimensional spontaneous aggregation characteristics of iPSC-derived pancreatic islet cells during culture, this invention constructs a two-stage architecture combining deep learning semantic segmentation and geometric parameter spatial demodulation: the front-end deep learning model filters out background noise and generates an initial binary mask, while the back-end geometric demodulation process separates independent cell spheroid instances, thereby achieving precise decomposition of complex overlapping regions. This design overcomes the inherent defects of traditional watershed algorithms, which are prone to oversegmentation or undersegmentation when dealing with extremely adhesive targets. Simultaneously, this system is not limited to outputting simple counts or equivalent values, but rather establishes a closed-loop data chain from microscopic image acquisition to clinically matched formulations—automatically back-calculating the biopotency concentration of the original solution based on precise sampling volume, and directly calculating the recommended extraction volume according to the clinical target dose, outputting a structured quality control report, which helps improve the standardized management level and clinical translation efficiency of stem cell therapy products. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the method flow of the first embodiment of the present invention for the automatic calculation method and system for standardized equivalents of iPSC-derived pancreatic islet cells; Figure 2 This is an example of the original grayscale image of iPSC islet cell spheres acquired under an inverted microscope after sampling, in the first embodiment of the automatic calculation method and system for standardized equivalents of iPSC-derived islet cells of the present invention. Figure 3 This is the mask image processed in step 2 in Embodiment 1 of the Automatic Calculation Method and System for Standardized Equivalents of iPSC-Derived Pancreatic Islet Cells of the Present Invention; Figure 4 This is a schematic diagram of the geometric demodulation process in Embodiment 1 of the automatic calculation method and system for standardized equivalents of iPSC-derived pancreatic islet cells of the present invention. Figure 5 This is a schematic diagram illustrating the effect of geometric demodulation processing in Embodiment 1 of the standardized equivalent automatic calculation method and system for iPSC-derived pancreatic islet cells of the present invention. Figure 6 This is a system architecture diagram of the first embodiment of the present invention for the automatic calculation method and system for standardized equivalents of iPSC-derived pancreatic islet cells; Figure 7 This is a statistical verification result graph of the present invention and the manual standard in Experiment Example 1 of the first embodiment of the automatic calculation method and system for standardized equivalent of iPSC-derived pancreatic islet cells. Figure 8 This is a comparison chart of the sampling convergence characteristics of the present invention and the manual method in Experiment Example 2 of Embodiment 1 of the present invention for the automatic calculation method and system for standardized equivalents of iPSC-derived pancreatic islet cells. Detailed Implementation

[0019] The following detailed explanation illustrates the specific implementation methods: Example 1 The basic implementation examples are as follows: Figure 1 The following steps are shown: An automated method for calculating the standardized equivalent of iPSC-derived pancreatic islet cells. Sample physical preparation steps: A precise volume of cell suspension sample is aspirated from the cell culture vessel and selectively diluted according to the estimated concentration; specifically, a precise volume is aspirated from a well-mixed iPSC-derived pancreatic islet cell suspension (total volume 30 mL) in the cell culture vessel using a calibrated pipette. The cell suspension was used as the initial sampling sample. Considering that the high-density clusters of the original solution could lead to insufficient sample representativeness (a major source of sampling error), the system allows the introduction of a dilution factor k based on the estimated concentration. In this embodiment, the dilution factor k = 1 (i.e., no additional dilution is performed).

[0020] The sample is injected into a flat-bottomed ultra-low adsorption plate, ensuring that the bottom of the plate is flat and free of air bubbles to avoid optical distortion caused by air bubbles. The plate is then placed on the stage of an inverted microscope, the transmission light source is turned on, and the light source intensity is adjusted.

[0021] Based on the density and size distribution of cell spheroids in the sample, objectives with magnifications of 1.25x, 2.5x, 4x, or 10x are selected for focusing and image acquisition. The introduction of ultra-low magnification (such as 1.25x and 2.5x) allows a single imaging session to cover a very large macroscopic field of view. This not only significantly reduces stitching errors and mechanical movement time during high-throughput scanning but also aligns well with the subsequent multi-frequency automatic sampling algorithm (step 5), enabling industrial-grade high-throughput full inspection to be achieved with extreme efficiency.

[0022] In this embodiment, an objective lens with a magnification of 4x was selected based on the estimated diameter distribution range of the cell spheres in this batch. This magnification can cover a field of view of approximately 2 mm × 1.5 mm in a single imaging session, which can balance the resolvability of individual cell spheres with statistical throughput.

[0023] Step 1: Acquire microscopic images of iPSC-derived pancreatic islet cell suspensions.

[0024] The operator lowers the microscope objective lens to near the bottom of the well plate using the coarse adjustment knob, and then slowly adjusts the fine adjustment knob until the edges of the cell spheres are clear and the internal structure is visible in the image. The CMOS camera connected to the computer displays the image in real time on the software. After confirming that the focus is correct, the scale is matched with the magnification (in this embodiment, the physical size of each pixel under the 4x objective lens is 1.35 μm).

[0025] Click the acquire button to save the image as a TIFF file to the specified folder. Repeat the above sampling and imaging steps, performing a total of 3 initial samplings (m=3) to obtain 3 original microscopic images with independent fields of view.

[0026] like Figure 2 As shown, in typical raw microscopic images, cell spheres appear as dark, round or near-round areas of varying sizes against a bright, uniform background of culture medium. Because iPSC-derived pancreatic islet cells tend to spontaneously aggregate during culture, the images show complex patterns of multiple cell spheres adhering to or stacking together. Some cell clusters exhibit a "grape bunch" structure with blurred edges and uneven internal grayscale.

[0027] Step 2: Input the microscopic image into a pre-trained deep learning segmentation model to obtain a binary mask representing the candidate region of pancreatic islet spheroids.

