A two-dimensional cellular two-photon image analysis system and method

Abstract

This invention relates to the field of biomedical image processing technology, specifically a two-dimensional cellular two-photon image analysis system and method, comprising: an image preprocessing subsystem for acquiring and preprocessing calcium imaging video to obtain feature-enhanced images; a subcellular segmentation subsystem for segmenting the image into subcellular structures using a hybrid detection strategy, wherein the hybrid detection strategy consists of manual annotation and adaptive thresholding; and a calcium response feature extraction and analysis subsystem for detecting the fluorescence curve of the calcium response and extracting calcium response event parameter features from the segmented image. This solution can segment astrocytes of different morphologies, distinguish subcellular structures, and employ adaptive thresholding for calcium response detection and analysis, thereby improving segmentation accuracy and analysis efficiency while reducing manual and time costs.

Description

Technical Field

[0001] This invention relates to the field of biomedical image processing technology, specifically to a two-dimensional cellular two-photon image analysis system and method. Background Technology

[0002] Astrocytes are the most widely distributed type of cell in the mammalian brain and the largest type of glial cell. Using classic metallization techniques (silver staining), these glial cells are shown to be star-shaped, with numerous long, branching processes extending from their cell bodies to fill the spaces between the cell bodies and processes of nerve cells, thus supporting and separating them. In recent years, increasing evidence suggests that astrocytes are considered passive supporting cells that play a crucial role in human physiological activities. However, due to the complexity of astrocyte structure, existing subcellular segmentation methods often fail to meet the precision requirements of scientific research; furthermore, the detection and analysis of calcium signals within astrocytes has always been a major challenge in research, as traditional manual detection methods are time-consuming and costly.

[0003] Specifically, the existing technology for analyzing calcium imaging data of astrocytes has the following shortcomings: 1. Previous studies have mainly focused on the astrocyte cell body, main stem, and terminal foot, while lacking analysis of the micro-regions (micro-domains) in addition to the above cellular structures; 2. Astrocytes have complex and dynamic morphology, and traditional segmentation methods are difficult to adapt to their heterogeneity. The accurate differentiation of subcellular structures within a single cell still needs further optimization. 3. Currently, calcium signal detection relies on manual methods. The empirical formula method for calcium signal analysis requires manual adjustment of the threshold, and single-sample analysis often takes about 2-3 hours, which is time-consuming.

[0004] Therefore, there is an urgent need for a two-dimensional cellular two-photon image analysis system and method that can segment astrocytes of different morphologies, distinguish subcellular structures, and perform threshold adaptation, as well as calcium response detection and analysis, in order to improve segmentation accuracy and analysis efficiency, and reduce labor and time costs. Summary of the Invention

[0005] One of the objectives of this invention is to provide a two-dimensional cellular two-photon image analysis system that can segment astrocytes of different morphologies, distinguish subcellular structures, and has an adaptive threshold, as well as perform calcium response detection and analysis, thereby improving segmentation accuracy and analysis efficiency, and reducing labor and time costs.

[0006] The basic solution provided by this invention is: a two-dimensional cell two-photon image analysis system, comprising: The image preprocessing subsystem is used to acquire calcium imaging video and perform preprocessing to obtain feature-enhanced images; The subcellular segmentation subsystem is used to segment images into subcellular structures using a hybrid detection strategy, which consists of manual annotation and adaptive thresholding. The calcium reaction feature extraction and analysis subsystem is used to detect the fluorescence curve of the calcium reaction and extract the calcium reaction event parameter features from the segmented images.

[0007] The second objective of this invention is to provide a two-dimensional cellular two-photon image analysis method that can segment astrocytes of different morphologies, distinguish subcellular structures, and has an adaptive threshold. It can also perform calcium response detection and analysis to improve segmentation accuracy and analysis efficiency, and reduce labor and time costs.

[0008] The present invention provides a second basic solution: a two-dimensional cell two-photon image analysis method, which uses the above-mentioned two-dimensional cell two-photon image analysis system.

[0009] Beneficial effects: First, acquire calcium imaging video and preprocess it to obtain feature-enhanced images, making the structure of astrocytes more obvious, which facilitates subsequent subcellular segmentation and helps improve the segmentation accuracy. Then, a hybrid detection strategy is adopted to segment the image into subcellular structures, that is, to use manual annotation and adaptive thresholding to segment the image into subcellular structures. When segmenting micro-domains, the adaptive thresholding can perform targeted segmentation, which is an improvement over the traditional fixed thresholding method, especially in the ability to capture subcellular micro-structures. Moreover, no manual thresholding is required, saving labor costs and setting time. Finally, the segmented images are subjected to fluorescence curve detection of calcium reaction and extraction of calcium reaction event parameter features, thereby reducing reliance on manual labor, reducing analysis time, and saving labor and time costs.

