Conjunctival sac pathogen detection method and system using fluorescent staining image analysis

Abstract

This invention discloses a method and system for detecting pathogens in the conjunctival sac using fluorescence staining image analysis, belonging to the field of medical image processing technology. The method includes: acquiring multi-wavelength fluorescence images and performing inter-channel differential operations to construct a spectral response surface; identifying regions of abrupt curvature changes to determine candidate pathogen locations; extracting radial fluorescence intensity distribution curves; calculating attenuation gradients and transition band width matching rules to determine pathogen types; acquiring temporal fluorescence intensity; screening locations where the intensity change rate exceeds an activity threshold as active pathogen locations; identifying the diffusion axis based on spatial distribution; calculating the spatial density gradient based on statistical density changes; coupling the intensity change rate to calculate the diffusion rate and determine the warning region; identifying abnormal pixels within the warning region; verifying continuous growth characteristics through temporal tracking; confirming and marking early infection areas. This invention improves detection accuracy and timeliness, enabling early warning.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a method and system for detecting pathogens in the conjunctival sac using fluorescence staining image analysis. Background Technology

[0002] Existing fluorescence detection methods typically employ a single-wavelength excitation source, identifying pathogens by observing the morphological characteristics of fluorescently stained samples under a microscope. However, the fluorescence characteristics of different pathogen types overlap, making it difficult to accurately distinguish pathogen types based solely on single-wavelength fluorescence images. Current methods primarily focus on static image analysis, lacking dynamic monitoring of pathogen activity, thus failing to effectively identify pathogens at different growth stages and easily leading to false positive or false negative results.

[0003] In pathogen spread monitoring, traditional methods rely primarily on manual observation and experience, lacking quantitative assessment tools for spread rates. Currently available gene testing methods are expensive, making them unaffordable for the average person. Laboratory bacterial and fungal cultures take long times (more than 3 days), and the positive rate of conjunctival sac culture is extremely low. Therefore, clinicians typically rely on subjective assessments based on changes in the infected area, a method heavily influenced by individual experience and unsuitable for early warning. Especially in the early stages of infection, when pathogen numbers are low and distribution is dispersed, manual identification can easily miss potential areas of infection spread, delaying optimal treatment.

[0004] Current image analysis techniques for processing fluorescence images often employ traditional methods such as thresholding or morphological filtering. These methods require high image quality, and their detection accuracy drops significantly when the fluorescence signal is weak or the background noise is high. Traditional image analysis methods lack in-depth analysis of the spatial distribution characteristics of pathogens and fail to fully utilize the spatial statistical characteristics of pathogens, such as directionality and aggregation, exhibited during their spread, resulting in insufficient predictive ability for infection development trends. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting pathogens in the conjunctival sac using fluorescent staining image analysis, aiming to at least solve one of the technical problems existing in the prior art.

[0006] The technical solution of this invention is: a method for detecting pathogens in the conjunctival sac by fluorescence staining image analysis, comprising the following steps: Fluorescence images of conjunctival sac samples at multiple wavelengths were acquired. Inter-channel difference operations were performed on the fluorescence images at different wavelengths to construct spectral response surfaces. Regions of curvature abrupt change on the spectral response surfaces were identified to determine candidate locations of pathogens. Fluorescence intensity distribution curves are extracted radially from the candidate pathogen location, and the attenuation gradient and transition band width are calculated to form boundary feature parameters. The boundary feature parameters are then matched with preset pathogen type discrimination rules to determine the pathogen type. The time-series fluorescence intensity of pathogen candidate sites is collected, the rate of intensity change between adjacent time points is calculated, and pathogen candidate sites with an intensity change rate exceeding the activity threshold are selected as active pathogen sites. Based on the spatial distribution of active pathogen locations, the diffusion axis is identified, the spatial density gradient is calculated by statistically analyzing the density changes of active pathogen locations along the diffusion axis, and the rate of intensity change is coupled with the spatial density gradient to calculate the pathogen diffusion rate and determine the warning area. Abnormal pixels that deviate from the background baseline within the warning area are identified. The abnormal pixels are then tracked over time to verify their continuous growth characteristics. Abnormal pixels that meet the continuous growth characteristics are identified as early infection areas and marked.

[0007] Fluorescence images of conjunctival sac samples at multiple wavelengths were acquired. Inter-channel difference operations were performed on the fluorescence images at different wavelengths to construct spectral response surfaces. Regions of curvature abrupt changes on the spectral response surfaces were identified to determine candidate pathogen locations, including: The conjunctival sac sample was sequentially irradiated with a multi-band excitation light source, and fluorescence images were acquired at each excitation wavelength to obtain fluorescence images at multiple wavelengths. Fluorescence images at adjacent wavelengths are selected for inter-channel difference calculation to calculate the fluorescence intensity difference between the adjacent wavelength fluorescence images and generate inter-channel difference data of fluorescence images. The differential data between fluorescence image channels are arranged in wavelength order to construct a spectral response surface. The fluorescence intensity gradient value and curvature value of each point on the spectral response surface are calculated. Based on fluorescence intensity gradient values ​​and curvature values, regions on the spectral response surface whose gradient change rate exceeds a preset change threshold and whose curvature values ​​change abruptly are identified as candidate pathogen regions. Fluorescence intensity distribution features are extracted within the candidate pathogen region, and the distribution parameters and morphological parameters of the fluorescence intensity distribution features are calculated. Based on distribution and morphological parameters, feature matching is performed on candidate regions of pathogens, and regions that meet the preset matching conditions are identified as candidate locations of pathogens.

[0008] Fluorescence intensity distribution curves are extracted radially from the candidate pathogen location. The attenuation gradient and transition band width are calculated to form boundary feature parameters. These boundary feature parameters are then matched with preset pathogen type discrimination rules to determine the pathogen type, including: A circular sampling area is constructed with the candidate pathogen location as the center, and the sampling direction is set along the circumference. Fluorescence intensity values ​​are collected in the sampling direction to generate a fluorescence intensity distribution curve. The fluorescence intensity distribution curve was smoothed, the trend of fluorescence intensity with radial distance was calculated, and the peak position of fluorescence intensity was extracted. The rate of change of fluorescence intensity from the peak position of fluorescence intensity to the sampling boundary is calculated along the radial direction to generate a fluorescence intensity attenuation gradient; The points of maximum and minimum rate of change are identified on the fluorescence intensity decay gradient, and the radial distance between the points of maximum and minimum rate of change is determined as the transition band width. The fluorescence intensity attenuation gradient and the transition band width are combined to generate boundary feature parameters; The boundary feature parameters are matched with the preset pathogen type discrimination rules, the matching scores of the boundary feature parameters under each pathogen type discrimination rule are calculated and normalized, a pathogen type probability distribution is generated, and the pathogen type is determined based on the maximum probability value in the pathogen type probability distribution.

