Method, device and equipment for determining scanning height of microfluidic chip and storage medium

Abstract

The present disclosure provides a method, device and equipment for determining the scanning height of a microfluidic chip, and a storage medium. The method sets multiple sampling points in a scanning area and collects gray-scale microscopic images at each sampling point. A trained deep learning model is used to identify candidate region cells and determine the corresponding focal length offset. The focal length aggregation offset is obtained by comprehensive analysis. A scanning height field model of the scanning area is constructed based on the corresponding focal length aggregation offset at each sampling point. The obtained scanning height field model is used to determine the target scanning height corresponding to each position in the scanning area of the microfluidic chip, i.e. the best focal plane. The scanning height field model is used as the basis for determining the best scanning height, so that the focusing operation does not need to be performed frequently, the scanning efficiency is improved, and the wear of the scanning system is reduced.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method, apparatus, device and storage medium for scanning the height of a microfluidic chip. Background Technology

[0002] Microfluidic chips are widely used in cell analysis, clinical testing, and life science research due to their advantages such as low sample consumption, high integration, and high detection efficiency. When performing microscopic imaging on stained cells on microfluidic chips, it is usually necessary to perform continuous, high-resolution scanning of a large area of ​​the microfluidic chip to obtain high-quality microscopic images that can be used for analysis and interpretation.

[0003] In related technologies, to ensure the clarity of microscopic images, frequent focusing is usually required during the scanning process to determine the optimal focal plane for obtaining a clear image, i.e., to determine the optimal scanning height. However, frequent searches for the optimal scanning height not only affect scanning efficiency, but excessive mechanical movement also exacerbates wear and tear on the scanning system. Summary of the Invention

[0004] This disclosure provides a method, apparatus, device, and storage medium for determining the scanning height of a microfluidic chip, in order to at least solve the above-mentioned technical problems existing in the prior art.

[0005] A first aspect of this disclosure provides a method for determining the scanning height of a microfluidic chip, the method comprising:

[0006] Determine at least one sampling point within the scanning area corresponding to the microfluidic chip, and acquire a grayscale microscopic image of the sampling point at a preset scanning height; A deep learning model is used to identify candidate cell regions in the grayscale micrograph and determine the focal length offset corresponding to each candidate cell region, and the focal length aggregate offset is obtained by combining them. The scanning height field model is determined based on the focal length aggregation offset. The scanning height field model is used to determine the target scanning height corresponding to each position within the scanning area of ​​the microfluidic chip.

[0007] In one possible implementation, identifying candidate cell regions in the grayscale micrograph includes: The grayscale microscopic image is subjected to stepwise feature extraction to obtain the corresponding cell semantic feature map; Perform a two-dimensional convolution transformation on the cell semantic feature map, and determine the channel weight coefficients corresponding to the cell semantic feature map after the two-dimensional convolution transformation; Based on the channel weighting coefficients, feature enhancement is performed on the cell-scale features in the grayscale microscopic image to obtain an enhanced feature map; The enhanced feature map is subjected to full convolution processing to determine each candidate cell region and its corresponding confidence value.

[0008] In one possible implementation, determining the focal length offset corresponding to each candidate cell region includes: Determine the single cell image corresponding to each candidate cell region in the grayscale micrograph; Multi-level convolution feature extraction is performed on the single cell image to obtain the defocus feature characterizing the single cell image at the preset scanning height; After compressing the defocus feature, the focal length degradation feature is obtained; Linear regression processing is performed on the focal length degradation feature to obtain the corresponding focal length offset.

[0009] In one possible implementation, the step of performing multi-level convolutional feature extraction on a single cell image to obtain defocus features characterizing the single cell image at the preset scan height includes: Perform at least one two-dimensional convolution operation on the single cell image, and then normalize and process the two-dimensional convolution operation result with a nonlinear activation function to extract the corresponding local key features of the cell. The key local features of the cell are sequentially subjected to depthwise convolution and pointwise convolution to extract the cell structure degradation features corresponding to the preset scanning height. The cell structure degradation features are then processed by a nonlinear activation function to obtain the corresponding mid-to-high-level features. The mid-to-high-level features are subjected to multi-scale downsampling, and the features obtained at different scales are fused to obtain the corresponding out-of-focus features.

[0010] In one possible implementation, before determining the focal length offset corresponding to each candidate cell region, the method further includes: Based on the confidence value corresponding to each candidate cell region, target candidate cell regions that meet the confidence threshold are selected.

[0011] In one possible implementation, determining the scan height field model based on the focal length convergence offset includes: For each sampling point, the target scanning height corresponding to that sampling point is determined based on the focal length aggregation offset and the preset scanning height. For each position within the scanning area corresponding to the microfluidic chip, a corresponding weighting coefficient is determined based on the distance between the position and each sampling point, and a scanning height field model is determined based on the weighting coefficient and the target scanning height of each sampling point.

[0012] In one possible implementation, it further includes: The focal length aggregation offset corresponding to the real-time acquired grayscale microscopic image is determined using a trained deep learning model, thereby determining the corresponding real-time target scanning height, and the scanning height field model is updated exponentially based on the real-time target scanning height.

[0013] A second aspect of this disclosure provides a scanning height determination device for a microfluidic chip, the device comprising: The acquisition module is used to determine at least one sampling point within the scanning area corresponding to the microfluidic chip, and to acquire a grayscale microscopic image of the sampling point at a preset scanning height. The processing module is used to identify candidate cell regions in the grayscale micrograph using a trained deep learning model, and determine the focal length offset corresponding to each candidate cell region, and obtain the aggregated focal length offset. A construction module is used to determine a scanning height field model based on the focal length aggregation offset. The scanning height field model is used to determine the target scanning height corresponding to each position within the scanning area of ​​the microfluidic chip.