[0028] The deep learning segmentation model is a semantic segmentation network based on convolutional neural networks, visual transformers, or a combination of both. It can also be other deep learning segmentation models capable of recognizing cell spheroid regions and outputting masks. Furthermore, the deep learning segmentation model can be trained using a training dataset consisting of microscopic images of pancreatic islet cells from iPSCs and corresponding manually labeled masks. The training dataset is preferably composed of sample images from different differentiation batches, different cell spheroid size ranges, different degrees of adhesion, and different imaging conditions.

[0029] During training, the pancreatic islet cell spheroids can be labeled as the foreground, while the culture medium background, well plate noise, air bubbles, cell debris, and other non-target areas can be labeled as the background. To improve the model's adaptability to different sample states and imaging conditions, data augmentation techniques such as random rotation, scaling, brightness adjustment, contrast adjustment, blurring, and noise perturbation can be used during training.

[0030] Model training can be conducted using supervised training methods. The loss function can be cross-entropy loss, Dice loss, FocalLoss, or a combination thereof. The optimizer can be Adam, SGD, or other optimization algorithms. The training process can continue until the validation set loss converges, or until the preset number of training rounds is reached.

[0031] After training, the microscopic image to be analyzed is input into the deep learning segmentation model, and the model outputs a probability map or initial mask corresponding to the input image. The system then generates a binary mask based on a preset threshold or the model output result, which is used for subsequent target recognition, geometric completion, instance separation and IEQ calculation.

[0032] The acquired microscopic images (such as...) Figure 2 As shown, the input is fed into the trained model above. The model extracts the texture, edge and morphological features of the cell spheres through multi-layer convolution and pooling operations, and finally outputs a probability map of the same size as the input image, where each pixel value represents the probability that the position belongs to a cell sphere.

[0033] Using 0.5 as the classification threshold, the probability map is converted into a binary mask: pixels with a probability ≥ 0.5 are marked as white (foreground, representing candidate regions for cell spheres), and pixels with a probability < 0.5 are marked as black (background).

[0034] like Figure 3 As shown, in the processed mask image, background noise has been effectively filtered out, and the cell spheroid regions appear as white connected regions. For separated cell spheroids, the mask can clearly distinguish them; for adhered cell clusters, the mask still shows them as single connected regions, and these regions will proceed to the next step for decoupling.

[0035] In this step, the powerful data-driven feature extraction capabilities of deep learning models are leveraged to overcome the shortcomings of traditional morphological algorithms (such as thresholding and edge detection) that are sensitive to changes in illumination and struggle to capture weak signals. Through large-sample training, the deep learning model in this approach can establish a non-linear mapping from pixel grayscale to biological semantics. Whether the target is an independent microcell (<100μm) or a complex aggregate, it can be accurately identified through the learned feature representation. Combined with subsequent geometric demodulation strategies, the accuracy of quantitative calculations across the entire size range is significantly improved while maintaining morphological realism.

[0036] Step 3: Perform geometric demodulation processing based on parameter space voting on the connected regions in the binarized mask to separate independent cell sphere instances.

[0037] In this embodiment, the binary mask output by the deep learning segmentation model is not used to directly send all connected regions to geometric reconstruction. Instead, the target region is first distinguished into normal cell spheres, lateral adhesions, and occlusion defects.

[0038] Specifically, after the deep learning segmentation model outputs a binary mask, an adaptive decision-making strategy based on area ratio is used to determine whether there is adhesion, stacking, or occlusion in each connected region, and based on this, decides whether to trigger geometric demodulation processing. The specific decision-making process is as follows: For each connected region, its edge contour information is extracted. In this embodiment, the Canny edge detection algorithm is preferably used to obtain the contour information. Then, the edge contour points are mapped to a three-dimensional parameter space (circle center x-coordinate a, circle center y-coordinate b, radius r) using the circular Hough transform, and parameter accumulation voting is performed to construct a parameter space accumulator array.

[0039] The center and radius of potential cell spheres are determined by detecting local maxima in the accumulator array. A fitted circle is generated based on the detected center and radius, and concentric circles or repeated detection results can be eliminated by incorporating IoU constraints and non-maximum suppression NMS.

[0040] For each connected region, let the target mask area output by the deep learning segmentation model be... The minimum circumcircle radius corresponding to the target is Then the area of ​​the smallest circumcircle is And calculate the area ratio. This ratio is used in the first stage of the determination: to determine whether there are any adhesions or occlusions in the current connected region that require intervention.

[0041] when When the area ratio is greater than or equal to the preset threshold, the target is determined to be a normal cell sphere or a laterally adhered target. The original segmentation mask is retained for subsequent calculation of equivalent diameter and islet equivalent. At this time, geometric demodulation processing is not triggered.

[0042] when When the area ratio is less than the preset threshold, the target is determined to have occlusion, depression or abnormal defects, triggering the geometric completion (demodulation) procedure, and subsequent instance reconstruction and islet equivalent calculation are performed based on the completed geometric parameters.

[0043] In a preferred embodiment, the preset area ratio threshold can be set to 0.65. That is, when When, retain the original mask; when When this occurs, geometric completion is triggered. The aforementioned thresholds can be adjusted based on imaging magnification, pixel scale, cell spheroid size distribution, and sample condition.

[0044] The geometric demodulation process based on parameter space voting includes the following sub-steps: Extract the edge contour information of the connected regions; obtain a set of edge pixel coordinates.

[0045] The circular Hough transform is used to map the edge contour points to a three-dimensional parameter space, namely the x-coordinate of the center, the y-coordinate of the center, and the radius. The system then performs cumulative voting. Specifically, for each edge point, the system searches within a preset radius range. Perform a circle parameter search within the circle. and It can be set according to the imaging magnification, pixel scale, cell sphere size distribution in the same batch, historical calibration data, or the minimum circumscribed circle radius of the current target connected region.