[0009] The temporal fluorescence intensity of candidate pathogen sites is collected, the rate of intensity change between adjacent time points is calculated, and candidate pathogen sites with an intensity change rate exceeding the activity threshold are selected as active pathogen sites, including: Fluorescence images are continuously acquired at preset time intervals, and the fluorescence intensity values ​​of pathogen candidate locations at continuous time points are extracted to construct the temporal fluorescence intensity of pathogen candidate locations. Adaptive filtering is applied to the temporal fluorescence intensity at candidate pathogen sites to generate filtered temporal fluorescence intensity. Calculate the rate of intensity change between adjacent time points in the filtered time-series fluorescence intensity to construct a sequence of intensity change rates at candidate pathogen locations; Dynamic change features are extracted from the intensity change rate sequence of pathogen candidate sites to form activity feature indicators of pathogen candidate sites; The activity characteristic indicators of pathogen candidate sites are compared with preset activity thresholds to screen out pathogen candidate sites whose activity characteristic indicators exceed the preset activity thresholds. Spatial correlation analysis was performed on the screened pathogen candidate locations to identify pathogen candidate locations with spatial clustering as active pathogen locations.

[0010] Based on the spatial distribution of active pathogen locations, the main diffusion axis is identified. The spatial density gradient is calculated by statistically analyzing the density changes of active pathogen locations along the main diffusion axis. The rate of intensity change is coupled with the spatial density gradient to calculate the pathogen diffusion rate and determine the warning area, including: Based on the spatial distribution of the location of the active pathogen, the spatial coordinates of the location of the active pathogen are extracted, an adjacency graph is constructed, the connected link with the most connected nodes in the adjacency graph is identified, and the extension direction of the connected link is determined as the diffusion axis. A continuous measurement segment is set up along the main diffusion axis. The number of active pathogen locations in each measurement segment is counted to form a location density sequence. The inflection point of the sign conversion of the density difference between adjacent measurement segments in the location density sequence is identified. The measurement segment between the inflection point and the end is extracted as the diffusion wavefront segment. The density difference between adjacent measurement segments in the diffusion wavefront segment is calculated as the spatial density gradient. The intensity change rate of the active pathogen location within the diffusion wavefront is obtained, and the intensity change rate sequence is formed by sorting the active pathogen locations according to their projection positions on the diffusion principal axis. The increasing trend coefficient of the intensity change rate sequence is then calculated. Calculate the consistency between the increasing trend coefficient and the spatial density gradient, determine the coupling weight based on the consistency, and couple the intensity change rate with the spatial density gradient to generate the pathogen diffusion rate. Starting from the boundary position of the diffusion wavefront, the propagation distance is calculated along the main diffusion axis according to the pathogen diffusion rate, and the spatial area covered by the propagation distance is determined as the warning area.

[0011] Within the warning area, identify anomalous pixels that deviate from the background baseline. Perform time-series tracking to verify the continuous growth characteristics of these anomalous pixels. Identify anomalous pixels that meet the continuous growth characteristics as early infection areas and mark them accordingly. Extract the fluorescence intensity of each pixel within the warning area, select a reference area outside the boundary of the warning area where no active pathogens appear, statistically analyze the fluorescence intensity distribution characteristics of pixels within the reference area, and construct a background baseline; Calculate the degree of deviation between the fluorescence intensity of each pixel within the warning area and the background baseline, and filter out pixels whose deviation exceeds the preset fluctuation range as abnormal pixels; Fluorescence intensity is continuously collected for abnormal pixels at preset time intervals to construct a time-series fluorescence intensity trajectory for abnormal pixels. Continuous rising segments are extracted from the time-series fluorescence intensity trajectory, and the duration and cumulative increase of the continuous rising segments are calculated. Abnormal pixels whose duration and cumulative increase simultaneously meet the growth determination criteria are identified as abnormal pixels that meet the continuous growth characteristics. Spatial clustering is performed on abnormal pixels that exhibit continuous growth characteristics. Connected regions composed of spatially adjacent abnormal pixels that exhibit continuous growth characteristics are identified, and these connected regions are confirmed as early infection regions. Contour annotation information is generated on the boundaries of the early infection regions.

[0012] This invention provides a conjunctival sac pathogen detection system based on fluorescence staining image analysis, the system comprising: The image processing module is used to acquire fluorescence images of conjunctival sac samples at multiple wavelengths, perform inter-channel difference operations on fluorescence images at different wavelengths to construct spectral response surfaces, and identify curvature abrupt change regions on the spectral response surfaces to determine candidate locations of pathogens. The type discrimination module is used to extract the fluorescence intensity distribution curve radially with the pathogen candidate location as the center, calculate the attenuation gradient and transition band width to form boundary feature parameters, and match the boundary feature parameters with the preset pathogen type discrimination rules to determine the pathogen type; The activity recognition module is used to collect the time-series fluorescence intensity of pathogen candidate sites, calculate the rate of intensity change between adjacent time points, and screen pathogen candidate sites whose rate of intensity change exceeds the activity threshold as active pathogen sites. The diffusion analysis module is used to identify the main diffusion axis based on the spatial distribution of the location of active pathogens, calculate the spatial density gradient by statistically analyzing the density changes of active pathogens along the direction of the main diffusion axis, and couple the intensity change rate with the spatial density gradient to calculate the pathogen diffusion rate and determine the warning area. The region labeling module is used to identify abnormal pixels that deviate from the background baseline within the warning area, perform time-series tracking to verify the continuous growth characteristics of the abnormal pixels, and identify and label the abnormal pixels that meet the continuous growth characteristics as early infection areas.

[0013] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0014] One technical solution provided in this embodiment of the invention is a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the steps in any of the aforementioned methods.

[0015] This invention constructs a spectral response surface through inter-channel differential operations on multi-wavelength fluorescence images and utilizes curvature abrupt changes to locate pathogens, significantly improving the accuracy of pathogen identification and effectively reducing misjudgments caused by overlapping fluorescence features of different pathogens. By extracting the attenuation gradient and transition band width of the radial fluorescence intensity distribution as boundary feature parameters, quantitative discrimination of pathogen types is achieved, overcoming the shortcomings of strong subjectivity and poor repeatability of manual observation. Using time-series fluorescence intensity monitoring and calculating the intensity change rate, it can accurately distinguish between active pathogens and inactive substances, avoiding false positive results easily produced by static image analysis. Based on the spatial distribution of active pathogen locations, the diffusion axis is identified, and the diffusion rate is calculated by coupling the intensity change rate with the spatial density gradient, achieving a quantitative assessment of the infection development trend. Within the warning area, time-series tracking verifies the continuous growth characteristics of abnormal pixels, enabling the identification of potential diffusion areas in the early stages of infection, achieving an early warning function, and facilitating timely intervention to prevent further spread of infection. This invention improves the timeliness and accuracy of detection, providing an effective technical means for the diagnosis and monitoring of conjunctival sac pathogen infections. Attached Figure Description

[0016] Figure 1 A flowchart of a method for detecting pathogens in the conjunctival sac using fluorescence staining image analysis provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the conjunctival sac pathogen detection system for fluorescence staining image analysis, according to an embodiment of the present invention. Detailed Implementation

[0017] like Figure 1 As shown, Figure 1 A flowchart of a method for detecting pathogens in the conjunctival sac using fluorescent staining image analysis provided in an embodiment of the present invention, the method comprising the following steps: Step 101: Obtain fluorescence images of conjunctival sac samples at multiple wavelengths, perform inter-channel difference operations on fluorescence images at different wavelengths to construct spectral response surfaces, and identify curvature abrupt change regions on the spectral response surfaces to determine candidate pathogen locations.