[0010] In summary, this scheme can segment astrocytes of different morphologies, distinguish subcellular structures, and has an adaptive threshold. It also performs calcium response detection and analysis to improve segmentation accuracy and analysis efficiency while reducing labor and time costs. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the structure of a two-dimensional cell two-photon image analysis system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the interface of the image preprocessing subsystem in Embodiment 1 of the two-dimensional cell two-photon image analysis system of the present invention; Figure 3This is a schematic diagram of the interface structure of the sub-cell segmentation subsystem in Embodiment 1 of the two-dimensional cell two-photon image analysis system of the present invention; Figure 4 This is a schematic diagram of the process for extracting cell bodies in an embodiment of the two-dimensional cell two-photon image analysis system of the present invention; Figure 5 This is a schematic diagram of the Region of Interest (ROI) in Embodiment 1 of the two-dimensional cell two-photon image analysis system of the present invention; Figure 6 This is a schematic diagram of the target portion in an embodiment of a two-dimensional cell two-photon image analysis system of the present invention; Figure 7 This is a schematic diagram illustrating the disconnection of the cell body from the main stem and micro-domain in Embodiment 1 of the two-dimensional cell two-photon image analysis system of the present invention. Figure 8 This is a schematic diagram of the first mask in an embodiment of a two-dimensional cell two-photon image analysis system of the present invention; Figure 9 This is a schematic diagram of the second mask in an embodiment of a two-dimensional cell two-photon image analysis system of the present invention; Figure 10 This is a schematic diagram of the cell body in Embodiment 1 of the two-dimensional cell two-photon image analysis system of the present invention; Figure 11 This is a schematic diagram of the process for extracting micro-domains in a first embodiment of the two-dimensional cell two-photon image analysis system of the present invention; Figure 12 This is a schematic diagram of the micro-domain in Embodiment 1 of the two-dimensional cell two-photon image analysis system of the present invention; Figure 13 This is a schematic diagram of the interface of the calcium reaction feature extraction and analysis subsystem in Embodiment 1 of the two-dimensional cell two-photon image analysis system of the present invention; Figure 14 This is the ROC curve diagram in Embodiment 1 of the two-dimensional cell two-photon image analysis system of the present invention; Figure 15 This is an image with enhanced features from a second embodiment of the two-dimensional cell two-photon image analysis method of the present invention; Figure 16 This is a diagram showing the segmentation results of the main branch, cell body, and terminal foot in Embodiment 2 of the two-dimensional cell two-photon image analysis method of the present invention. Figure 17 This is a schematic diagram of the micro-domain in Embodiment 2 of the two-dimensional cell two-photon image analysis method of the present invention; Figure 18 This is a schematic diagram of the structure for detecting the fluorescence curve of the calcium reaction in Example 2 of the two-dimensional cell two-photon image analysis method of the present invention; Figure 19This is a feature map of calcium response event parameters in Example 2 of the two-dimensional cell two-photon image analysis method of the present invention; Figure 20 This is a display diagram of the multi-view interactive subsystem in Embodiment 2 of the two-dimensional cell two-photon image analysis method of the present invention. Detailed Implementation

[0012] The following detailed description illustrates the specific implementation method: Example 1 This embodiment is basically as shown in the appendix. Figure 1 As shown: A two-dimensional cell two-photon image analysis system, comprising: an image preprocessing subsystem, a subcellular segmentation subsystem, a calcium response feature extraction and analysis subsystem, and a multi-view interaction subsystem; The image preprocessing subsystem is used to acquire calcium imaging video and perform preprocessing to obtain feature-enhanced images; In this embodiment, the image preprocessing subsystem preprocesses the imported AVI format video, i.e., the calcium imaging video, to achieve feature enhancement of the calcium image. Its interface is as follows: Figure 2 As shown; Specifically, the image preprocessing subsystem includes: an input module, a main processing module, an additional processing module, and a video frame management module. The input module is used to input the video of calcium imaging; The main processing module is used to generate a mean image and dynamically adjust the contrast and brightness based on the input calcium imaging video. An additional processing module is used to generate a maximum value image and a standard deviation image based on the input calcium imaging video; The video frame management module is used to view dynamic changes in the video. The generation methods of mean image, maximum image, and standard deviation image can all enhance the image, making the structure of astrocytes more obvious. At the same time, the dynamic changes of the video can be viewed through the video frame management module with an accuracy of 1 frame. The subsequent subsystem selects one type of image for processing. In this embodiment, the mean image is selected by default. The mean image is generated by calculating the average value of the pixel positions corresponding to each frame in the video. in, The first of the mean image The intensity of each pixel It is a mean function. For the video number The first frame of the image Intensity of each pixel The number of frames in the video. This represents the total number of frames in the video. The maximum value image is generated by extracting the highest brightness value at the corresponding pixel location in each frame of the video. in, The first of the maximum value images The intensity of each pixel It is a function for maximizing the value; The standard deviation image is generated by extracting the standard deviation of the pixel positions corresponding to each frame in the video. in, The first of the mean image The intensity of each pixel This is the standard deviation function.