[0046] In a preferred embodiment, the minimum search radius The maximum search radius can be set to 20 pixels. When the minimum circumscribed circle radius of the target is less than or equal to 20 pixels, it can be determined that the target size is too small or the geometric fitting reliability is insufficient, and geometric completion is skipped. The radius can be set according to the maximum expected radius of the cell spheres in the sample, for example, it can be set to 150–200 pixels, or it can be set to a preset multiple of the minimum circumcircle radius of the current connected region.

[0047] For example, under imaging conditions of 4x objective lens and a pixel scale of approximately 1.35 μm / pixel, 20 pixels correspond to a physical radius of approximately 27 μm; 150–200 pixels correspond to a physical radius of approximately 202.5–270 μm. These radius ranges are merely example values ​​and can be adjusted according to different imaging magnifications, pixel scales, iPSC-derived pancreatic islet cell spheroid size distributions, and sample densities.

[0048] Local maxima in the parameter space are detected to determine the true center coordinates and fitted radius of the occluded cell sphere. Specifically, the parameter space accumulator is traversed to detect local maxima. A peak threshold is set (in this embodiment, it is set to 0.7 times the maximum value of the accumulator), and peaks exceeding this threshold are identified as candidate cell sphere instances. Each peak corresponds to the true center coordinates and radius of an occluded cell sphere.

[0049] like Figure 4 and Figure 5 As shown, the boundary contours of independent cell spheres are reconstructed based on the detected center coordinates and fitting radius, and the originally connected regions are decomposed into multiple independent cell sphere instances, thus completing the instance separation of the adhered clusters.

[0050] For example: Figure 4 In step 1, there is a cluster containing two adherent cell spheres. After the circular Hough transform, two significant peaks (a1,b1,r1) and (a2,b2,r2) are detected in the parameter space, which correspond to the physical centers of the two cell spheres, respectively. Based on this, the system segments the original mask to generate two independent sub-regions, corresponding to two independent cell sphere instances.

[0051] Preferably, to avoid erroneous decoupling or excessive geometric completion, after completing the parameter space energy peak detection based on the circular Hough transform, a reliability verification condition based on the ratio of the mask area to the fitted geometric shape area is used to verify and screen the candidate decoupling results. The specific judgment process is as follows: First, calculate the target mask area of ​​the currently unverified connected region output by the deep learning segmentation model. The area ratio between the fitted geometric shape and the area of ​​the target. Let the minimum circumcircle radius of the target be... Then the area of ​​the smallest circumcircle is The area ratio is: .

[0052] This ratio reflects the degree of fit between the segmentation mask and the ideal circular contour: a ratio close to 1 indicates that the mask shape is full and without significant defects; a ratio significantly less than 1 suggests the presence of occlusion, depression, or abnormal defects. Furthermore, this ratio is used in the second stage to verify the geometric reliability of the demodulation results.

[0053] Specifically, when When the area ratio is greater than or equal to a preset threshold, the target is determined to be a normal cell sphere or a laterally adhered target, the original segmentation mask is retained (i.e., the instance is retained), and no additional geometric completion is performed; when When the area ratio is less than a preset threshold, the target is determined to have occlusion, depression or abnormal defects, triggering a geometric completion procedure (e.g., re-execute edge extraction and parameter space mapping, or use more refined parameters for secondary decoupling) to reconstruct the complete outline of the cell sphere.

[0054] In a preferred embodiment, the preset area ratio threshold can be set to 0.65. That is, when When, geometric completion is triggered; when At that time, the original mask is preserved.

[0055] Furthermore, to avoid false completion due to small targets or unreliable fitting, the following auxiliary screening conditions are set (all thresholds are preset parameters; this embodiment is for illustrative purposes only): (1) When the minimum circumscribed circle radius of the target If the radius is less than or equal to a preset radius threshold (e.g., 20 pixels), the area is determined to be noise or fragment and is skipped directly without any decoupling or completion operation.

[0056] (2) Reliability screening is performed by combining the number of convex hull points, the length of the minor axis of the ellipse fitting, and the ratio of the major axis to the minor axis: Ellipse fitting is performed on the candidate region. The major axis and minor axis of the fitted ellipse are required to be greater than the preset lower limit (in this example, the minor axis is >40 pixels), and the ratio of the major axis to the minor axis is <3.0 (i.e., the shape is not excessively narrow). If the ellipse fitting result does not meet the above conditions or the number of convex hull points is insufficient, the region is considered to be non-cellular spherical impurities or severely distorted cell debris, and the completion is also skipped.

[0057] The above thresholds (including the ratio threshold of 0.65, the lower limit of radius of 20 pixels, the lower limit of minor axis of 40 pixels, and the upper limit of major-minor axis ratio of 3.0) are all preferred examples given in this embodiment. In actual operation, they can be preset or dynamically adjusted according to the magnification of the imaging system, the pixel scale, the typical size distribution of cell spheres, and the specific state of the sample (such as the number of days of culture, differentiation efficiency, etc.) to achieve a balance between sensitivity and specificity.

[0058] Step 4: Based on the isolated individual cell sphere instances, calculate the equivalent diameter of each cell sphere, and calculate the islet equivalent of each cell sphere according to the continuous volume integral formula, and sum them up to obtain the total islet equivalent of the field of view.

[0059] The equivalent diameter is calculated as follows: the physical area Area is obtained by converting the number of pixels occupied by the independent cell sphere instance, and then the equivalent circle diameter d is calculated from the physical area. .

[0060] The formula for continuous volume integral is: Where d is in micrometers, and the denominator 150 represents the reference diameter of the standard islet equivalent. Total islet equivalent is the islet equivalent of each individual cell sphere. The sum of the values. Unlike the traditional Ricordi lookup method (which classifies the diameter into multiple discrete intervals and multiplies them by a fixed coefficient), this formula can be calculated continuously for any diameter, eliminating the jump error at the interval boundaries.

[0061] The total pancreatic islet equivalent in the current field of view is obtained by traversing all independent cell sphere instances and summing them up, as shown in the following formula: ; Where N is the total number of independent cell sphere instances in this field of view.