[0018] In some embodiments of the present invention, step 101 may specifically include the following sub-steps: Sub-step 1011: The conjunctival sac sample is sequentially irradiated with a multi-band excitation light source, and fluorescence images at each excitation wavelength are acquired to obtain fluorescence images at multiple wavelengths. Sub-step 1012: Select fluorescence images of adjacent wavelengths and perform inter-channel difference operation to calculate the fluorescence intensity difference value between the fluorescence images of adjacent wavelengths and generate fluorescence image inter-channel difference data. Sub-step 1013: Arrange the differential data between fluorescence image channels in wavelength order to construct a spectral response surface, and calculate the fluorescence intensity gradient value and curvature value at each point on the spectral response surface; Sub-step 1014: Based on the fluorescence intensity gradient value and curvature value, identify regions on the spectral response surface where the gradient change rate exceeds a preset change threshold and the curvature value changes abruptly as pathogen candidate regions. Sub-step 1015: Extract fluorescence intensity distribution features within the pathogen candidate region and calculate the distribution parameters and morphological parameters of the fluorescence intensity distribution features; Sub-step 1016: Perform feature matching on the pathogen candidate region based on distribution parameters and morphological parameters, and determine the region that meets the preset matching conditions as the pathogen candidate location.

[0019] First, a tunable laser with a wavelength range of 350-550 nm was used as the excitation source. An excitation wavelength was generated every 5 nm using a step-wavelength adjustment, resulting in a total of 41 different excitation wavelengths. The conjunctival sac samples were pre-stained with fluorescent dyes.

[0020] When obtaining conjunctival sac samples, the inner side of the patient's conjunctival sac was gently wiped with a sterile cotton swab, the sample was smeared onto a glass slide, a mixed fluorescent dye solution was sprayed on, and the sample was incubated at room temperature for 10 minutes before observation using a fluorescence microscope. The samples were sequentially irradiated with an excitation light source, with each excitation wavelength irradiated for 200 ms. A CCD camera simultaneously acquired fluorescence images, with the image resolution set to 1.5 μm / pixel.

[0021] During the inter-channel difference calculation of adjacent wavelength fluorescence images, pixel-level difference calculations are performed on each pair of adjacent wavelength fluorescence images. The original fluorescence images are preprocessed, including background noise removal and brightness normalization. Background noise removal involves sampling the background area outside the sample region, calculating the average background brightness value and standard deviation, and marking pixels exceeding three times the standard deviation of the average value as valid signals. Brightness normalization normalizes each image by referencing the fluorescence intensity of a standard fluorograph, eliminating the influence of light source intensity fluctuations.

[0022] During the generation of inter-channel differential data, the image at wavelength n and the image at wavelength (n+1) are differentially analyzed at the pixel level, generating 40 differential images. In the differential images, positive values ​​indicate an increase in fluorescence intensity with increasing wavelength, negative values ​​indicate a decrease in fluorescence intensity with increasing wavelength, and values ​​near zero indicate that the fluorescence change in that wavelength range is not significant. The set of differential images constitutes a spectral response surface, which corresponds to the two-dimensional distribution of the samples in the spatial dimension and the rate of change of fluorescence intensity in the wavelength dimension.

[0023] After constructing the spectral response surface, the fluorescence intensity gradient along the wavelength dimension is calculated for each spatial location. The gradient is calculated using the central difference method, where the gradient value for a wavelength position λ is the average of the differences between two adjacent wavelength points. The curvature value is calculated based on the rate of change of the gradient value, representing the degree of curvature of the spectral response curve at that point. Regions with high curvature values ​​typically correspond to locations where fluorescence properties change significantly, and are key features for identifying different pathogens.

[0024] Pathogen candidate region identification is based on two indicators: gradient rate of change and curvature value. Regions with a gradient rate of change exceeding a preset threshold of 0.15 and a curvature value greater than 0.25 are marked as potential pathogen regions. These regions typically exhibit fluorescence characteristic change patterns that are significantly different from the background. Bacterial pathogens usually show peak gradient rate of change in the 420-460 nm wavelength range, while viral pathogens show characteristic curvature changes in the 480-520 nm range.

[0025] Within the defined candidate region, fluorescence intensity distribution features are extracted, including the statistical distribution and spatial morphological features of pixel fluorescence intensity within the region. Statistical distribution features include average fluorescence intensity, standard deviation, skewness, and kurtosis, describing the central tendency and dispersion of fluorescence intensity within the region. Spatial morphological features include region area, perimeter, circularity, and texture complexity, describing the geometric shape of the candidate region.

[0026] The feature matching process uses a pre-established pathogen feature library for comparison. The feature library contains fluorescence spectral feature templates of common ocular pathogens, such as Staphylococcus aureus, which is represented by a region area of ​​30-50 μm. 2 The circularity is 0.85-0.95, and it has a fluorescence intensity peak at 450 nm; herpes simplex virus shows a region area of ​​10-25 μm. 2 The region exhibits an irregular shape with a circularity of 0.65-0.80 and a characteristic fluorescence peak at 500 nm. Matching accuracy is calculated using a weighted Euclidean distance, with weighting coefficients dynamically adjusted based on the discriminative power of different features. Candidate regions with a matching accuracy exceeding a preset threshold of 0.85 are identified as the final pathogen locations.

[0027] This invention achieves rapid and accurate detection of pathogens in the conjunctival sac by combining multi-wavelength fluorescence imaging technology with spectral response surface analysis. Utilizing the differential characteristics of different pathogens in their fluorescence spectra, this invention constructs a spectral response surface through inter-channel difference operations and identifies pathogen locations based on curvature abrupt changes, avoiding the drawbacks of long culture cycles and complex operations in traditional methods. Combining fluorescence intensity distribution characteristics with morphological parameter matching analysis further improves the specificity and sensitivity of the detection. This invention can complete the detection process in a short time, reducing patient waiting time and providing important evidence for the early diagnosis and treatment of ophthalmic infectious diseases.

[0028] Step 102: Extract the fluorescence intensity distribution curve radially from the candidate pathogen location, calculate the attenuation gradient and transition band width to form boundary feature parameters, and match the boundary feature parameters with the preset pathogen type discrimination rules to determine the pathogen type.