[0014] A third aspect of this disclosure provides an electronic device comprising: At least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the scan height determination method for a microfluidic chip as described in this disclosure.

[0015] A fourth aspect of this disclosure provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the microfluidic chip scanning height determination method described in this disclosure.

[0016] This disclosure discloses a method for determining the scanning height of a microfluidic chip. Multiple sampling points are set within the scanning area, and grayscale microscopic images are acquired at each sampling point. A trained deep learning model is used to identify candidate region cells and determine the corresponding focal length offset, which is then aggregated to obtain a focal length aggregate offset. Based on the focal length aggregate offset at each sampling point, a scanning height field model of the scanning area is constructed. This model is used to determine the target scanning height, i.e., the optimal focal plane, at each position within the scanning area of ​​the microfluidic chip. By using the scanning height field model as the basis for determining the optimal scanning height, frequent focusing operations are eliminated, improving scanning efficiency and reducing wear on the scanning system.

[0017] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0018] The above and other objects, features, and advantages of this disclosure will become readily apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. Several embodiments of this disclosure are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0019] Figure 1 A flowchart illustrating a method for determining the scanning height of a microfluidic chip according to an embodiment of the present disclosure is shown. Figure 2 This diagram illustrates the network structure of the cell semantic perception module in a microfluidic chip scanning height determination method according to an embodiment of the present disclosure. Figure 3 This diagram illustrates the structural composition of a single-cell focal length determination module in a microfluidic chip scanning height determination method according to an embodiment of the present disclosure. Figure 4 A 3D surface plot of the scanning height field in a method for determining the scanning height of a microfluidic chip according to an embodiment of the present disclosure is shown. Figure 5 This diagram illustrates the modular structure of a scanning height determination device for a microfluidic chip according to an embodiment of the present disclosure. Figure 6 A structural diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0020] To make the objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0021] In related technologies, to ensure the clarity of microscopic imaging, the optimal focal plane is typically determined through multi-layer Z-axis searching during the scanning process. This means that the Z-axis height of the scanning equipment needs to be adjusted multiple times, such as adjusting the Z-axis height of the microscopic platform, to acquire a clear microscopic image. This results in low scanning efficiency and makes it difficult to balance image quality and scanning efficiency. Furthermore, frequent mechanical movement also exacerbates wear and tear on the scanning system, increasing maintenance costs.

[0022] Based on this, this disclosure provides a method, apparatus, device, and storage medium for determining the scanning height of a microfluidic chip to solve the aforementioned problems. The method of this disclosure can be directly applied to a microscopic imaging platform, or to electronic devices other than microscopic imaging platforms. It works in conjunction with the microscopic imaging platform, controlling the platform to acquire images, training a model based on the acquired images, constructing a scanning height field, and determining the optimal scanning height based on the scanning height field so that the microscopic imaging platform performs scanning at the corresponding scanning height. The electronic device can be a smart terminal or a server. The smart terminal can be a mobile phone, a personal digital assistant (PAD), a tablet computer, a laptop computer, a desktop computer, etc. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN) services, and big data and artificial intelligence platforms.

[0023] This disclosure provides a method for determining the scanning height of a microfluidic chip, such as... Figure 1 As shown, the method includes: S101. Determine at least one sampling point within the scanning area corresponding to the microfluidic chip, and acquire a grayscale microscopic image of the sampling point at a preset scanning height.

[0024] In this step, the two-dimensional geometric dimensions of the scanning area in the microscopic platform coordinate system are obtained based on the scanning area range of the microfluidic chip. The scanning area is determined based on the scan start position and scan end position. The scan start position and scan end position can be selected based on actual requirements.

[0025] Furthermore, combining the geometric dimensions of the scanning area and the preset sampling density parameters, multiple spatial locations within the scanning area are determined as height sampling points, i.e., the x-axis and y-axis coordinates of multiple sampling points are determined within the scanning area. These height sampling points are regularly distributed within the scanning area, preferably arranged in a two-dimensional grid pattern, i.e., the spacing between sampling points is set according to the preset sampling density parameters along the row direction (x-axis direction) and column direction (y-axis direction) of the scanning area. It should be noted that the planar distance between adjacent sampling points needs to be greater than the imaging range of a single field of view during subsequent continuous scanning to reduce the amount of height sampling point data and thus improve overall scanning efficiency.

[0026] Furthermore, after determining multiple sampling points within the scanning area, the planar motion mechanism of the microscope platform is controlled to move sequentially to the spatial position corresponding to each sampling point. The Z-axis height of the microscope platform remains unchanged. At the preset initial height, the imaging system of the microscope platform is driven to acquire grayscale microscopic images of each sampling point at the preset height according to the preset imaging parameters, which serve as the image basis for subsequently determining the focal length offset.

[0027] S102. Use a deep learning model to identify candidate cell regions in a grayscale micrograph and determine the focal length offset corresponding to each candidate cell region, and then combine them to obtain the aggregated focal length offset.

[0028] In this step, a trained deep learning model is used to identify candidate cell regions in each grayscale microscopic image, determine the focal length offset corresponding to each candidate cell region, and synthesize them to obtain the aggregated focal length offset. It should be noted that the deep learning model in this step includes a cell semantic perception module, a single-cell focal length determination module, and a focal length robustness aggregation module. Specifically, the cell semantic perception module infers the candidate cell regions in a single frame of grayscale microscopic image to determine the corresponding candidate cell regions, and the single-cell focal length determination module generates a focal length evaluation result for each candidate cell region, i.e., the focal length offset state under the current imaging conditions. Finally, the focal length robustness aggregation module aggregates each focal length evaluation result, i.e., the focal length offset, to obtain the aggregated focal length offset. The aggregated focal length offset can be obtained for each grayscale microscopic image acquired at each sampling point, serving as the basis for constructing the subsequent scanning height field model.