[0062] After the above calculations, the IEQ values ​​of the three fields of view sampled in the first instance of this embodiment are 190, 200, and 210 respectively (unit: IEQ / field of view).

[0063] Step 5: Perform sampling quality control. Multiple pre-sampling operations were performed using an initial dilution factor k, with each sampling volume being [missing data]. Steps 1 to 4 are performed on the microscopic images sampled each time to calculate the total pancreatic islet equivalent in the field of view of each sample. The sample variance was calculated based on the results of multiple pre-sampling operations. and sample mean In this embodiment, Where m=3, Let I be the total IEQ of the field of view in the j-th sample. Substituting the value from step 4, we get... .

[0064] Substitute The values ​​obtained in step 4 are... .

[0065] Based on the set confidence threshold Z and the allowable relative error Calculate the absolute error limit and using the formula Estimate the optimal total number of samples n required to achieve acceptable estimation accuracy.

[0066] In this embodiment, the confidence threshold Z corresponds to 1.96 at a 95% confidence level, and the allowable relative error... The threshold is set to be less than or equal to 5%. This 5% threshold is based on the following: (i) Clinical safety boundary: The total dose of islet transplantation usually needs to be greater than 5000 IEQ / kg, and a 5% error corresponds to 250 IEQ / kg, which is acceptable within the safety window; (ii) The International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) guidelines for the validation of analytical methods recommend a relative error of ≤5% for the quantitative detection of live cell biological products; (iii) Physical quantification limit: When the mean field of view is about 200 IEQ, an absolute error of 10 IEQ is approximately equivalent to the volume of a large cell sphere with a diameter of about 320 μm, representing monomer-level control precision.

[0067] Substitute the values, ; Since the number of samples must be an integer, and to ensure a conservative estimate, we round up to obtain n = 4 times.

[0068] If the actual number of samplings is less than n, a prompt will be made to supplement the sampling until the cumulative number of samplings reaches n. Then, the average total islet equivalent and sampling volume of the multiple samplings will be used as the basis for the calculation. Dilution factor k can be used to infer the concentration of the original solution. .

[0069] In this embodiment, three samples have been completed, but one more is needed to reach the optimal number of samples n=4. The following prompt message is generated: "Three samples have been completed. One more sample needs to be performed to meet the accuracy requirement of 5% relative error at 95% confidence level." The operator completes the fourth independent sampling and imaging according to the aforementioned sample physical preparation steps and the method described in step 1. Steps 1 to 4 are repeated for the image of this field of view, and the total IEQ of the fourth field of view is calculated to be 200 IEQ.

[0070] The average value was recalculated based on the cumulative four sampling results, and the updated value was... Then, the concentration y of the original solution (unit: IEQ / mL) was calculated backwards: IEQ / mL.

[0071] By implementing this step, this solution significantly reduces sampling errors at the physical source, achieving high-precision convergence with less sampling cost. Addressing the common pain points of insufficient representativeness in high-density cell suspension sampling and the lack of a quantitative standard for sample number in the industry, this solution innovatively combines dynamic dilution with an optimal sampling number estimation model based on statistical variance. It dynamically calculates the minimum total number of samples required to meet a given confidence level based on the variance distribution of the initial pre-sampling and the clinically set tolerance threshold. Because the underlying algorithm of the automated system eliminates subjective fluctuations and discrete lookup table errors, its sample variance is significantly lower than that of traditional manual methods. Therefore, it achieves the same qualified estimation accuracy with fewer samplings, greatly reducing physical sampling costs and the loss of precious cells.

[0072] Step 6, guide clinical solution preparation: Based on clinical target dose Based on the original solution concentration y obtained by reverse calculation, the required volume of formulation to be extracted is automatically calculated. It also outputs a quality control report containing the concentration of the original solution and the extraction volume.

[0073] For example, if the clinically required transplant dose is 5000 IEQ / kg, then the total required dose is... The volume of the formulation to be extracted. .

[0074] Finally, the quality control report contains the following: Macroscopic concentration of stock solution: 2000 IEQ / mL; Minimum number of samples required to achieve 95% confidence level and 5% relative error: 4. Clinically recommended volume of solution to be prepared: 175 mL; Detailed list of IEQ for each field of view (190, 200, 210, 200).

[0075] like Figure 6 As shown, this embodiment also provides an automatic calculation system for standardized equivalents of iPSC-derived pancreatic islet cells, used to execute the automatic calculation method for standardized equivalents of iPSC-derived pancreatic islet cells as described above; including: An optical signal acquisition module is used to acquire microscopic images of iPSC-derived pancreatic islet cell suspensions; The intelligent feature extraction and segmentation module has a built-in pre-trained deep learning segmentation model, which is used to receive the microscopic image and output a binary mask characterizing the candidate region of pancreatic islet spheres. The overlapping signal geometric demodulation module is used to perform geometric demodulation processing based on parameter space voting on the adhesive connected regions in the binarized mask to separate independent cell sphere instances; The adaptive calculation and error control module includes: The continuous volume integral unit is used to calculate the equivalent diameter of each cell sphere based on the isolated individual cell sphere instances, and to calculate the islet equivalent of each cell sphere using the continuous volume integral formula, and to sum them up to obtain the total islet equivalent of the field of view; The sampling number estimation unit is used to calculate the sample variance and sample mean based on the total pancreatic islet equivalent of multiple pre-sampled visual fields, and to estimate the optimal total number of samplings required to achieve qualified estimation accuracy based on the set confidence threshold and allowable relative error, and to prompt supplementary sampling based on the comparison results between the actual number of samplings and the optimal total number of samplings. The concentration back-calculation unit is used to back-calculate the stock solution concentration based on the average total islet equivalent, single sample volume, and dilution factor after the cumulative sampling number reaches the optimal total sampling number, and outputs clinical solution preparation guidance.