[0029] In some embodiments of the present invention, step 102 may specifically include the following sub-steps: Sub-step 1021: Construct a circular sampling area centered on the candidate pathogen location, set the sampling direction along the circumference, collect fluorescence intensity values ​​in the sampling direction, and generate a fluorescence intensity distribution curve; Sub-step 1022: Smooth the fluorescence intensity distribution curve, calculate the variation trend of fluorescence intensity with radial distance, and extract the peak position of fluorescence intensity; Sub-step 1023: Calculate the rate of change of fluorescence intensity from the peak position of fluorescence intensity to the sampling boundary along the radial direction to generate the fluorescence intensity attenuation gradient; Sub-step 1024: Identify the maximum and minimum rate of change points on the fluorescence intensity decay gradient, and determine the radial distance between the maximum and minimum rate of change points as the transition band width. Sub-step 1025: Combine the fluorescence intensity attenuation gradient with the transition band width to generate boundary feature parameters; Sub-step 1026: Match the boundary feature parameters with the preset pathogen type discrimination rules, calculate the matching score of the boundary feature parameters under each pathogen type discrimination rule and normalize it to generate a pathogen type probability distribution, and determine the pathogen type based on the maximum probability value in the pathogen type probability distribution.

[0030] A circular sampling region with a radius of 100 μm was constructed centered on the candidate pathogen location, sufficient to cover the fluorescence distribution characteristics of the target pathogen and surrounding tissues. Sixteen sampling directions were uniformly arranged along the circumference of the circular region, with an angular interval of 22.5°, ensuring omnidirectional capture of pathogen boundary features. In each sampling direction, samples were taken radially outward from the center point at equal intervals of 2 μm, until the boundary of the circular sampling region was reached, forming 16 radial sampling lines.

[0031] Fluorescence intensity values ​​were collected along each radial sampling line, and the fluorescence intensity and corresponding radial distance of each sampling point were recorded, resulting in 16 sets of radial fluorescence intensity distribution data. To eliminate noise interference, Gaussian filtering was applied to smooth each set of radial fluorescence intensity distribution data, with a filter window size of 5 sampling points and a standard deviation of 1.2. The smoothed fluorescence intensity distribution curves more accurately reflect the true characteristics of the pathogen boundary and reduce interference caused by image noise.

[0032] Analyze the smoothed fluorescence intensity distribution curve to locate the peak fluorescence intensity positions. Typically, the core region of the pathogen exhibits high fluorescence intensity, appearing as a distinct peak on the curve. The peak fluorescence intensity position is determined by locating the point of maximum fluorescence intensity on each radial sampling line; this position usually corresponds to the core region of the pathogen. Some pathogens may have multiple peaks; in such cases, the dominant peak is selected as the characteristic point.

[0033] Starting from the peak fluorescence intensity position, the rate of change of fluorescence intensity is calculated radially outward to generate a fluorescence intensity decay gradient curve. The rate of change is calculated based on the ratio of the difference in fluorescence intensity between two adjacent sampling points to the difference in their radial distance. Pathogen boundary regions typically exhibit abrupt changes in fluorescence intensity, which are reflected as significant fluctuations in the decay gradient curve. For mononuclear pathogens, such as Staphylococcus aureus, the decay gradient curve usually exhibits a single-peak shape; for pleomorphic pathogens, such as Pseudomonas aeruginosa, the decay gradient curve may exhibit multi-peak characteristics.

[0034] Identify the points of maximum and minimum rate of change on the fluorescence intensity decay gradient curve. The point of maximum rate of change typically corresponds to the inner interface of the pathogen boundary, where the fluorescence intensity rapidly transitions from high to low values; the point of minimum rate of change usually corresponds to the outer interface of the pathogen boundary, where the fluorescence intensity change tends to be gradual. The width of the transition zone is determined by calculating the radial distance between the points of maximum and minimum rate of change.

[0035] Statistical analysis was performed on the maximum and minimum values ​​of fluorescence intensity attenuation gradients and the width of the transition band measured in 16 sampling directions. The mean, standard deviation, skewness, and kurtosis were calculated to generate a set of boundary feature parameters. This set of boundary feature parameters contains multi-dimensional feature information, comprehensively describing the boundary characteristics of the pathogen. The mean reflects the overall intensity of the boundary gradient, the standard deviation characterizes the uniformity of the boundary gradient, and the skewness and kurtosis describe the morphological characteristics of the boundary gradient distribution.

[0036] The extracted boundary feature parameters are matched against a pre-defined pathogen type discrimination rule library. This library stores boundary feature parameter templates for common conjunctival sac pathogens, including typical boundary features of various bacteria, viruses, fungi, and other pathogens. The matching process employs a feature space distance calculation method to measure the similarity between the boundary feature parameters of the sample to be tested and the features of each pathogen in the template library.

[0037] For each pathogen type, a matching score is calculated between the sample boundary feature parameters and the template for that type. The matching score calculation considers the weights of different feature dimensions, with important feature dimensions assigned higher weights. For bacterial pathogens, the gradient mean and transition band width have higher weights; for viral pathogens, the gradient standard deviation and kurtosis have higher weights. The matching scores for all pathogen types are normalized and converted into probability values ​​between 0 and 1, forming a pathogen type probability distribution.

[0038] Based on probability distribution, the pathogen type corresponding to the highest probability value is selected as the final determination result. To improve the reliability of the judgment, a minimum probability threshold of 0.65 is set. If the highest probability value is lower than this threshold, it is marked as "undetermined type" and further verification using other detection methods is required. For example, the boundary characteristics of Staphylococcus aureus are characterized by a high gradient mean (approximately 0.28), a moderate transition zone width (approximately 5.5 μm), and a uniform gradient distribution; while herpes simplex virus is characterized by a low gradient mean (approximately 0.15), a narrow transition zone (approximately 2 μm), and a non-uniform gradient distribution.

[0039] By analyzing the boundary feature parameters and matching the type discrimination rules described above, combined with the spectral response characteristics from the previous step, accurate identification and classification of conjunctival sac pathogens can be achieved, providing strong support for clinical diagnosis and treatment. For unclear discrimination results, cross-validation can be performed using traditional microbial culture or molecular biological detection methods to further improve diagnostic accuracy.

[0040] This invention achieves accurate classification and identification of pathogens in the conjunctival sac through pathogen boundary feature analysis technology. This technology utilizes the unique fluorescence intensity distribution pattern formed at the boundary after pathogen fluorescence staining to extract key feature parameters such as the radial fluorescence intensity gradient and transition zone width, establishing a complete pathogen type discrimination mechanism. Compared with traditional detection methods, this technology eliminates the need for cumbersome culture and staining steps, significantly shortening detection time and improving detection efficiency. Through multi-directional sampling and omnidirectional feature extraction, it overcomes the limitations of single-angle observation, enhancing the stability and reliability of detection results.

[0041] Step 103: Collect the time-series fluorescence intensity of pathogen candidate sites, calculate the intensity change rate between adjacent time points, and screen pathogen candidate sites with intensity change rates exceeding the activity threshold as active pathogen sites.