[0029] S103. Determine the scanning height field model based on the focal length aggregation offset. The scanning height field model is used to determine the target scanning height corresponding to each position in the scanning area of ​​the microfluidic chip.

[0030] In this step, the focal length aggregation offset of each sampling point is the estimated focal height at the field of view level for that sampling point. Based on this, a scanning height field model corresponding to the scanning area is constructed by combining the focal length offsets of all sampling points. The 3D scanning height field model in this step is in the form of a 3D curved surface, used to determine the optimal focal plane height corresponding to any spatial location within the scanning area. This effectively solves the frequent focusing problem often encountered in related technologies and improves scanning efficiency.

[0031] This disclosure discloses a method for determining the scanning height of a microfluidic chip. Multiple sampling points are set within the scanning area, and grayscale microscopic images are acquired at each sampling point. A trained deep learning model is used to identify candidate region cells and determine the corresponding focal length offset, which is then aggregated to obtain a focal length aggregate offset. Based on the focal length aggregate offset at each sampling point, a scanning height field model of the scanning area is constructed. This model is used to determine the target scanning height, i.e., the optimal focal plane, at each position within the scanning area of ​​the microfluidic chip. By using the scanning height field model as the basis for determining the optimal scanning height, frequent focusing operations are eliminated, improving scanning efficiency and reducing wear on the scanning system.

[0032] In one possible implementation, identifying candidate cell regions in a grayscale micrograph includes: The grayscale microscopic image is subjected to stepwise feature extraction to obtain the corresponding cell semantic feature map; Perform a two-dimensional convolution transformation on the cell semantic feature map and determine the channel weight coefficients corresponding to the cell semantic feature map after the two-dimensional convolution transformation. Based on the channel weighting coefficients, feature enhancement is performed on the cell-scale features in the grayscale microscopic image to obtain the enhanced feature map; The enhanced feature map is processed by full convolution to determine each candidate cell region and its corresponding confidence value.

[0033] In this embodiment, as Figure 2 As shown, the cell semantic perception module includes a feature extraction backbone network, a cell-scale feature focusing module, and a candidate region prediction module. The model as a whole adopts a fully convolutional structure, and the above modules are connected in sequence.

[0034] It should be noted that, due to the presence of interference factors such as uneven staining, impurities, air bubbles, and background noise in cell micrographs, related techniques using full-field grayscale statistics or gradient information for focus evaluation struggle to distinguish between cellular and non-target cell regions. This makes them susceptible to interference from impurities and background, leading to unstable focus determination, especially when cells are sparsely distributed or partially occluded, further reducing focus accuracy and robustness. Therefore, this embodiment constructs a cell semantic awareness training dataset when training the deep learning model. This dataset serves as the training data foundation, enabling the model to identify cellular regions from grayscale micrographs and eliminate interference from non-cellular regions such as impurities, air bubbles, background noise, and other factors in focus determination.

[0035] Accordingly, when constructing the training dataset for cell semantic perception, multiple preset height offsets are actively introduced along the Z-axis of the microfluidic chip microscopic imaging system to acquire corresponding cell grayscale microscopic images at different height positions, thereby obtaining a cell imaging sample set covering multiple imaging clarity states.

[0036] Specifically, the original grayscale microscopic images acquired by the imaging system in this step have a resolution of 1024×1024 pixels. Based on a reference imaging height, multiple discrete height offset positions are set above and below this reference height. The interval between each height offset position can be set according to the depth-of-field range and imaging resolution of the imaging system. By acquiring grayscale microscopic images at each height offset position, the resulting image dataset can simultaneously include cell image samples in sharp, slightly out-of-focus, and severely out-of-focus states, thereby enhancing the subsequent model's adaptability to cell morphology changes under different focal length conditions.

[0037] Furthermore, after obtaining the aforementioned cell image samples, the spatial locations of the cells in the images are labeled. Preferably, candidate cell regions in the image are labeled using bounding rectangles. The candidate cell regions in this step represent the location and corresponding range of a single cell in the image. It should be noted that the labeling process only targets the cell regions used as the objectives, without requiring fine segmentation of the cell's internal structure. This reduces the difficulty of creating the training dataset and ensures consistency in labeling.

[0038] In addition, non-cellular interference samples were introduced as negative samples during model training when constructing the training dataset. These non-cellular interference samples included high-grayscale patches formed by dye deposition, debris particles generated inside the microfluidic chip, bubble edge structures, and background texture regions generated by the chip substrate or flow. By labeling these non-cellular regions and using them in conjunction with cellular samples for model training, the model can explicitly distinguish between cellular and non-cellular structures while learning cellular features, thereby improving the model's ability to suppress impurities and background interference in actual microfluidic chip scanning environments.

[0039] Furthermore, in order to enable the model to distinguish the contribution of cell regions to focal length determination at different imaging heights, an imaging validity label is introduced for the cell candidate regions. The imaging validity label is used to characterize the reliability of the cell candidate region as a basis for focal length determination at the current imaging height.