[0076] This embodiment provides a standardized equivalent automatic calculation method and system for iPSC-derived islet cells. While maintaining a high degree of consistency with human evaluation results, it achieves automated output of target identification, adhesion decoupling, islet equivalent calculation, and sampling decision. It can significantly reduce human subjective error, shorten detection time, and reduce the number of samplings required to achieve target estimation accuracy. It is suitable for large-scale and standardized quality control applications of iPSC-derived islet cell products.

[0077] In particular, this solution breaks through the segmentation limits of highly adhesive biological samples and eliminates clinical-grade quantification losses: Compared with existing general particle detection or conventional blood cell morphology systems, this invention specifically targets the highly adhesive and stacked physical characteristics of iPSC pancreatic islet cell populations, solving the problem of individual cell disassembly of complex clusters through an original energy clustering demodulation method. More importantly, this invention does not stop at the level of conventional cell counting or physical size output, but innovatively applies a continuous mathematical integral model to the conversion of clinically specific IEQ standards, fundamentally eliminating the discrete truncation error inherent in the Ricordi lookup table method used in the industry. It achieves high-precision, non-destructive conversion of microscopic features to macroscopic clinical dosage indicators, filling the technical gap in the automated quality control of stem cell preparation lines where it is difficult to balance the segmentation of extremely adhesive samples with accurate reconstruction of clinical equivalents.

[0078] Secondly, through the coordinated operation of steps 4 and 5, this scheme forms a mutually reinforcing positive technical closed loop. Specifically, by replacing the traditional discrete lookup table method with the continuous volume integral formula, the step cutoff error is fundamentally eliminated, significantly reducing the fluctuation of IEQ measurements between different fields of view within the same batch and minimizing sample variance. This is far lower than manual methods. Because the optimal number of sampling times n and the variance... Proportional to this, lower variance directly means fewer samples are needed to achieve the same clinical confidence level. Reducing the number of samples lowers physical sampling costs and the loss of precious cell samples, while also enabling high-throughput full-scale testing. Simultaneously, fewer samples further reduce operational variability introduced by multiple sampling, which in turn helps maintain stable performance with low variance. Therefore, this approach can create a virtuous cycle: higher algorithm accuracy → smaller measurement variance → fewer required samples → higher detection efficiency → lower cost. This provides a quantifiable and optimizable technical path for the standardized quality control of stem cell preparations from a mathematical perspective.

[0079] To verify the superiority of this scheme, several experimental examples are provided below.

[0080] Experimental Example 1 This example aims to verify the quantification accuracy, clinical interchangeability, and sampling convergence advantages of the method described in Example 1.

[0081] (1) Construction of the validation set Two independently differentiated batches of iPSC-derived pancreatic islet cell suspensions (batch A and batch B) were selected.

[0082] Each batch was sampled and measured independently by three trained professional technicians. Each technician randomly selected nine microscopic fields of view and measured the maximum diameter of each pancreatic islet cell sphere in the field of view in situ (without image acquisition) using a micrometer on the microscope eyepiece. Simultaneously, high-resolution microscopic images of the field of view were acquired using the same imaging system as in Example 1. A total of 54 strictly paired pairs of "manual measurement value - microscopic image" data were obtained (2 batches × 3 people × 9 fields of view = 54 pairs).

[0083] (2) Establishment of artificial gold standard (control group) For each set of data, based on the diameter of each cell spheroid obtained through manual measurement, and according to the islet equivalent discrete conversion standard shown in Table 1, the diameter of each cell spheroid is assigned to the corresponding size range, and converted to the standard islet equivalent according to the corresponding islet equivalent conversion index. Finally, the total IEQ for that field of view is accumulated. The islet equivalent conversion index in Table 1 is compiled with reference to the "Technical Operation Specifications for Clinical Islet Preparation in China (2023 Edition)" and the traditional IEQ standard proposed by Ricordi et al., where a spherical islet with a diameter of 150 μm is used as 1 IEQ. The mathematical principle is: with 150 μm as the standard equivalent (IEQ=1), the volume is proportional to the cube of the diameter. However, because traditional manual table lookup uses the representative diameter of the range (e.g., the representative diameter of 151-200 μm is 175 μm), there is a deviation from the actual continuous diameter. This embodiment strictly follows the conversion according to the table.

[0084] Table 1. Islet equivalent conversion index corresponding to islet cell spheres of different diameters.

[0085] (3) Automated calculation of this system (experimental group) Fifty-four simultaneously acquired microscopic images were input into this system, and the system performed calculations according to the method described in Example 1 (deep learning segmentation → geometric demodulation → continuous volume integral formula), outputting the automated IEQ total value for each image. The continuous volume integral formula directly adopts... No interval merging is performed on the diameter.

[0086] Table 2 shows the quantitative results output by this method after automated calculation of multiple fields of view and multiple batches of samples, including total pancreatic islet equivalent and cell spheroids.

[0087] Table 2 Examples of Quantification Results

[0088] (4) Linear regression and consistency analysis Scatter plots were created using "Automated IEQ" as the ordinate and "Artificial Gold Standard IEQ" as the abscissa for the 54 paired data sets, and linear regression analysis was performed.

[0089] like Figure 7 As shown, the regression equation is: Coefficient of determination The results indicate a very high linear correlation between the automated calculation results of this system and the artificial gold standard, and the overall quantitative trend of this system almost perfectly matches the traditional clinical standards.

[0090] The Bland-Altman method was used for consistency analysis, calculating the difference between each pair of data points (automated IEQ - manual IEQ) and their average. The results showed a mean deviation of +4.53 IEQ (95% confidence interval: +2.17 to +6.89 IEQ), with the vast majority of data points (52 / 54, 96.3%) falling within the 95% consensus limit (mean ± 1.96 × standard deviation). This positive deviation is not a systematic error, but rather a result of the continuous volume integral formula used in this system accurately reproducing the true volume of each cell spheroid. Traditional manual lookup methods, however, produce a "step-cutoff loss" of volume when the diameter crosses interval boundaries (e.g., when the diameter changes from 149 μm to 151 μm, the lookup coefficient jumps from 0.648 to 1.685). The continuous integration of this system compensates for this loss, achieving a lossless quantification conversion from microscopic image pixels to macroscopic clinically standardized equivalents. Therefore, the deviation is systematic, stable, and physiologically reasonable.