[0042] In some embodiments of the present invention, step 103 may specifically include the following sub-steps: Sub-step 1031: Continuously acquire fluorescence images at preset time intervals, extract fluorescence intensity values ​​of pathogen candidate locations at continuous time points, and construct the temporal fluorescence intensity of pathogen candidate locations; Sub-step 1032: Adaptive filtering is performed on the temporal fluorescence intensity of the pathogen candidate location to generate the filtered temporal fluorescence intensity; Sub-step 1033: Calculate the rate of change of intensity between adjacent time points in the filtered time-series fluorescence intensity, and construct the rate of change of intensity sequence of pathogen candidate locations; Sub-step 1034: Extract the dynamic change features from the intensity change rate sequence of pathogen candidate sites to form activity feature indicators of pathogen candidate sites. Sub-step 1035: Compare the activity characteristic indicators of the pathogen candidate locations with the preset activity threshold, and screen out the pathogen candidate locations whose activity characteristic indicators exceed the preset activity threshold. Sub-step 1036: Perform spatial correlation analysis on the screened pathogen candidate locations to identify pathogen candidate locations with spatial clustering as active pathogen locations.

[0043] The fluorescence image acquisition time interval was set to 5 seconds, and 40 fluorescence images were continuously acquired, with a total observation time of 200 seconds. From the 40 continuously acquired fluorescence images, the fluorescence intensity value of each identified pathogen candidate location was extracted. A circular region of interest with a radius of 10 μm was set at the center of each pathogen candidate location, and the average fluorescence intensity of all pixels within this region was calculated. For larger pathogen candidate locations, such as *Pseudomonas aeruginosa* colonies, the region of interest could be expanded to 15 μm; for smaller pathogen candidate locations, such as herpes simplex virus, the region of interest could be reduced to 5 μm. In this way, a time-series fluorescence intensity data sequence containing 40 time points was constructed for each pathogen candidate location.

[0044] Time-series fluorescence intensity data for pathogen candidate sites often contain random noise and systematic fluctuations, requiring signal filtering. An adaptive median filtering method is used to process this data, with the filter window size dynamically adjusted based on the signal-to-noise ratio (SNR). When the SNR is low, the window size is set to 5 time points; when the SNR is high, the window size can be reduced to 3 time points. The filtering process preserves abrupt changes in the data while smoothing out random fluctuations. Adaptive median filtering effectively removes outliers and preserves the true trend of fluorescence intensity changes, which is crucial for the detection of active pathogens.

[0045] For the filtered time-series fluorescence intensity data, the ratio of the fluorescence intensity difference between two adjacent time points to the time interval was calculated, yielding intensity change rate data for 39 time points. Active pathogens exhibit significant fluorescence intensity fluctuations, reflecting changes in their metabolic activity and membrane structure. The intensity change rate data for active pathogens show a regular fluctuation pattern, with a fluctuation period typically between 20 and 35 seconds. In contrast, dead or dormant pathogens exhibit smaller, random fluctuations, lacking obvious periodic characteristics. The average intensity change rate for active *Pseudomonas aeruginosa* strains was 7.8% / 5 seconds, while the rate for dead strains was only 1.2% / 5 seconds.

[0046] Dynamic change characteristics are extracted from the intensity change rate sequence, including the mean change rate, fluctuation range, fluctuation frequency, and fluctuation pattern. The mean change rate is calculated as the arithmetic mean of 39 change rate data points, representing the overall intensity change trend. The fluctuation range is calculated as the difference between the maximum and minimum change rate values, reflecting the magnitude of intensity change. The fluctuation frequency is obtained by statistically analyzing the number of alternating positive and negative change rates and dividing by the total observation time. The fluctuation pattern analyzes the periodicity and regularity of the change rate sequence. These characteristics combined constitute the activity characteristic indicators of pathogen candidate sites. Typical activity characteristic indicators of active pathogens include a mean change rate > 3% / 5s, a fluctuation range > 10%, a fluctuation frequency of 0.15-0.25Hz, and a fluctuation pattern exhibiting obvious periodicity.

[0047] The activity characteristics of each pathogen candidate location are compared with preset activity thresholds. These thresholds are determined based on the characteristics of different pathogen types. For bacterial pathogens, the activity threshold is set as a mean change rate > 2.5% / 5s, a fluctuation range > 8%, and a fluctuation frequency > 0.12Hz. For viral pathogens, the activity threshold is adjusted accordingly to a mean change rate > 1.8% / 5s, a fluctuation range > 6%, and a fluctuation frequency > 0.08Hz. Only when all parameters of the activity characteristics exceed the corresponding thresholds is the pathogen candidate location marked as a potential active pathogen location.

[0048] Spatial correlation analysis was performed on the screened potential active pathogen locations to identify candidate locations with spatial clustering. The spatial correlation analysis was based on a distance threshold of 50 μm. Two potential active pathogen locations were considered spatially correlated if the distance between them was less than the threshold. An association network was constructed using spatial correlation analysis, where nodes represent potential active pathogen locations and edges represent spatially correlated pairs of locations. Highly connected subnetworks were then identified within the association network; these subnetworks represent spatially clustered groups of active pathogens. Bacterial infections typically exhibit larger spatial clusters, with 10-30 pathogens per cluster; viral infections, on the other hand, exhibit smaller spatial clusters, typically containing 3-8 pathogens per cluster.

[0049] Spatial correlation analysis also considers the temporal synchronicity of active characteristics; active pathogens located in the same infection area often exhibit similar fluorescence intensity change patterns and temporal synchronicity. Temporal synchronicity is assessed by calculating the correlation coefficient of the intensity change rate sequences of two pathogen locations; pathogen locations with a correlation coefficient > 0.7 are considered to be temporally synchronized. The combined effect of temporal synchronicity and spatial correlation improves the accuracy of active pathogen population identification.

[0050] This invention achieves precise detection of the active state of pathogens in the conjunctival sac through fluorescence temporal image analysis technology. This technology overcomes the limitations of traditional static fluorescence imaging by introducing a temporal analysis dimension, capturing the dynamic changes in fluorescence characteristics during pathogen activity and metabolism. By accurately calculating and analyzing the rate of change of temporal fluorescence intensity, combined with spatial correlation and temporal synchronicity characteristics, it effectively distinguishes between active and inactive pathogens, overcoming the limitation of traditional methods in assessing pathogen activity. This invention can not only identify the location of pathogens but also determine their activity state, significantly improving the clinical value and application scope of conjunctival sac pathogen detection.

[0051] Step 104: Identify the main diffusion axis based on the spatial distribution of the active pathogen's location, calculate the spatial density gradient by statistically analyzing the density changes of the active pathogen's location along the main diffusion axis, and couple the intensity change rate with the spatial density gradient to calculate the pathogen's diffusion rate and determine the warning area.