[0013] In this embodiment, the contrast and brightness are dynamically adjusted, including: the contrast adjustment is done using the adapthisteq function in MATLAB, which is a function in MATLAB that implements contrast-limited adaptive histogram equalization (CLAHE); the brightness adjustment is done by changing the brightness of each pixel.

[0014] The subcellular segmentation subsystem employs a hybrid detection strategy to segment images into subcellular structures; this strategy combines manual annotation and adaptive thresholding. The interface of the subcellular segmentation subsystem is shown below. Figure 3 As shown; Subcellular structures include: single cell, main branch, endfeet, microdomain, and soma. Specifically, the subcellular segmentation subsystem uses manual drawing of ROI (Region of Interest) for single cells, main branches, and terminal feet, that is, it obtains the ROI of single cells, main branches, and terminal feet through manual annotation. Morphological opening operation processing combined with adaptive threshold segmentation is used to extract cell bodies, i.e., extract cell body ROIs; A noise model combined with a spatial domain is used for multi-level screening, and adaptive threshold segmentation is used to extract micro-domains, i.e., micro-domain ROIs are extracted. For cell body extraction, since cell bodies have high contrast and large volume (3-5 μm in diameter) in two-photon microscopy imaging, this scheme uses morphological opening operation processing combined with adaptive threshold segmentation (Otsu algorithm, also known as Otsu method) for automatic cell body extraction. The traditional Otsu method is sensitive to noise and performs poorly when the target and background ratios are unbalanced. Therefore, we add local standard deviation weighting to the Otsu method for subsequent calculations. We calculate the neighborhood standard deviation (3×3 window) of each pixel to generate a standard deviation weight map, and then merge the weight map with the original image. in This operation, by adjusting the coefficient, enhances the contrast of high-texture areas and suppresses uniform backgrounds.

[0015] Morphological opening operations combined with adaptive threshold segmentation are used to extract cell bodies, such as... Figure 4 As shown, the specific process is as follows: Obtain the drawn region of interest (ROI) for a single cell; in this embodiment, the images.roi.Freehand function in MATLAB is used to manually draw the ROI for a single cell, such as... Figure 5 As shown; By calculating the maximum inter-class variance for single-cell ROIs, binarizing the image, and employing a local standard deviation enhancement method, the image is highlighted to highlight regions with significant differences in fluorescence signal intensity, thus obtaining the single-cell ROI as the target region, i.e., the main cell portion. Figure 6 As shown; the maximum inter-class variance (MOV) is achieved by finding a threshold to maximize the variance between the foreground and background, thus realizing effective image segmentation. Specifically, it iterates through all gray levels in the image, calculates the inter-class variance after dividing the image into two parts (foreground and background) using each gray level as a threshold; assuming the total number of gray levels in the image is... The threshold is The images were divided into two categories. and ,in Includes all grayscale values ​​less than C2 contains all pixels with gray values ​​greater than or equal to 10 ... pixels, inter-class variance for: in and They are respectively and The average gray value; The binarized target portion is subjected to an opening operation to disconnect the cell body from non-cell body parts (such as: main branches, micro-domains, and terminals), for example... Figure 7 As shown, the largest connected component is selected as the first mask, as follows. Figure 8 As shown; the opening operation includes: an octagonal erosion operation and an octagonal expansion operation; Based on the first mask, adaptive threshold segmentation is performed to obtain the cell body; Specifically, it includes: Calculate the centroid of the first mask and the average distance from the centroid to the edge of the connected domain; Using the average distance as the radius and the centroid as the center, fit a circle as the second mask, such as... Figure 9 As shown; Calculate the intersection of the first and second masks; the common portion of the intersection is used as the final automatically detected cell body, such as... Figure 10 As shown.