[0091] Experiment Example 2 This example aims to compare sampling convergence characteristics.

[0092] This embodiment further verifies the advantages of the system in reducing the number of sampling times. Two batches of sample data, identical to those in Experiment 1 (27 fields of view per batch, totaling 54 data sets, but analyzed independently by batch), were used.

[0093] Two methods were used to calculate the IEQ for each field of view: (1) Manual method (Ricordi discrete lookup table method). Technicians measure the diameter of each pancreatic islet cell in the microscopic image one by one, look up the conversion factor according to the standard size range shown in Table 1, and sum them up to obtain the total IEQ of the field of view.

[0094] (2) Automated method (the continuous volume integration method of Embodiment 1 of this invention). For each batch, the Bootstrap resampling method (1000 random permutations) is used to simulate the estimation error under different cumulative sampling numbers (n=1,2,…,27). The relative estimation error (RE) is calculated according to the following formula: ; in, Let be the mean of the accumulated samples from n fields of view; This represents the overall mean of all 27 fields of view in this batch. Record the minimum number of samples required for each resampling curve to first fall below the 5% error threshold. CV is the coefficient of variation (standard deviation / mean × 100%).

[0095] like Figure 8 As shown ( Figure 8 In Table 3, A and C represent the error convergence curves for the two batches, and B and D represent the 95% confidence intervals of the cumulative mean. The error reduction rate of the automated method (green line) is significantly faster than that of the manual method (red line). Specific values ​​are summarized in Table 3. Table 3 Summary of the convergence results of the two batches of sampling experiments

[0096] The results show that the automated method of the present invention reduces the number of samples required to achieve clinically acceptable accuracy (RE≤5%) in two independent batches by approximately 28.6% (from 7 to 5) and 40% (from 5 to 3) compared to the manual method, respectively, demonstrating that it achieves more efficient and economical batch concentration estimation by reducing inter-sample variance.

[0097] In summary, this approach achieves quantitative accuracy and systematic optimization highly consistent with manual standards (with clinical interchangeability): the IEQ results calculated by this method show extremely high overall consistency with the manual gold standard measured independently by multiple experts in a double-blind manner. Statistical validation based on 54 rigorously paired samples shows that the coefficient of determination between the automated calculation results of this system and expert manual measurements reaches 0.9951, which confirms that this system has the reliability and interchangeability to replace routine manual microscopy in clinical quality control in terms of macroscopic trends.

[0098] Furthermore, Bland-Altman consistency analysis revealed a mean bias of +4.53 IEQ. Analysis showed that this small and stable positive bias primarily stemmed from the system's use of a smooth, continuous volume function integral, which objectively compensated for volume truncation losses caused by crossing size thresholds during traditional manual discrete grading and lookup processes. Therefore, this approach not only effectively avoids subjective visual errors from human operation but also refines the traditional discrete standards through a continuous mathematical model, further improving the quantitative resolution of islet equivalents while ensuring the continuity of clinical standards.

[0099] Example 2 An automated method for calculating the standardized equivalent of iPSC-derived pancreatic islet cells, based on Example 1, further includes step 7, which performs error control on the total islet equivalent in the field of view: The coefficient of variation (CV) is calculated based on the islet equivalents from multiple independent fields of view within the same batch. When the CV is less than a preset threshold, the samples in this batch are considered to have good homogeneity, the calculation results are reliable, and a report can be directly output. When the CV exceeds the preset threshold, a prompt is issued to increase the number of samplings or check the sample homogeneity.

[0100] The coefficient of variation reflects the relative dispersion among samples, and is calculated using the following formula: ; in, The sample standard deviation of IEQ values ​​for each visual field; This is the sample mean.

[0101] In this embodiment, the preset CV threshold can be set according to quality control requirements, sample type, historical batch data, or clinical release standards. Since iPSC-derived pancreatic islet cell spheres may exhibit inherent variability across different differentiation batches, culture densities, and sampling conditions, the preset CV threshold is not limited to a fixed value.

[0102] In a preferred embodiment, the preset CV threshold can be set to 15% or 20%; in another embodiment, the preset CV threshold can also be dynamically determined based on the CV distribution of historical qualified batches.

[0103] To further improve data reliability, outlier detection rules for single-view IEQ can be set in this step. The system can employ 3... Abnormal visual fields are identified using rules, Grubbs' test, interquartile range (IQR) rules, or other statistical outlier detection methods. In a preferred embodiment, the system employs the 3σ rule to calculate the deviation between the IEQ value of each visual field and the sample mean; if the absolute value of the deviation between a certain visual field's IEQ value and the sample mean exceeds a preset multiple of the standard deviation, for example, exceeding three times the standard deviation, then... If so, the viewpoint is marked as a suspicious outlier.

[0104] In another implementation, when the system uses the Grubbs test, it calculates the corresponding test statistic and compares it with the critical value at a preset significance level to determine whether there are outliers.

[0105] If an outlier is detected, the following actions will be taken: The system interface displays the message: "The IEQ value (XX IEQ) of field of view X deviates significantly from the mean (YY IEQ). Please review the original image of this field of view."

[0106] The system automatically retrieves the original image of the field of view and marks abnormal areas with red borders on the image, allowing operators to determine whether there are sampling bubbles, cell debris aggregation, or imaging quality problems.

[0107] After manual verification, the operator may choose to remove the abnormal field of view and resample, or retain the data but must indicate this in the report.