[0052] In some embodiments of the present invention, step 104 may specifically include the following sub-steps: Sub-step 1041: Extract the spatial coordinates of the active pathogen locations based on the spatial distribution of the active pathogen locations, construct an adjacency graph, identify the connected links with the most connected nodes in the adjacency graph, and determine the extension direction of the connected links as the diffusion axis. Sub-step 1042: Set up continuous measurement segments along the main diffusion axis, count the number of active pathogen locations in each measurement segment to form a location density sequence, identify the inflection point of the sign conversion of the density difference between adjacent measurement segments in the location density sequence, extract the measurement segment between the inflection point and the end as the diffusion wavefront segment, and calculate the density difference between adjacent measurement segments in the diffusion wavefront segment as the spatial density gradient. Sub-step 1043: Obtain the intensity change rate of the active pathogen position within the diffusion wavefront, sort the active pathogen positions according to their projection positions on the diffusion principal axis to form an intensity change rate sequence, and calculate the increasing trend coefficient of the intensity change rate sequence. Sub-step 1044: Calculate the consistency between the increasing trend coefficient and the spatial density gradient, determine the coupling weight based on the consistency, and couple the intensity change rate with the spatial density gradient to generate the pathogen diffusion rate. Sub-step 1045: Starting from the boundary position of the diffusion wavefront, calculate the propagation distance along the main diffusion axis according to the pathogen diffusion rate, and determine the spatial area covered by the propagation distance as the warning area.

[0053] Spatial coordinates were extracted based on the spatial distribution of identified active pathogen locations. Each active pathogen location was recorded as a two-dimensional coordinate point in the image, with coordinate values ​​in pixels. For ease of analysis, pixel coordinates were converted to actual physical distances at a conversion ratio of 0.5 μm / pixel. In conjunctival sac samples, active pathogen locations typically exhibit a certain degree of aggregation and directional distribution, reflecting the diffusion path of pathogens within the conjunctival sac tissue. For Pseudomonas aeruginosa infected samples, a typical observation area was 800 μm × 600 μm, containing approximately 120 active pathogen locations. In actual processing, a set of coordinate points was used to represent the spatial distribution of active pathogens, and an adjacency graph was constructed based on a distance threshold. The distance threshold was set to 30 μm, meaning that a connection was established in the adjacency graph when the distance between two active pathogen locations was less than 30 μm.

[0054] The adjacency graph is represented by an undirected graph structure, where each node represents a location of an active pathogen, and edges between nodes indicate an adjacency relationship between the two locations. In the constructed adjacency graph, the connected path with the most connected nodes is identified; this path represents the most significant pathogen aggregation path in space. A depth-first search algorithm is used to traverse the adjacency graph to find the longest connected path, which typically contains 25-40 locations of active pathogens. Least squares fitting is performed on the locations of active pathogens in the connected path to obtain a straight line; the direction of this line is the pathogen diffusion axis. In *Pseudomonas aeruginosa* samples, the direction of the diffusion axis usually aligns with the direction of conjunctival folds, indicating a trend of pathogen diffusion along tissue structures. The diffusion axis can also be verified using principal component analysis; the obtained directional angle deviation is typically less than 5 degrees, demonstrating the stability of the method.

[0055] Continuous measurement segments were set along the identified diffusion axis, each segment being 15 μm wide and the length equal to the image width. The measurement segments were perpendicular to the diffusion axis and arranged sequentially along the axis, with a total of 40 segments covering the entire observation area. The number of viable pathogen locations within each measurement segment was counted, forming a location density sequence. Location density is the ratio of the number of viable pathogen locations within a measurement segment to the area of ​​that segment. The location density sequence reflects the distribution pattern of viable pathogens along the diffusion axis, typically showing a trend of first increasing and then decreasing.

[0056] Identify the inflection point where the density difference sign changes from increasing to decreasing in the location density sequence. Calculate the density difference between adjacent measurement segments; when the difference changes from positive to negative, the corresponding measurement segment is the inflection point. Define the measurement segment from the inflection point to the end of the sequence as the diffusion front, representing the active front of pathogen diffusion. Within the diffusion front, calculate the density difference between adjacent measurement segments as the spatial density gradient. The spatial density gradient reflects the rate of change of the active pathogen location density with spatial location. Within the diffusion front, the spatial density gradient is typically negative, and its absolute value gradually decreases with distance from the inflection point, showing a decreasing trend in pathogen diffusion intensity.

[0057] Intensity change rate data for all active pathogen locations within the diffusion wavefront were obtained, derived from the previously calculated pathogen activity characteristics. The active pathogen locations were sorted according to their projection positions on the diffusion principal axis, forming an intensity change rate sequence. Within the diffusion wavefront, the intensity change rate of active pathogens near the inflection point was lower, approximately 4-6% / 5s, while the intensity change rate of active pathogens near the wavefront was higher, reaching 7-9% / 5s. The increasing trend coefficient of the intensity change rate sequence was calculated, obtained through linear regression, representing the trend of intensity change rate with spatial location. A positive increasing trend coefficient indicates that the intensity change rate increases as it moves towards the wavefront, a typical characteristic of active diffusion processes. In *Pseudomonas aeruginosa* samples, the increasing trend coefficient was typically between 0.15 and 0.25, indicating a significant increasing intensity change rate.

[0058] The consistency between the increasing trend coefficient and the spatial density gradient is calculated. This consistency is measured by their correlation. When the increasing trend coefficient is positive and the spatial density gradient is negative, it indicates that the rate of change of intensity increases with increasing spatial location, while the spatial density decreases. In this case, their directions of change are opposite, and the consistency is negative. The larger the absolute value of the consistency, the stronger the correlation between the intensity change characteristics and the spatial distribution characteristics. Coupling weights are determined based on the consistency. When the absolute value of the consistency is in the range of 0.7-0.9, the coupling weights for the rate of change of intensity and the spatial density gradient are set to 0.6 and 0.4, respectively; when the absolute value of the consistency is in the range of 0.5-0.7, the coupling weights are set to 0.5 and 0.5, respectively; and when the absolute value of the consistency is less than 0.5, the coupling weights are set to 0.4 and 0.6, respectively.

[0059] The intensity change rate and spatial density gradient are coupled using defined coupling weights. The coupling method is a weighted combination, where the normalized intensity change rate and the absolute value of the spatial density gradient are combined according to their weights to obtain the diffusion rate index. The diffusion rate represents the speed at which pathogens spread within the conjunctival sac tissue. The diffusion rate is related to factors such as pathogen type, activity intensity, and conjunctival sac environmental conditions, and is a comprehensive reflection of the pathogen's diffusion ability.

[0060] Starting from the boundary of the diffusion wavefront, the propagation distance within a certain future timeframe is estimated along the main diffusion axis based on the calculated pathogen spread rate. The propagation time is set at 60 minutes, a critical time window before clinical intervention to predict potential pathogen spread. Propagation distance = diffusion rate × propagation time; in *Pseudomonas aeruginosa* samples, the propagation distance is typically 900-1500 μm. The spatial area covered by this propagation distance is defined as the warning zone, representing the potential extent of pathogen spread in the short term. The warning zone extends along the main diffusion axis from the boundary of the diffusion wavefront, with its width perpendicular to the axis equal to the width of the diffusion wavefront plus a 30 μm buffer zone on each side. The warning zone is marked with a specific color on the image, providing a visual indication of the risk area.