[0040] Accordingly, imaging validity labels can be constructed using a multi-level discrete labeling approach, including three categories: first imaging validity label, second imaging validity label, and third imaging validity label. Cell candidate regions labeled with the first imaging validity label are considered suitable as high-confidence input regions for focal length determination at the current imaging height, meaning they have higher weight in model training and subsequent focal length determination. Similarly, cell candidate regions labeled with the second imaging validity label are considered suitable as auxiliary focal length determination input regions for field-of-view aggregation at the current imaging height, but their contribution is lower than that of the cell regions corresponding to the first imaging validity label. Cell candidate regions labeled with the third imaging validity label are considered unsuitable as focal length determination input regions at the current imaging height and are used to guide the model in learning to suppress low-quality imaging regions during model training.

[0041] It should also be noted that the ratio of cell samples to non-cell samples is set according to the actual scanning environment. For example, the proportion of cell samples is 70%, and the corresponding proportion of non-cell samples is 30%. Furthermore, before model training, the training dataset is divided into a 70% training set, a 15% validation set, and a 15% test set to ensure that different resolution states and different types of samples are covered in each set.

[0042] Therefore, the cell semantic perception training dataset constructed based on the above method can not only cover multiple focal length imaging states, but also systematically include common non-cellular interference factors in the scanning process of microfluidic chips, providing a sufficient data foundation for subsequent cell region identification.

[0043] In this embodiment, the input of the cell semantic perception module is a single-channel 1024×1024 pixel original grayscale microscopic image acquired by the microscopic imaging system. During the inference process, the input image is not cropped, segmented, or rearranged. Instead, the entire field of view is used as the whole input, thereby ensuring that the spatial distribution relationship of cells in the scanning field of view is not destroyed, which is beneficial to subsequent field-level focal length aggregation and height field modeling.

[0044] In this embodiment, the feature extraction backbone network adopts a multi-level cascaded structure, progressively extracting features from the grayscale microscopic image to obtain the corresponding cell semantic feature map. Preferably, a four-level feature extraction is used, corresponding to four feature extraction modules. Each feature extraction stage uses a downsampling method with a step size of 2 to progressively reduce the spatial resolution of the feature map and progressively increase the number of channels. Accordingly, the spatial resolution of the feature map is adjusted sequentially from the original 1024×1024 to 512×512, then to 512×512, then to 128×128, finally outputting a 64×64 high-level semantic feature map. Specifically, in the first-level feature extraction stage, the shallow feature extraction module extracts cell edges, contours, and local texture features; in the second-level feature extraction stage, the cell morphology feature extraction module extracts cell morphology and scale-related features; in the third-level feature extraction stage, the structural integrity feature extraction module extracts cell structural integrity and regional stability features; and in the fourth-level feature extraction stage, the high-level semantic feature extraction module extracts high-level semantic features.

[0045] Each of the above feature extraction modules extracts corresponding spatial features based on depthwise separable convolution operations. Pointwise convolution operations are used to achieve feature fusion between channels, normalization operations are used to stabilize feature distribution, and nonlinear activation operations are used to enhance the model's expressive power.

[0046] The feature mapping process of any of the above-mentioned first-level feature extraction modules can be represented as:

[0047] In the formula, DWConv( represents the input feature map of the previous layer) ) represents depthwise separable spatial convolution; PWConv( () represents pointwise convolution; This represents the normalization operation; This represents a non-linear activation function.

[0048] By progressive downsampling, the feature map spatial resolution is made to satisfy:

[0049] in, The final output feature map has a resolution of .

[0050] Therefore, this embodiment can significantly reduce the number of model parameters and computational complexity by using a feature extraction backbone network to ensure sufficient extraction of cell edge, morphological and structural information, thus meeting the needs of industrial-grade real-time inference.

[0051] In this embodiment, the cell-scale feature focusing module is positioned between the feature extraction backbone network and the candidate region prediction model. It enhances feature responses that conform to the expected cell size range and suppresses background features that significantly deviate from the cell size range. The cell-scale feature focusing module includes a local feature enhancement submodule, a channel weight adaptive adjustment submodule, and a feature recalibration submodule.

[0052] It should be noted that the local feature enhancement submodule is used to perform two-dimensional convolutional transformation on the cell semantic feature map output by the feature extraction backbone network to enhance the spatial response related to local cell structure. The corresponding formula is as follows:

[0053] In the formula, This is the high-level feature map output by the backbone network, namely the cellular semantic feature map; This represents a two-dimensional convolution.

[0054] The channel weight adaptive adjustment submodule is used to perform global statistics on the cell semantic feature map after 2D convolution transformation, and generate the weight coefficients corresponding to each channel to reflect the importance of different channels to cell-scale expression. The corresponding formula is as follows: .

[0055]

[0056] In the formula, For channel description vectors, For channel weights, This represents a nonlinear mapping function.

[0057] The feature weight calibration submodule applies the adjusted channel weights to the original grayscale micrograph, enhancing features within the cell scale range and suppressing features significantly deviating from the cell scale range, thus obtaining scale-aware enhanced features, as shown below:

[0058] Therefore, this embodiment, through the cell-scale feature focusing module, enables the model to remain stably focused on the real cell region even in the complex microfluidic chip imaging environment, significantly reducing the interference of impurities, bubble edges and background textures on the generation of candidate cell regions.

[0059] The candidate region prediction module in this embodiment, based on the enhanced feature map output by the cell-scale feature focusing module, uses a fully convolutional approach to predict the location and scale of possible cell regions within the scanning field of view. The candidate region prediction module includes two independent prediction branches: location-scale prediction and confidence prediction. The location-scale prediction branch outputs the center offset and width / height scale parameters of the candidate region relative to the feature map grid position, characterizing the spatial location and size of the candidate cell region in the original input image. Each spatial location corresponds to one candidate region prediction result, independent of predefined anchor boxes or fixed scale parameters. The confidence prediction branch outputs a confidence value indicating whether the corresponding candidate region is suitable as a focal length determination input under the current imaging conditions. The confidence value reflects whether the candidate region contains structurally intact and image-quality-suitable effective cell regions for focal length determination, providing a reliable screening basis for subsequent single-cell focal length determination and field-of-view aggregation. The candidate region prediction module outputs a predicted feature map with a spatial resolution of 64×64, with each spatial location corresponding to at least one candidate region and its confidence information. The formula for the candidate region prediction result is as follows:

[0060] in, Indicates the first The center position of each candidate region in the original image coordinate system Indicates the width and height of the candidate region. This represents the confidence level of the corresponding candidate region.