[0016] A noise model combined with a spatial domain is used for multi-level screening, and adaptive threshold segmentation is used to extract micro-domains, such as... Figure 11 As shown, the specific process is as follows: Construct a noise model; Specifically, in this embodiment, the noise model is a Gaussian noise model, which is represented by a noise spatial distribution matrix; In complex biological image processing, accurate noise modeling and efficient noise suppression are the core prerequisites for ensuring the accuracy of subsequent analysis. In this embodiment, the noise model construction step is based on three types of key images generated by the preprocessing subsystem: the mean image reflects the overall level of regional grayscale, the maximum value image captures the peak features of local signals, and the standard deviation image directly depicts the dispersion of pixel values ​​and is the core carrier for noise feature extraction. To achieve quantitative characterization of noise, we first analyze the probability distribution characteristics of pixel values ​​in the standard deviation image using statistical methods. Based on these probability distribution characteristics, we determine the mean intensity of the noise. and variance These two parameters constitute the basic description of the global characteristics of the noise, where the statistical method is Gaussian distribution fitting; Simultaneously, by combining the spatial fluctuation patterns of gray values ​​in the mean image, a noise model incorporating spatial domain correlation is further established, thus overcoming the limitation of traditional pure statistical models that ignore spatial correlation. Then, a 3×3 sliding window is used to perform local traversal calculations on the standard deviation image, generating noise variance data reflecting spatial heterogeneity pixel by pixel, ultimately constructing a complete noise spatial distribution matrix. This provides a precise spatial positioning basis for subsequent targeted noise reduction; in It follows a standard normal distribution. The noise mean. This represents the noise variance.

[0017] A spatial domain joint denoising model is used to perform differential denoising on the original image to obtain the denoised image; the original image is an image of the ROI with single cells, main branches, and terminal feet already drawn; Specifically, the CLAHE (Contrast Limiting Adaptive Histogram Equalization) algorithm is used to preprocess the original image. By dividing the original image into multiple sub-blocks and performing histogram equalization on each, the detail representation in weak signal areas is effectively enhanced, while avoiding the noise amplification problem caused by traditional equalization, thus laying a signal foundation for subsequent denoising. Based on the noise spatial distribution matrix Differential filtering is performed: For complex regions formed by sub-blocks where the local noise variance exceeds a preset variance threshold, a 5×5 window bilateral filter is used, with its weighting coefficients designed to be similar to... The filtering weights are inversely proportional, meaning that the stronger the noise, the greater the filtering weight. This effectively suppresses noise while preserving details by maintaining edge gradients. For smooth regions formed by sub-blocks with local noise variance less than a preset variance threshold, a 3×3 window median filter is used to achieve accurate removal of impulse noise with higher computational efficiency. The local noise variance is: Local mean within the 3×3 window ; Bilateral filtering is: in Spatial weights, , For grayscale weights, ; Noise adaptive adjustment: The larger the noise variance, the stronger the grayscale weight constraint. Median filtering is: Median filtering is suitable for impulse noise and is computationally efficient; The dynamic switching mechanism of differential filtering perfectly balances the contradiction between noise suppression and edge preservation. According to actual tests, the peak signal-to-noise ratio (PSNR) of the denoised image is consistently improved by ≥3dB compared with the basic preprocessing result, providing high-quality image input for subsequent micro-domain detection.