[0061] This invention integrates dynamic changes in fluorescence intensity with spatial distribution information to accurately predict the spread trend of pathogens in the conjunctival sac. It not only determines the current spatial extent of pathogen infection but also predicts areas where infection may spread in the near future, providing a time window and precise targets for clinical intervention. Identifying the warning area allows physicians to administer targeted local medications and treatments, reducing the overuse of antibiotics and other drugs and improving treatment efficiency. The invention also quantitatively characterizes the spread characteristics of different types of pathogens, providing new technical means and analytical perspectives for the study of conjunctival sac infections.

[0062] Step 105: Identify abnormal pixels that deviate from the background baseline within the warning area, perform time-series tracking to verify the continuous growth characteristics of the abnormal pixels, and identify and mark the abnormal pixels that meet the continuous growth characteristics as early infection areas.

[0063] In some embodiments of the present invention, step 105 may specifically include the following sub-steps: Sub-step 1051: Extract the fluorescence intensity of each pixel within the warning area, select a reference area outside the boundary of the warning area where no active pathogens appear, statistically analyze the fluorescence intensity distribution characteristics of pixels within the reference area, and construct a background baseline; Sub-step 1052: Calculate the degree of deviation between the fluorescence intensity of each pixel in the warning area and the background baseline, and filter out pixels whose deviation exceeds the preset fluctuation range as abnormal pixels; Sub-step 1053: Continuously collect fluorescence intensity of abnormal pixels at preset time intervals, construct the time-series fluorescence intensity trajectory of abnormal pixels, extract continuous rising segments on the time-series fluorescence intensity trajectory, calculate the duration and cumulative increase of the continuous rising segments, and determine abnormal pixels that meet the growth judgment conditions in terms of both duration and cumulative increase as abnormal pixels that meet the continuous growth characteristics. Sub-step 1054 involves spatially clustering abnormal pixels that conform to the continuous growth characteristic, identifying spatially adjacent abnormal pixels that conform to the continuous growth characteristic as connected regions, confirming the connected regions as early infection regions, and generating contour annotation information on the boundaries of the early infection regions.

[0064] Once the warning area is determined, the fluorescence intensity values ​​of all pixels within that area are extracted. The fluorescence intensity range for each pixel is 0-255. The warning area typically contains potentially infected conjunctival sac tissue and has an area of ​​approximately 1.5 × 10⁶ μm. 2 The corresponding image pixel count is approximately 6 × 10⁶. A reference region is selected 50-100 μm outward from the boundary of the warning area. The reference region is annular with a width of 50 μm and must not contain any identified active pathogens. The reference region typically contains uninfected healthy conjunctival sac tissue, and its fluorescence intensity distribution reflects the background fluorescence characteristics of normal tissue. For Pseudomonas aeruginosa infected samples, the average fluorescence intensity of pixels within the reference region is typically between 35 and 45, with a standard deviation between 4 and 8. The fluorescence intensity of the reference region exhibits an approximately normal distribution and has stable background characteristics, making it suitable as a benchmark for judging abnormalities.

[0065] The fluorescence intensity of all pixels within the reference area was statistically analyzed, and the mean and standard deviation of the fluorescence intensity were calculated to construct a background baseline. The background baseline was defined by the mean fluorescence intensity and its fluctuation range, which was determined by the standard deviation. The background baseline reflects the characteristics of normal tissue under fluorescent staining and is an important reference for identifying abnormal fluorescence. In Pseudomonas aeruginosa infected samples, the mean of the background baseline was 40, the standard deviation was 6, and the fluctuation range was set to mean ± 2 times the standard deviation, meaning pixels with fluorescence intensity values ​​between 28 and 52 were considered within the normal fluctuation range. The construction of the background baseline considered factors such as tissue autofluorescence and the non-specific binding of fluorescent staining to ensure the accuracy of abnormality detection. The background baseline remained constant during the analysis, providing a stable reference for subsequent identification of abnormal pixels.

[0066] The difference between the fluorescence intensity of each pixel within the warning area and the mean of the background baseline is calculated, and this difference is compared with a preset fluctuation range to determine whether the pixel's fluorescence intensity is abnormal. The preset fluctuation range is three times the standard deviation of the background baseline; that is, a pixel is considered abnormal when the absolute value of the difference between the fluorescence intensity and the mean of the background baseline exceeds 18. For different types of pathogen infections, the multiple of the preset fluctuation range can be adjusted according to the fluctuation characteristics of the background baseline. For Pseudomonas aeruginosa infection, pixels with a fluorescence intensity greater than 58 or less than 22 are considered abnormal. Generally, the fluorescence intensity of pixels in early infection areas is higher than the background baseline; therefore, the focus is mainly on abnormal pixels whose fluorescence intensity exceeds the upper limit threshold. Within the warning area, approximately 2-5% of pixels are typically identified as abnormal pixels; these pixels may represent early signs of infection.

[0067] The identified abnormal pixels were time-series tracked and verified to confirm whether their fluorescence intensity changes conformed to the characteristics of infection spread. A preset time interval of 5 seconds was set, and fluorescence images were continuously acquired at 15 time points, with a total observation time of 75 seconds. For each abnormal pixel, fluorescence intensity values ​​at 15 time points were extracted to construct a time-series fluorescence intensity trajectory. The time-series fluorescence intensity trajectory reflects the fluorescence intensity change trend of the abnormal pixel over a short period. Continuous rising segments were extracted from the time-series fluorescence intensity trajectory. A continuous rising segment was defined as a sequence of time points where fluorescence intensity continuously increased, allowing for small fluctuations of no more than two time points. The duration and cumulative increase of the continuous rising segment were calculated. The duration was calculated by multiplying the number of time points contained in the continuous rising segment by the time interval, and the cumulative increase was calculated as the difference in fluorescence intensity between the end and the beginning time point of the continuous rising segment.

[0068] The criteria for determining growth were a duration of at least 25 seconds and a cumulative increase of at least 12 fluorescence intensity units. For Pseudomonas aeruginosa infection, typical abnormal pixels exhibiting continuous growth characteristics show an increase in fluorescence intensity from approximately 60 to over 80 within a 75-second observation period, with a continuous increase lasting 35-50 seconds and a cumulative increase of 15-25 fluorescence intensity units. Abnormal pixels in non-infected areas typically show a pattern of brief increase followed by a rapid decline or fluctuating patterns, with a shorter duration and limited cumulative increase. Verification of continuous growth characteristics effectively filters out false-positive abnormal pixels caused by random fluctuations and transient interference, improving the accuracy of early infection area identification. Within the warning area, typically 30-50% of abnormal pixels conform to the continuous growth characteristics.