[0061] The overall scale and computational complexity of the model parameters in this embodiment are limited to a range suitable for real-time operation on industrial-grade computing platforms. The model can complete cell candidate region inference for a single frame of grayscale microscopic image with millisecond-level latency during the scanning process. This eliminates the need for frequent traditional search-based autofocus operations during continuous scanning of the microfluidic chip, providing stable input for online correction and adaptive updates of the scanning height field model.

[0062] In one possible implementation, determining the focal length offset corresponding to each candidate cell region includes: Determine the single cell image corresponding to each candidate cell region in the grayscale micrograph; Multi-level convolution feature extraction is performed on a single cell image to obtain the defocus features that characterize the single cell image at a preset scanning height; After compressing the out-of-focus features, the focal length degradation features are obtained. Linear regression was performed on the focal length degradation characteristics to obtain the corresponding focal length offset.

[0063] In this embodiment, the deep learning model also includes a single-cell focal length determination module, which generates a corresponding focal length evaluation result based on the local imaging features of a single cell, and is used to characterize the focal shift state of the cell under the current imaging conditions.

[0064] It should be noted that a single-cell focal length determination dataset is created to train, debug, and evaluate the single-cell focal length determination model, enabling the model to learn the correspondence between structural degradation features and focal length offsets in cell images. Specifically, the optimal imaging focal plane corresponding to a selected field of view is first determined manually or automatically, and denoted as the reference height. A non-fixed step height offset is introduced along the Z-axis by controlling the microscopy platform to construct a height sampling sequence to cover clear, slightly out-of-focus, and severely out-of-focus states near the focal point. Preferably, multiple positive and negative offsets are set within symmetrical intervals. Correspondingly, grayscale microscopic images of the corresponding field of view are acquired. Based on a cell semantic perception model or manual annotation, the spatial position of a single cell is located in the field of view image, and a fixed-size single-cell region image is cropped based on the cell center. The cropped single-cell image size is preferably 128×128 pixels in single-channel grayscale format. For each single-cell image, a corresponding focal length offset label is constructed based on the difference between its acquired height position and the reference height position. In this embodiment, the focal length offset label is a continuous numerical value, used to supervise the regression training of the single-cell focal length determination model. In addition, to improve the model's generalization ability in real-world scanning environments, this embodiment also introduces a sample diversity strategy during the construction of the single-cell focal length determination dataset. This involves selecting cell samples of different shapes, sizes, and staining intensities, corresponding to cell images with slight overlap or incomplete edges, and removing samples with obvious occlusion, severe overlap, or incomplete cropping. The dataset is divided into training, validation, and test sets, preferably in a 7:2:1 or 8:1:1 ratio.

[0065] Specifically, such as Figure 3 As shown, the single-cell focal length determination model adopts a lightweight convolutional neural network structure, which includes a multi-level convolutional feature extraction module, a focal length degradation feature encoding module, and a focal length offset regression module arranged in sequence, forming an end-to-end regression model structure.

[0066] For each candidate cell region, the corresponding image patch is cropped from the original grayscale micrograph to obtain the corresponding single-cell image. The corresponding representation is as follows:

[0067] In the formula, Indicates from the original image The The center location of each candidate region Crop the image to 128×128. .

[0068] Accordingly, a multi-level convolutional feature extraction module is used to extract multi-scale features highly correlated with focal length changes from grayscale microscopic images of individual cells. The focal length degradation feature encoding module transforms spatially distributed blurred features into a compact representation suitable for regression. It performs global feature aggregation on the feature maps output by the multi-level convolutional feature extraction module, compressing the spatially blurred patterns and structural degradation information into low-dimensional feature vectors, thereby obtaining a focal length degradation feature representation reflecting the overall image sharpness degradation degree of a single cell. Through this feature encoding method, the model can avoid relying solely on local texture or a single edge response for judgment, thus improving robustness to focal length changes under different cell morphologies and slight noise perturbations. The corresponding formula is as follows:

[0069] In the formula, This is the focal length degradation feature vector.

[0070] In addition, the focal length offset regression module in this embodiment is based on focal length degradation feature representation and adopts a fully connected mapping structure to map the encoded feature vector into a single continuous output value Δz, which represents the focal length offset of the cell relative to the optimal focal plane. It should be noted that the last layer of the regression module in this embodiment does not use a classification activation function, but directly outputs the linear regression result to ensure that the model can accurately characterize the focal length offset, making it suitable for the subsequent construction and online updating of the scanning height field model. The corresponding formula is as follows:

[0071] In the formula, For the regression weight vector, This is a bias term.

[0072] After obtaining multiple cell focal length offsets using the single-cell focal length determination module in this embodiment, the focal length robustness aggregation module is used to statistically fuse the multiple cell focal length offsets within the same field of view to obtain a stable and reliable field-of-view focal length offset estimation result, which is the focal length aggregated offset.