[0018] For the denoised image, segment the known structures that are considered as interference regions to obtain micro-domain candidate regions; Specifically, the images.roi.Freehand function in MATLAB was used to manually and finely segment the known structures to generate baseline contour data, which was used to label the known structures in astrocytes. The known structures included the visible cell bodies, main branches, and terminal feet of astrocytes. After removing the known structures, the remaining image regions were designated as micro-domain candidate regions. The known structures were precisely removed as interference regions, thus significantly reducing the complexity of subsequent analysis by eliminating known non-target structures. A noise mask is generated based on the noise model. This mask is then used to remove high-noise regions from the micro-domain candidate regions, resulting in an updated micro-domain candidate region. The specific process is as follows: According to the noise model Generate a noise mask and set the threshold to 0. : The threshold setting is used to distinguish regions that are significantly higher than the global noise level; Based on the noise mask, high-noise regions are automatically eliminated from the micro-domain candidate regions. These regions are deemed to have no value for micro-domain detection due to insufficient signal reliability. For the preserved region, the bwboundaries function in MATLAB is used to detect active pixels and obtain seed points for connected component construction; the boundary pixels with gray-level abrupt changes are accurately identified to provide seed points for connected component construction. Based on the seed points, connected components are grown to obtain updated micro-domain candidate regions. During the connected component growth process, noise constraints are set; the noise constraint condition is that the noise variance of all pixels within the connected component is ≤ ,Right now: in For connected components, To ensure the overall signal-to-noise ratio of the region meets the standard for connected pixels, a rigorous noise level screening process is used to ensure that the candidate regions have the basic signal-to-noise ratio requirements. The area range of the final selected micro-domain candidate regions can be flexibly adjusted according to specific experimental needs, which not only ensures the biological rationality of the region, but also reserves a sufficient sample size for subsequent analysis.

[0019] The noise model is enhanced to generate an enhanced noise parameter matrix, which is then co-calculated with the gray-level statistical features of the micro-domain candidate region to generate an adaptive threshold in the subcellular segmentation algorithm. The micro-domain candidate region is then subcellularly segmented based on the adaptive threshold to obtain the micro-domain, such as... Figure 12 As shown; Specifically, for the noise model Calculate the value within the window by sliding window The standard deviation is used to construct the enhanced noise parameter matrix. : in The local mean within a 5×5 sliding window is used to strengthen the boundary by amplifying the local gradient. In this embodiment, the noise variance matrix obtained from noise modeling is used. As input, the standard deviation image generation method in the additional processing module is used to calculate the value within the window using a 5×5 sliding window. The standard deviation is used to obtain the enhanced noise parameter matrix. By amplifying the spatial gradient change of noise variance, the originally blurred subcellular boundary region (usually accompanied by abrupt changes in noise features) is significantly enhanced, providing clearer boundary guidance for the segmentation algorithm. Based on the enhanced noise parameter matrix By combining the gray-scale statistical features of the micro-domain candidate region for collaborative calculation, an adaptive threshold is generated. ,in The brightness threshold. The mean gray level of the candidate micro-domain is denoted as . Let be the standard deviation of the grayscale values ​​of the candidate micro-domains. The noise adjustment weighting coefficient ranges from 0.3 to 0.5; the gray-scale statistical characteristics include: mean. Standard deviation ; Determine whether the brightness of a pixel in the micro-domain candidate region is greater than the brightness threshold. If it is, then the pixel is determined to be a micro-domain; otherwise, the pixel is determined not to be a micro-domain. Pixels identified as micro-domains are used to form micro-thresholds; The dynamic calculation mechanism of the segmentation threshold is the core of achieving adaptive spatial segmentation. The adaptive threshold retains the basic statistical logic of the traditional mean and 0.5 standard deviation, while innovatively introducing a noise parameter adjustment term; in edge regions with sharp noise gradients, Large values ​​lead to threshold This reduces the number of edge pixels, thus preventing them from being misidentified as background and avoiding missed edge pixel classification; in the internal uniform region, A smaller threshold value keeps the threshold relatively stable, ensuring overall consistency in segmentation. The dynamic adjustment mechanism improves segmentation accuracy compared to traditional fixed threshold methods, especially significantly enhancing the ability to capture subcellular microstructures.

[0020] In addition, multidimensional validation analysis was performed on the obtained microthreshold regions to ensure their biological validity. The specific process is as follows: The primary verification step combines the analysis results of the patented calcium response feature extraction subsystem: calculate the relative fluorescence change value (ΔF / F0) of the micro-domain region, and retain only the region where ΔF / F0 > 0.2 and the duration exceeds 5 frames - this standard is determined based on a large amount of neuroscience experimental data to ensure that the selected micro-domains have significant calcium signal response activity. Simultaneously, the patented multi-view interactive subsystem enables the correlation verification of spatial and temporal features: it links the three-dimensional spatial location information of the micro-domain with the curve characteristics of its fluorescence intensity changing over time to check whether the spatial distribution conforms to known cellular physiological structural patterns and whether the time curve exhibits a typical calcium signal fluctuation pattern. This dual verification mechanism checks both biological activity (calcium response intensity) and structural integrity (spatial location rationality), ensuring that the final output micro-domain dataset meets both morphological analysis requirements and possesses clear functional significance, providing highly reliable foundational data for subsequent neural cell function research.