[0108] This embodiment provides a standardized equivalent automatic calculation method and system for iPSC-derived pancreatic islet cells. Compared with Embodiment 1, it adds an error control mechanism to ensure the robustness of the system under non-ideal sample conditions and the reliability of the report.

[0109] Example 3 A standardized equivalent automatic calculation method for iPSC-derived pancreatic islet cells, based on Example 1 or Example 2, further includes a functional correction equivalent calculation step based on measured functional data.

[0110] This embodiment does not change the standard islet equivalent calculation logic in Embodiment 1; that is, the system still first calculates the islet equivalent (IEQ) of each cell sphere based on the continuous volume integral formula. On this basis, the system can further introduce a functional correction factor to correct the islet equivalent, obtaining the functionally corrected islet equivalent.

[0111] The functional correction factor can be correlated with cell spheroid diameter, insulin secretion, ATP content, oxygen consumption rate, glucose-stimulated insulin secretion index and / or other functional test data, and is used to adapt to the functional heterogeneity between different iPSC differentiated cell lines, different culture conditions or cell spheroids of different sizes.

[0112] Specifically, the standard continuous volume integral formula This system is used to obtain standardized islet equivalent (IEQ) results. For certain iPSC differentiated cell lines or cell spheroids formed under specific culture conditions, there may be differences in functional activity per unit volume among cell spheroids of different sizes. To address this, the system can further introduce a functional correction factor based on the standard IEQ results to obtain a functionally corrected IEQ, thereby assisting in assessing functional differences among cell spheroids of different sizes or under different culture conditions.

[0113] The system-preset or user-imported modified models have the following general forms: ; The correction factor is a nonlinear correction factor related to the cell spheroid diameter d and / or measured functional data. The value range of the correction factor is not limited to a fixed range and can be determined based on functional validation experiments, cell line type, culture conditions, or normalization criteria. In one example, the correction factor can be set to 0.5 to 1.0; in other embodiments, it can also be set to other positive ranges.

[0114] The methods for determining the correction factor include, but are not limited to, the following: Method 1: Determined through fitting of measured data. For cell spheroids of different diameter ranges (e.g., 50 cell spheroids each with diameters <150μm, 150-250μm, 250-350μm, and >350μm), the insulin secretion or ATP content per unit cell spheroid was measured. The expected IEQ calculated using the standard formula was used as the independent variable, and the measured functional value (normalized) was used as the dependent variable. Regression analysis was then used to determine the IEQ. The mathematical expression for .

[0115] Method 2: User-defined import. For example, the system provides a graphical configuration interface that allows users to customize the diameter-correction factor mapping table by importing a comma-separated values ​​(CSV) file. File format requirements: the first column is the diameter (μm), and the second column is the correction factor. The system uses linear interpolation to calculate the corresponding diameter. value.

[0116] Furthermore, in the automated calculation system for standardized equivalents of iPSC-derived pancreatic islet cells, the selection and configuration of the correction model are as follows: The system provides an "Equivalent Calculation Parameters" menu in the user interface, which includes the following options: Standard mode: The standard IEQ is calculated using the continuous volume integral formula, which is suitable for conventional quality control scenarios.

[0117] Correction Mode: Based on the standard IEQ calculation results, it calls the function correction factor determined by fitting measured function data and outputs the function correction IEQ or function correction equivalent.

[0118] Hybrid mode: Outputs standard IEQ for cell spheres with a diameter smaller than a preset threshold; for cell spheres with a diameter greater than or equal to the preset threshold, outputs a functionally modified IEQ based on the standard IEQ. The preset threshold can be, for example, 250 μm, or can be adjusted according to the results of functional verification experiments.

[0119] Furthermore, the system automatically pops up a reminder when the user selects a correction mode: "Using a correction model will change the IEQ calculation results. It is recommended to confirm the applicability of the correction factor through a functional validation experiment before use. All reports will be automatically labeled with the model type and version used to ensure data traceability."

[0120] This embodiment provides a standardized equivalent automatic calculation method and system for iPSC-derived pancreatic islet cells. Compared with Embodiment 1 or 2, it adds a correction model configuration function, which provides the system with the ability to flexibly adapt to different application scenarios, and further expands the scope of application of the present invention.

[0121] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. An automated method for calculating the standardized equivalent of iPSC-derived pancreatic islet cells, characterized in that, Includes the following steps: Step 1: Acquire microscopic images of iPSC-derived pancreatic islet cell suspensions; Step 2: Input the microscopic image into a pre-trained deep learning segmentation model to obtain a binary mask representing the candidate region of pancreatic islet spheres; Step 3: Perform geometric demodulation processing based on parameter space voting on the connected regions in the binarized mask to separate independent cell sphere instances; The geometric demodulation process based on parameter space voting includes the following sub-steps: Extract the edge contour information of the connected regions; The edge contour points are mapped to the parameter space using the circular Hough transform and then accumulated and voted. Local energy maxima are detected in the parameter space to determine the true center coordinates and fitting radius of the occluded cell spheres. The boundary contours of the independent cell spheres are reconstructed based on the center coordinates and the fitted radius to complete the separation of the adhering clusters. Based on the reliability verification condition of the ratio of mask area to fitted geometric shape area, the verification and screening are carried out, and the process is as follows: Calculate the area ratio R between the target mask area and the target fitted geometry area of ​​the currently unverified connected region output by the deep learning segmentation model; when When the area ratio is greater than or equal to a preset threshold, the original segmentation mask is retained; when... If the area ratio is less than the preset threshold, the target is determined to have occlusion, depression or abnormal defects, and the geometric completion procedure is triggered to reconstruct the complete outline of the cell sphere. To avoid false completion due to small targets or unreliable fitting, the following auxiliary screening criteria are set: When the minimum circumscribed circle radius of the target If the radius is less than or equal to the preset radius threshold, the area is determined to be noise or debris and is skipped without completion. Reliability screening is performed by combining the number of convex hull points, the length of the minor axis of the ellipse fitting, and the ratio of the major axis to the minor axis. If the ellipse fitting result does not meet the conditions or the number of convex hull points is insufficient, the region is considered to be non-cellular spherical impurities or severely distorted cell debris, and the completion is skipped. Step 4: Based on the isolated individual cell sphere instances, calculate the equivalent diameter of each cell sphere, and calculate the islet equivalent of each cell sphere according to the continuous volume integral formula, and sum them up to obtain the total islet equivalent of the field of view.