[0073] Specifically, the input to the field-of-view-level focal length robustness aggregation module is a set of cell focal length observations acquired within the same scanning field of view, represented as follows:

[0074] In the formula, This represents the spatial position of a single cell in the field of view coordinate system. This represents the focal length offset output by the cell focal length determination model. This represents the confidence level corresponding to the cell semantic perception model, used to measure the reliability of the observation. This represents the number of effective cells participating in aggregation within the current field of view.

[0075] To suppress extreme focal length deviations caused by abnormal cell morphology or local noise, robust constraints are applied to the focal length observations before aggregation, and a robust screening interval is constructed using the median and absolute deviation:

[0076]

[0077] Only observations that meet the following conditions will be retained for subsequent aggregation:

[0078] In the formula, This is a preset robustness factor.

[0079] Furthermore, for the focal length shift of single cells that pass the robust screening, weights are assigned based on their confidence levels:

[0080] In the formula, The number of valid observations after robust screening.

[0081] Field of view focal length offset Obtained through a weighted average:

[0082] Therefore, by aggregating multiple cell-level focal length evaluation results (i.e., focal length offsets) through robust suppression and confidence-weighted aggregation, a field-of-view (FOV) level focal height estimate for the sampling point can be obtained, serving as the basis for constructing the scanning height field model. Robust aggregation of individual cell focal length determination results through confidence weighting and anomaly suppression avoids the decisive influence of single cells or local imaging anomalies on the focal length estimation results, making the FAV level focal length determination more stable and ensuring reliable focal length estimation results even under conditions of uneven cell distribution or fluctuating image quality.

[0083] In one embodiment, multi-level convolutional feature extraction is performed on a single cell image to obtain defocus features characterizing the single cell image at a preset scanning height, including: Perform at least one two-dimensional convolution operation on a single cell image, and then normalize and process the two-dimensional convolution operation results sequentially using nonlinear activation functions to extract the corresponding local key features of the cell. The key local features of the cell are sequentially subjected to depthwise convolution and pointwise convolution to extract the cell structure degradation features corresponding to the preset scanning height. The cell structure degradation features are then processed by a nonlinear activation function to obtain the corresponding mid-to-high-level features. Multi-scale downsampling is performed on mid-to-high-level features, and the features obtained from different scales are fused to obtain the corresponding out-of-focus features.

[0084] In this embodiment, the multi-level convolutional feature extraction module includes a local edge and contrast feature extraction module, a structural integrity and blur diffusion modeling module, and a defocusing trend perception feature extraction module, which are connected in sequence.

[0085] The local edge and contrast feature extraction module is used to extract the local high-frequency structural information most sensitive to focus changes in cell images. This module includes two layers of two-dimensional convolution operations using 3×3 convolution kernels, combined with batch normalization and non-linear activation functions to extract the corresponding key local cell features. By setting the downsampling stride, the feature map resolution is reduced from 128×128 to 64×64, thus reducing subsequent computational complexity while preserving key edge information. The local edge and contrast feature extraction module can effectively extract cell contours, nuclear membrane edges, and internal gray-level contrast features, providing basic information for subsequent focus determination. The corresponding formula is as follows:

[0086] In the formula, This represents the input image.

[0087] The structural integrity and fuzzy diffusion modeling module is used to characterize the structural degradation features of cells under different defocus states. Accordingly, this module adopts a depthwise separable convolutional structure, and through multi-level stacking and downsampling operations, it gradually reduces the spatial resolution of the feature map and improves the semantic level to obtain a mid-to-high-level feature representation that can reflect the overall structural integrity of the cell. The corresponding formula is as follows:

[0088] The defocusing trend perception feature extraction module performs multi-scale downsampling processing on the above features to obtain a high-level feature map with a spatial size of approximately 16×16. This feature map simultaneously encodes information such as the overall cell contour degradation, the degree of internal texture loss, and the blur diffusion trend. It is a comprehensive representation of the defocusing state of a single cell at the current imaging height, and the corresponding formula is as follows:

[0089] In the formula, This indicates multi-scale downsampling and feature fusion operations.

[0090] In one possible implementation, before determining the focal length offset corresponding to each candidate cell region, the method further includes: Based on the confidence value corresponding to each candidate cell region, target candidate cell regions that meet the confidence threshold are selected.

[0091] In this embodiment, to avoid interference from background, impurities, and overlapping cell regions on the focal length determination result, the candidate cell regions output by the cell-scale feature focusing module are further screened before being input into the single-cell focal length determination module. Only candidate regions that meet the threshold conditions are retained as target candidate cell regions. The corresponding formula is as follows:

[0092] In the formula, The reliability threshold can be preset and set according to actual needs.

[0093] This ensures that the input samples that subsequently enter the focal length determination model come from single-cell regions with high imaging quality and intact structure.

[0094] In one possible implementation, determining the scan height field model based on the focal length convergence offset includes: For each sampling point, the target scanning height corresponding to that sampling point is determined based on the focal length aggregation offset and the preset scanning height. For each position within the scanning area of ​​the microfluidic chip, the corresponding weighting coefficient is determined based on the distance between that position and each sampling point, and the scanning height field model is determined based on the weighting coefficient and the target scanning height of each sampling point.

[0095] In this embodiment, the sampling point is defined as follows:

[0096] It should be noted that at each height calibration sampling point, a grayscale microscopic image is acquired at the corresponding location. The actual imaging height of the scanning platform at that sampling point is... The acquired grayscale microscopic images are input into a trained deep learning model, which then predicts the field-of-view focal length observation results at that location. The optimal plane height observation value at this sampling point is obtained. The corresponding formula is as follows:

[0097] Accordingly, the scanning height field is defined as:

[0098] The construction of the scanning height field is formulated as the following optimization problem:

[0099] In the formula, the first term is used to ensure that the height field is fitted to the field-of-view focal length observation at the sampling point; For balance coefficient, The spatial smoothing regularization term for the height field is used to suppress violent oscillations in the height field in space, so that the height field variation conforms to the actual geometry of the microfluidic chip, wherein:

[0100] The weight function in the formula Defined as:

[0101] like Figure 4 As shown, a scanning height field model is determined to determine the target scanning height corresponding to each position within the scanning area of ​​the microfluidic chip, so as to achieve continuous microscopic scanning without the need for frequent focusing.