[0021] The calcium response feature extraction and analysis subsystem is used to detect fluorescence curves of calcium responses and extract calcium response event parameter features from images segmented into cell bodies, main branches, and microdomains using a hybrid segmentation strategy. The interface of the calcium response feature extraction and analysis subsystem includes features such as... Figure 13 As shown; The specific process is as follows: Preprocessing of astrocyte calcium signal data included: averaging the fluorescence intensity of the selected region of interest (ROI) from each frame of the video during analysis, and constructing a mean fluorescence intensity curve over time. ; To eliminate random fluctuations in the data, A smoothing window of size 20 is used for processing, and the smoothed data sequence is labeled as follows. ; The fluorescence intensity data at each time point in the data sequence were standardized using the following normalization formula: in, Indicates at a point in time The smoothed fluorescence intensity value, and As the baseline value, in this embodiment it is set to be based on the time point The 25th percentile within a 300-second sliding window starting from the given point: in, For at a certain point in time of Value, i.e., fluorescence intensity index, Represents the 25th quantile. It is the first Fluorescence intensity values ​​of ROIs in the frame after smoothing, and The range of values ​​is arrive Between these, a normalized fluorescence intensity index that changes over time can be obtained using this formula as a baseline value for pixel fluorescence intensity, and the relative fluorescence change curve of the corresponding ROI can be extracted based on the obtained fluorescence intensity index. Based on the relative fluorescence change curve, the features of calcium reaction event parameters were extracted. These features include: amplitude, duration, area, rise time, fall time, rise slope, fall slope, half width at half maximum (WHM), number of peaks, number of troughs, trough value being half the peak value, and duration of peak. Amplitude (AMP) represents the distance between the maximum peak value and the initial value; Duration (Dura) represents the time from the start point to the end point; Area (area) represents the area enclosed by the fluorescence signal from the starting point to the ending point; Rise time (UT) represents the time from the starting point to the first peak point; Descent time (DT) represents the time from the last peak to the end point; The upward slope (US) represents the slope between the starting point and the first peak point. The descent slope (DS) represents the slope from the last peak point to the end point. Half width (HW) represents the width at a position halfway from the maximum peak value; Number of peaks (NP) represents the number of peak values ​​within a calcium signal that reach the signal level. Number of valleys (NV) represents the number of valleys between peaks within a calcium signal; A trough value is half the peak value (bump, Bump), indicating whether there is a trough value within the calcium signal that is less than half the size of the adjacent highest peak; Main Peak Duration (MPDura) refers to the duration of the highest peak within the calcium signal.

[0022] The parametric features of the above 12 kinetics are calculated from each calcium response event, and the structured parametric features of each subcellular structure can be viewed in the calcium signal feature region. Random forests, with swarm intelligence as their core idea, construct diverse decision tree forests through double randomness, and perform well in most scenarios. Despite shortcomings such as high computational cost and weak interpretability, their stability and generalization ability still make them a classic tool in the field of machine learning. This solution can ensure the generalization ability of the model, enabling it to more accurately predict calcium signals in unknown data, thus playing a greater role in practical applications. Therefore, in this embodiment, a random forest model is used to detect whether a calcium signaling activity event occurs in the ROI. A positive label indicates the presence of a calcium signal, and a negative label indicates the absence of a calcium signal. The lowest peak value was selected as the peak threshold from the peak values ​​of all calcium signal relative fluorescence change curves to reduce the false detection rate of negative samples. Based on expert testing, a peak value exceeding the threshold was considered positive, and vice versa. In this experiment, data were collected from 9 mice. Under this threshold, there were 1134 negative samples and 265 positive samples.

[0023] To construct a calcium signal detection model based on random forest, the experiment was conducted by grouping mice into groups and extracting the features of the 12 calcium response events mentioned above as model inputs. A 9-fold cross-validation method was adopted, in which the data of one mouse was used as the test set and the data of all other mice were used as the training set to train the model. This mouse-based grouping method helps to avoid feature crossing problems, thereby reducing the risk of model overfitting and local optima.

[0024] AUC (Area Under the Curve) is an important metric for evaluating model performance, especially in classification problems. In this experiment, the mean AUC obtained from 9-fold cross-validation was 0.984. Figure 14 As shown, this demonstrates that the model exhibits excellent performance and stability in detecting calcium signals in astrocytes. This method ensures the model's generalization ability, enabling it to more accurately predict calcium signals in unknown data, thus playing a greater role in practical applications.