2. The method for automatically calculating standardized equivalents of iPSC-derived pancreatic islet cells according to claim 1, characterized in that, The deep learning segmentation model is a semantic segmentation network based on a convolutional neural network or a visual transformer. The deep learning segmentation model is obtained through supervised training using a training dataset consisting of microscopic images of pancreatic islet spheres from iPSCs and manually labeled masks. The training dataset contains sample images of different differentiation batches, different culture densities, and different degrees of adhesion.

3. The method for automatically calculating standardized equivalents of iPSC-derived pancreatic islet cells according to claim 1, characterized in that, The equivalent diameter is calculated as follows: the physical area is obtained by converting the number of pixels occupied by the independent cell ball instance, and then the equivalent circle diameter d is calculated from the physical area; The formula for continuous volume integral is: Where d is in micrometers, and total islet equivalent is the islet equivalent of each individual cell sphere. sum.

4. The method for automatically calculating standardized equivalents of iPSC-derived pancreatic islet cells according to claim 1, characterized in that, It also includes step 5, which performs sampling quality control: Multiple pre-sampling operations were performed using an initial dilution factor k, with each sampling volume being [missing data]. Steps 1 to 4 are performed on the microscopic images sampled each time to calculate the total pancreatic islet equivalent in the field of view of each sample. The sample variance was calculated based on the results of multiple pre-sampling operations. and sample mean ; Based on the set confidence threshold Z and the allowable relative error Calculate the absolute error limit and using the formula Estimate the optimal total number of samples, n, required to achieve acceptable estimation accuracy; If the actual number of samplings is less than n, a prompt will be made to supplement the sampling until the cumulative number of samplings reaches n. Then, the average total islet equivalent and sampling volume of the multiple samplings will be used as the basis for the calculation. Dilution factor k can be used to infer the concentration of the original solution. ; The confidence threshold Z corresponds to 1.96 at a 95% confidence level, and the allowable relative error... Set to less than or equal to 5%.

5. The method for automatically calculating standardized equivalents of iPSC-derived pancreatic islet cells according to claim 4, characterized in that, It also includes step 6, which guides clinical solution preparation: Based on clinical target dose Based on the original solution concentration y obtained by reverse calculation, the required volume of formulation to be extracted is automatically calculated. It also outputs a quality control report containing the concentration of the original solution and the extraction volume.

6. The method for automatically calculating standardized equivalents of iPSC-derived pancreatic islet cells according to claim 1, characterized in that, The sample physical preparation step is included before step 1: A precise volume of cell suspension sample is aspirated from the cell culture vessel and selectively diluted according to the estimated concentration. The sample was injected into a flat-bottomed ultra-low adsorption plate, and then the plate was placed on the stage of an inverted microscope. Based on the density and size distribution of cell spheres in the sample, objective lenses with magnification of 1.25x, 2.5x, 4x, or 10x were selected for focusing and image acquisition.

7. The method for automatically calculating standardized equivalents of iPSC-derived pancreatic islet cells according to claim 1, characterized in that, It also includes step 7, which performs error control on the total pancreatic islet equivalent in the visual field: Based on the islet equivalent calculation results from multiple independent fields of view in the same batch, the coefficient of variation (CV) is calculated. When the CV exceeds a preset threshold, a prompt is issued to increase the number of samplings or check the sample homogeneity.

8. The method for automatically calculating standardized equivalents of iPSC-derived pancreatic islet cells according to claim 1, characterized in that, It also includes an equivalent correction step based on measured functional data: based on the islet equivalent obtained according to the continuous volume integral formula, a functional correction factor related to cell spheroid diameter, insulin secretion, ATP content and oxygen consumption rate is introduced to correct the islet equivalent to obtain a functionally corrected islet equivalent, so as to adapt to the functional heterogeneity of different iPSC differentiated cell lines or different culture conditions.

9. An automated system for calculating the standardized equivalent of iPSC-derived pancreatic islet cells, characterized in that: A method for performing the automatic calculation of standardized equivalents for iPSC-derived pancreatic islet cells as described in any one of claims 1-8; comprising: An optical signal acquisition module is used to acquire microscopic images of iPSC-derived pancreatic islet cell suspensions; The intelligent feature extraction and segmentation module has a built-in pre-trained deep learning segmentation model, which is used to receive the microscopic image and output a binary mask characterizing the candidate region of pancreatic islet spheres. The overlapping signal geometric demodulation module is used to perform geometric demodulation processing based on parameter space voting on the adhesive connected regions in the binarized mask to separate independent cell sphere instances; The adaptive calculation and error control module includes: The continuous volume integral unit is used to calculate the equivalent diameter of each cell sphere based on the isolated individual cell sphere instances, and to calculate the islet equivalent of each cell sphere using the continuous volume integral formula, and to sum them up to obtain the total islet equivalent of the field of view; The sampling number estimation unit is used to calculate the sample variance and sample mean based on the total pancreatic islet equivalent of multiple pre-sampled visual fields, and to estimate the optimal total number of samplings required to achieve qualified estimation accuracy based on the set confidence threshold and allowable relative error, and to prompt supplementary sampling based on the comparison results between the actual number of samplings and the optimal total number of samplings. The concentration back-calculation unit is used to back-calculate the stock solution concentration based on the average total islet equivalent, single sample volume, and dilution factor after the cumulative sampling number reaches the optimal total sampling number, and outputs clinical solution preparation guidance.

Images