[0102] In one possible implementation, it further includes: The focal length aggregation offset corresponding to the real-time acquired grayscale microscopic image is determined by using a trained deep learning model, so as to determine the corresponding real-time target scanning height, and the scanning height field model is updated exponentially based on the real-time target scanning height.

[0103] In this embodiment, during the scanning process, the trained deep learning model is used again to evaluate the focal height of the current field of view by combining the real-time acquired grayscale microscopic images, and the scanning height field model is adaptively updated accordingly to achieve continuous and highly consistent scanning of the microfluidic chip. The model evaluation result can be defined as:

[0104] In the formula, Indicates the scan timing. Indicates the current scan height The offset relative to the optimal focal plane.

[0105] It should be noted that this observation can be considered as a local constraint on the height field:

[0106] The scan height field model is adaptively updated using an exponential smoothing method:

[0107] In the formula, To update the coefficients.

[0108] In addition, to prevent highly abrupt changes introduced by misobservations, this embodiment sets restrictions on the update magnitude:

[0109] When the threshold is exceeded At that time, the update volume is truncated.

[0110] Therefore, this embodiment improves the single-point focal length judgment into a continuous spatial height model through the above-mentioned adaptive update mechanism for the scanning height field, supports learning and correction while scanning, significantly reduces the need for frequent autofocus across the entire field of view, and provides basic support for high-throughput and high-stability microscopy.

[0111] It should be noted that related technologies rely on static models to predict scanning height to reduce the number of focusing attempts. However, these typically assume a smooth chip surface, making them ill-suited for complex microfluidic chips or deformations and assembly errors that occur during actual use. Furthermore, these models usually lack online update capabilities, and any initial calibration errors can compromise subsequent image quality. In contrast, this embodiment combines real-time imaging results with online correction and adaptive updates to the scanning height field model. This dynamically compensates for focal plane shifts caused by microfluidic chip surface irregularities, assembly errors, and environmental changes, avoiding errors from static height models. This results in smooth focal plane changes and highly consistent image quality across a large scanning area. This not only effectively reduces focusing time and improves scanning efficiency and imaging consistency but also demonstrates good robustness against interference from impurities and localized imaging variations.

[0112] The second aspect of this disclosure, such as Figure 5 As shown, a scanning height determination device 500 for a microfluidic chip is provided, the device comprising: The acquisition module 501 is used to determine at least one sampling point in the scanning area corresponding to the microfluidic chip, and acquire the grayscale microscopic image corresponding to the sampling point at a preset scanning height. The processing module 502 is used to identify candidate cell regions in a grayscale micrograph using a trained deep learning model, and to determine the focal length offset corresponding to each candidate cell region, and to obtain the aggregated focal length offset. Module 503 is used to determine the scanning height field model based on the focal length aggregation offset. The scanning height field model is used to determine the target scanning height corresponding to each position in the scanning area of ​​the microfluidic chip.

[0113] In one embodiment, the processing module 502 is used to perform stepwise feature extraction on the grayscale microscopic image to obtain the corresponding cell semantic feature map; Perform a two-dimensional convolution transformation on the cell semantic feature map and determine the channel weight coefficients corresponding to the cell semantic feature map after the two-dimensional convolution transformation. Based on the channel weighting coefficients, feature enhancement is performed on the cell-scale features in the grayscale microscopic image to obtain the enhanced feature map; The enhanced feature map is processed by full convolution to determine each candidate cell region and its corresponding confidence value.

[0114] In one embodiment, the processing module 502 is further configured to determine the single cell image corresponding to each candidate cell region in the grayscale micrograph; Multi-level convolution feature extraction is performed on a single cell image to obtain the defocus features that characterize the single cell image at a preset scanning height; After compressing the out-of-focus features, the focal length degradation features are obtained. Linear regression was performed on the focal length degradation characteristics to obtain the corresponding focal length offset.

[0115] In one embodiment, the processing module 502 is further configured to perform at least one two-dimensional convolution operation on a single cell image, and to perform normalization and nonlinear activation function processing on the two-dimensional convolution operation result in sequence to extract the corresponding local key features of the cell. The key local features of the cell are sequentially subjected to depthwise convolution and pointwise convolution to extract the cell structure degradation features corresponding to the preset scanning height. The cell structure degradation features are then processed by a nonlinear activation function to obtain the corresponding mid-to-high-level features. Multi-scale downsampling is performed on mid-to-high-level features, and the features obtained from different scales are fused to obtain the corresponding out-of-focus features.

[0116] In one embodiment, the processing module 502 is further configured to filter target candidate cell regions that meet the confidence threshold based on the confidence value corresponding to each candidate cell region.

[0117] In one embodiment, the construction module 503 is used to determine the target scanning height corresponding to each sampling point based on the focal length aggregation offset and the preset scanning height. For each position within the scanning area of ​​the microfluidic chip, the corresponding weighting coefficient is determined based on the distance between that position and each sampling point, and the scanning height field model is determined based on the weighting coefficient and the target scanning height of each sampling point.

[0118] In one possible implementation, an optimization module is also included, which is used to determine the focal length aggregation offset corresponding to the real-time acquired grayscale microscopic image using a trained deep learning model, so as to determine the corresponding real-time target scanning height, and to perform exponential smoothing update of the scanning height field model based on the real-time target scanning height.