[0025] Sort by ROI distance The spatial location of each ROI is statistically analyzed, and the relative distances between each ROI are calculated based on these locations. This relative distance serves as a reference for sorting ROIs by distance relative to the geometric center of each other ROI. Calculate its distance from the center point.

[0026] The multi-view interactive subsystem is used for ROI display, and the calcium fluorescence imaging area, the relative fluorescence change curve display area, and the calcium reaction event parameter area are interconnected. This subsystem adopts an event-driven multi-view synchronous architecture, consisting of a visualization analysis module, a data linkage module, and a user interaction control module (see system architecture diagram for details). By establishing a centralized event bus mechanism, real-time data interaction among the following core components is achieved: Interactive Feature Object Association: After selecting the corresponding ROI number in the ROI management area, the spatial location outline of the ROI is highlighted in the main view; the ROI color in the main view corresponds to the ROI curve color in the waveform analysis view for ROIs with the same number.

[0027] This approach first acquires calcium imaging video and preprocesses it to obtain feature-enhanced images, making the structure of astrocytes more obvious, which facilitates subsequent subcellular segmentation and helps improve segmentation accuracy. Then, a hybrid detection strategy is adopted to segment the image into subcellular structures. This involves using both manual annotation and adaptive thresholding. For micro-domain segmentation, the adaptive threshold introduces a noise parameter adjustment term. In edge regions with sharp noise gradients, the threshold is lowered to avoid edge pixels being misclassified as background. In uniform internal regions, the threshold remains relatively stable to ensure overall segmentation consistency. This dynamic adjustment mechanism improves segmentation accuracy compared to traditional fixed threshold methods, especially significantly enhancing the ability to capture subcellular micro-structures. Furthermore, it eliminates the need for manual settings, saving labor costs and setup time. Finally, the segmented images are subjected to fluorescence curve detection of calcium reaction and extraction of calcium reaction event parameter features, thereby reducing reliance on manual labor, reducing analysis time, and saving labor and time costs.

[0028] In summary, this scheme can segment astrocytes of different morphologies, distinguish subcellular structures, and has an adaptive threshold. It also performs calcium response detection and analysis to improve segmentation accuracy and analysis efficiency while reducing labor and time costs.

[0029] Example 2 This embodiment is basically the same as the above embodiment, except that: a two-dimensional cell two-photon image analysis method is used, employing the above-mentioned two-dimensional cell two-photon image analysis system; The specific process is as follows: The image preprocessing subsystem acquires calcium imaging video and performs preprocessing to obtain feature-enhanced images. In this embodiment, the main body of the cell is highlighted to make its structure clearer, such as... Figure 15 As shown; The subcellular segmentation subsystem allows for manual segmentation of the subcellular structure of astrocytes, drawing structures such as main branches, cell bodies, and terminal feet. Figure 16 As shown; Through a subcellular segmentation subsystem, a noise model combined with a spatial domain is used for multi-level screening, and adaptive threshold segmentation is employed to detect micro-domains in astrocyte fluorescence images, such as... Figure 17 As shown; The calcium response feature extraction and analysis subsystem was used to detect the fluorescence curves of the calcium response in the aforementioned subcellular structural regions, such as... Figure 18 As shown; simultaneously, extract the calcium reaction event parameter features of the corresponding region, such as Figure 19 As shown; The multi-view interactive subsystem links the calcium fluorescence imaging area, the relative fluorescence change curve display area, and the feature parameter area together. Selecting the corresponding sequence number displays the specific location of the ROI in the fluorescence image area, and the color corresponds to the color of the relative fluorescence change curve. Figure 20 As shown.

[0030] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics 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. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A two-dimensional cellular two-photon image analysis system, characterized in that, include: The image preprocessing subsystem is used to acquire calcium imaging video and perform preprocessing to obtain feature-enhanced images; The subcellular segmentation subsystem is used to segment images into subcellular structures using a hybrid detection strategy, which consists of manual annotation and adaptive thresholding. The calcium reaction feature extraction and analysis subsystem is used to detect the fluorescence curve of the calcium reaction and extract the calcium reaction event parameter features from the segmented images.

2. The two-dimensional cellular two-photon image analysis system according to claim 1, characterized in that, The method employs a hybrid detection strategy to perform subcellular structure segmentation of the image, including: For single cells, main branches, and terminal foci, ROIs are drawn manually. Morphological opening operations combined with adaptive threshold segmentation are used to extract cell bodies. A noise model combined with a spatial domain is used for multi-level screening, and micro-domains are extracted by combining adaptive threshold segmentation.

3. The two-dimensional cellular two-photon image analysis system according to claim 2, characterized in that, The morphological opening operation, combined with adaptive threshold segmentation, extracts the cell body, including: Obtain the plotted single-cell ROI; The maximum inter-class variance is calculated for the ROI of a single cell, the image is binarized based on the maximum inter-class variance, and a method of fusing local standard deviation enhancement is used to obtain the single cell ROI as the target part. The binarized target part is subjected to an opening operation to disconnect the cell body from the non-cell body part, and the largest connected component is selected as the first mask. Based on the first mask, adaptive threshold segmentation is performed to obtain the cell body.

4. The two-dimensional cellular two-photon image analysis system according to claim 3, characterized in that, The step of performing adaptive threshold segmentation based on the first mask to obtain the cell body includes: Calculate the centroid of the first mask and the average distance from the centroid to the edge of the connected domain; Using the average distance as the radius and the centroid as the center, fit a circle as the second mask; Calculate the intersection of the first mask and the second mask, and use the common part of the intersection as the cell body for final automatic detection.

5. The two-dimensional cellular two-photon image analysis system according to claim 2, characterized in that, The process involves employing a noise model combined with a spatial domain for multi-level screening, followed by adaptive threshold segmentation to extract micro-domains, including: Construct a noise model; A spatial domain joint denoising model is used to perform differential denoising on the original image to obtain the denoised image; the original image is an image of the ROI with single cells, main branches, and terminal feet already drawn; For the denoised image, segment the known structures that are considered as interference regions to obtain micro-domain candidate regions; A noise mask is generated based on the noise model. The noise mask is then used to remove high-noise regions from the micro-domain candidate regions, and the micro-domain candidate regions are updated. The noise model is enhanced to generate an enhanced noise parameter matrix, which is then co-calculated with the gray-level statistical features of the micro-domain candidate region to generate an adaptive threshold. Based on the adaptive threshold, the micro-domain candidate region is sub-cell segmented to obtain the micro-domain.

6. The two-dimensional cellular two-photon image analysis system according to claim 5, characterized in that, The noise model is a Gaussian noise model, characterized by a noise spatial distribution matrix: ； in It follows a standard normal distribution. The noise mean. This represents the noise variance.

7. The two-dimensional cellular two-photon image analysis system according to claim 6, characterized in that, The adaptive threshold in the sub-cell segmentation algorithm includes: Calculate the noise model within the window using a sliding window. The standard deviation is used to construct the enhanced noise parameter matrix. : ； in This represents the local mean within the sliding window; Based on the enhanced noise parameter matrix An adaptive threshold is obtained by combining the gray-scale statistical features of the micro-domain candidate region with collaborative calculation. ； in The mean gray level of the candidate micro-domain is denoted as . Let be the standard deviation of the grayscale values ​​of the candidate micro-domains. These are noise adjustment weighting coefficients; the gray-scale statistical characteristics include: mean. Standard deviation .

8. The two-dimensional cellular two-photon image analysis system according to claim 1, characterized in that, The fluorescence curve detection and extraction of calcium reaction event parameters for the calcium reaction include: The fluorescence intensity of the ROIs obtained from the segmentation of each frame in the video is averaged to construct a mean fluorescence intensity curve that changes over time. ; Fluorescence intensity mean curve A smoothing window is used for processing, and the smoothed data sequence is labeled as follows: ; The fluorescence intensity data at each time point in the data sequence were standardized: in, Indicates at a point in time The smoothed fluorescence intensity value, Baseline value: in, For at a certain point in time of value, Represents the 25th quantile. It is the first Fluorescence intensity values ​​of ROIs in the frame after smoothing, and The range of values ​​is arrive Between these values, the fluorescence intensity index is used as an indicator, and the relative fluorescence change curve of the corresponding ROI is extracted based on the obtained fluorescence intensity index; Based on the relative fluorescence change curve, the features of calcium reaction event parameters were extracted.

9. The two-dimensional cellular two-photon image analysis system according to claim 1, characterized in that, Also includes: The multi-view interactive subsystem is used for ROI display, and the calcium fluorescence imaging area, the relative fluorescence change curve display area, and the calcium reaction event parameter area are interconnected.

10. A two-dimensional cellular two-photon image analysis method, characterized in that, The two-dimensional cell two-photon image analysis system according to any one of claims 1-9 above.

Images