[0119] By way of example, this application also provides an electronic device, including: processor; Memory used to store processor-executable instructions; The processor is used to read executable instructions from memory and execute the instructions to implement the above-described method for determining the scan height of a microfluidic chip.

[0120] By way of example, this application also provides a computer-readable storage medium storing a computer program for performing the above-described method for determining the scanning height of a microfluidic chip.

[0121] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0122] like Figure 6 As shown, device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 602 or a computer program loaded from storage unit 608 into random access memory (RAM) 603. RAM 603 may also store various programs and data required for the operation of device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via bus 604. Input / output (I / O) interface 605 is also connected to bus 604.

[0123] Multiple components in device 1000 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of monitors, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0124] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a method for determining the scan height of a microfluidic chip. For example, in some embodiments, a context processing method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of a method for determining the scan height of a microfluidic chip described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured by any other suitable means (e.g., by means of firmware) to perform a scan height determination method for a microfluidic chip.

[0125] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0126] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0127] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0128] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0129] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0130] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0131] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, and this is not limited herein.

[0132] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0133] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A method for determining the scanning height of a microfluidic chip, characterized in that, The method includes: Determine at least one sampling point within the scanning area corresponding to the microfluidic chip, and acquire a grayscale microscopic image of the sampling point at a preset scanning height; A deep learning model is used to identify candidate cell regions in the grayscale micrograph and determine the focal length offset corresponding to each candidate cell region, and the focal length aggregate offset is obtained by combining them. The scanning height field model is determined based on the focal length aggregation offset. The scanning height field model is used to determine the target scanning height corresponding to each position within the scanning area of ​​the microfluidic chip.

2. The method for determining the scanning height of a microfluidic chip according to claim 1, characterized in that, The identification of candidate cell regions in the grayscale micrograph includes: The grayscale microscopic image is subjected to stepwise feature extraction to obtain the corresponding cell semantic feature map; Perform a two-dimensional convolution transformation on the cell semantic feature map, and determine the channel weight coefficients corresponding to the cell semantic feature map after the two-dimensional convolution transformation; Based on the channel weighting coefficients, feature enhancement is performed on the cell-scale features in the grayscale microscopic image to obtain an enhanced feature map; The enhanced feature map is subjected to full convolution processing to determine each candidate cell region and its corresponding confidence value.

3. The method for determining the scanning height of a microfluidic chip according to claim 1, characterized in that, Determining the focal length offset corresponding to each candidate cell region includes: Determine the single cell image corresponding to each candidate cell region in the grayscale micrograph; Multi-level convolution feature extraction is performed on the single cell image to obtain the defocus feature characterizing the single cell image at the preset scanning height; After compressing the defocus feature, the focal length degradation feature is obtained; Linear regression processing is performed on the focal length degradation feature to obtain the corresponding focal length offset.

4. The method for determining the scanning height of a microfluidic chip according to claim 3, characterized in that, The step of performing multi-level convolution feature extraction on a single cell image to obtain the defocus features characterizing the single cell image at the preset scan height includes: Perform at least one two-dimensional convolution operation on the single cell image, and then normalize and process the two-dimensional convolution operation result with a nonlinear activation function to extract the corresponding local key features of the cell. The key local features of the cell are sequentially subjected to depthwise convolution and pointwise convolution to extract the cell structure degradation features corresponding to the preset scanning height. The cell structure degradation features are then processed by a nonlinear activation function to obtain the corresponding mid-to-high-level features. The mid-to-high-level features are subjected to multi-scale downsampling, and the features obtained at different scales are fused to obtain the corresponding out-of-focus features.

5. The method for determining the scanning height of a microfluidic chip according to claim 4, characterized in that, Before determining the focal length offset corresponding to each candidate cell region, the method further includes: Based on the confidence value corresponding to each candidate cell region, target candidate cell regions that meet the confidence threshold are selected.

6. The method for determining the scanning height of a microfluidic chip according to claim 1, characterized in that, The process of determining the scanning height field model based on the focal length convergence offset includes: For each sampling point, the target scanning height corresponding to that sampling point is determined based on the focal length aggregation offset and the preset scanning height. For each position within the scanning area corresponding to the microfluidic chip, a corresponding weighting coefficient is determined based on the distance between the position and each sampling point, and a scanning height field model is determined based on the weighting coefficient and the target scanning height of each sampling point.

7. The method for determining the scanning height of a microfluidic chip according to any one of claims 1-6, characterized in that, Also includes: The focal length aggregation offset corresponding to the real-time acquired grayscale microscopic image is determined using a trained deep learning model, thereby determining the corresponding real-time target scanning height, and the scanning height field model is updated exponentially based on the real-time target scanning height.

8. A device for determining the scanning height of a microfluidic chip, characterized in that, The device includes: The acquisition module is used to determine at least one sampling point within the scanning area corresponding to the microfluidic chip, and to acquire a grayscale microscopic image of the sampling point at a preset scanning height. The processing module is used to identify candidate cell regions in the grayscale micrograph using a trained deep learning model, and determine the focal length offset corresponding to each candidate cell region, and obtain the aggregated focal length offset. A construction module is used to determine a scanning height field model based on the focal length aggregation offset. The scanning height field model is used to determine the target scanning height corresponding to each position within the scanning area of ​​the microfluidic chip.

9. An electronic device, characterized in that, include: At least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform claim 1. The method for determining the scanning height of the microfluidic chip as described in any one of 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the method for determining the scanning height of a microfluidic chip according to any one of claims 1-7.

Images