[0069] Spatial clustering analysis was performed on anomalous pixels exhibiting continuous growth characteristics. The spatial clustering employed a region growing method, with the adjacency criterion being a Manhattan distance of no more than 3 pixels. Starting from each anomalous pixel exhibiting continuous growth characteristics, its surrounding pixels were examined. If adjacent pixels also exhibited continuous growth characteristics, they were included in the same connected region. This process was repeated until no further expansion was possible, forming a connected region. In practical applications, based on the biological characteristics of *Pseudomonas aeruginosa*, the minimum number of pixels in a connected region was set to 5, and the maximum number to 200. Connected regions that are too small may be noise, while those that are too large may no longer represent an early infection state. Connected regions meeting the criteria were identified as early infection regions. Each early infection region typically contains 20-100 anomalous pixels exhibiting continuous growth characteristics, with an area of ​​approximately 500-2500 μm. 2 .

[0070] Contour annotations are generated on the boundaries of confirmed early infection areas for visualization and clinical reference. The contour annotations employ a boundary tracking algorithm, generating closed curves along the outer boundary pixels of the early infection areas. For each early infection area, its center coordinates, area, average fluorescence intensity, and other characteristic parameters are calculated and displayed near the contour annotations. The annotations of early infection areas visually reflect the potential hotspots of pathogen spread in the conjunctival sac, providing precise spatial localization for clinical intervention. In Pseudomonas aeruginosa infection samples, early infection areas typically appear within a 5-15 μm range around the location of active pathogens, and are mostly distributed in the area anterior to the main spread axis. The shapes of early infection areas are usually irregular, reflecting the characteristics of pathogen spread along tissue structures.

[0071] This invention achieves highly sensitive identification of early signs of pathogen infection through detailed analysis and time-series tracking verification of the fluorescence intensity of pixels within the warning area. It improves the time lead and spatial accuracy of pathogen detection in the conjunctival sac, enabling the detection of infection signs at a stage where pathogen numbers are low and clinical symptoms are not yet obvious. By combining background baseline construction with abnormal pixel screening, non-specific fluorescence interference is effectively eliminated, improving detection specificity. The introduction of time-series tracking verification and spatial clustering analysis further enhances the reliability of early infection area identification.

[0072] like Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of a conjunctival sac pathogen detection system for fluorescence staining image analysis provided in an embodiment of the present invention. The system includes: Image processing module 201 is used to acquire fluorescence images of conjunctival sac samples at multiple wavelengths, perform inter-channel difference operations on fluorescence images of different wavelengths to construct spectral response surfaces, and identify curvature abrupt change regions on the spectral response surfaces to determine candidate locations of pathogens. The type discrimination module 202 is used to extract the fluorescence intensity distribution curve radially with the pathogen candidate location as the center, calculate the attenuation gradient and transition band width to form boundary feature parameters, and match the boundary feature parameters with the preset pathogen type discrimination rules to determine the pathogen type; The activity recognition module 203 is used to collect the time-series fluorescence intensity of pathogen candidate sites, calculate the intensity change rate between adjacent time points, and screen pathogen candidate sites whose intensity change rate exceeds the activity threshold as active pathogen sites. The diffusion analysis module 204 is used to identify the main diffusion axis based on the spatial distribution of the location of active pathogens, calculate the spatial density gradient by statistically analyzing the density change of the location of active pathogens along the direction of the main diffusion axis, and couple the intensity change rate with the spatial density gradient to calculate the pathogen diffusion rate and determine the warning area. The region labeling module 205 is used to identify abnormal pixels that deviate from the background baseline within the warning area, perform time-series tracking to verify the continuous growth characteristics of the abnormal pixels, and identify and label the abnormal pixels that meet the continuous growth characteristics as early infection areas.

[0073] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0074] One technical solution provided in this embodiment of the invention is a computer-readable storage medium storing a computer program, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0075] The specific embodiments described above are preferred embodiments of the present invention and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A method for detecting pathogens in the conjunctival sac using fluorescent staining image analysis, characterized in that, Includes the following steps: Fluorescence images of conjunctival sac samples at multiple wavelengths were acquired. Inter-channel difference operations were performed on the fluorescence images at different wavelengths to construct spectral response surfaces. Regions of curvature abrupt change on the spectral response surfaces were identified to determine candidate locations of pathogens. Fluorescence intensity distribution curves are extracted radially from the candidate pathogen location, and the attenuation gradient and transition band width are calculated to form boundary feature parameters. The boundary feature parameters are then matched with preset pathogen type discrimination rules to determine the pathogen type. The time-series fluorescence intensity of pathogen candidate sites is collected, the rate of intensity change between adjacent time points is calculated, and pathogen candidate sites with an intensity change rate exceeding the activity threshold are selected as active pathogen sites. Based on the spatial distribution of active pathogen locations, the diffusion axis is identified, the spatial density gradient is calculated by statistically analyzing the density changes of active pathogen locations along the diffusion axis, and the rate of intensity change is coupled with the spatial density gradient to calculate the pathogen diffusion rate and determine the warning area. Abnormal pixels that deviate from the background baseline within the warning area are identified. The abnormal pixels are then tracked over time to verify their continuous growth characteristics. Abnormal pixels that meet the continuous growth characteristics are identified as early infection areas and marked.

2. The method according to claim 1, characterized in that, Fluorescence images of conjunctival sac samples at multiple wavelengths were acquired. Inter-channel difference operations were performed on the fluorescence images at different wavelengths to construct spectral response surfaces. Regions of curvature abrupt changes on the spectral response surfaces were identified to determine candidate pathogen locations, including: The conjunctival sac sample was sequentially irradiated with a multi-band excitation light source, and fluorescence images were acquired at each excitation wavelength to obtain fluorescence images at multiple wavelengths. Fluorescence images at adjacent wavelengths are selected for inter-channel difference calculation to calculate the fluorescence intensity difference between the adjacent wavelength fluorescence images and generate inter-channel difference data of fluorescence images. The differential data between fluorescence image channels are arranged in wavelength order to construct a spectral response surface. The fluorescence intensity gradient value and curvature value of each point on the spectral response surface are calculated. Based on fluorescence intensity gradient values ​​and curvature values, regions on the spectral response surface whose gradient change rate exceeds a preset change threshold and whose curvature values ​​change abruptly are identified as candidate pathogen regions. Fluorescence intensity distribution features are extracted within the candidate pathogen region, and the distribution parameters and morphological parameters of the fluorescence intensity distribution features are calculated. Based on distribution and morphological parameters, feature matching is performed on candidate regions of pathogens, and regions that meet the preset matching conditions are identified as candidate locations of pathogens.

3. The method according to claim 1, characterized in that, Fluorescence intensity distribution curves are extracted radially from the candidate pathogen location. The attenuation gradient and transition band width are calculated to form boundary feature parameters. These boundary feature parameters are then matched with preset pathogen type discrimination rules to determine the pathogen type, including: A circular sampling area is constructed with the candidate pathogen location as the center, and the sampling direction is set along the circumference. Fluorescence intensity values ​​are collected in the sampling direction to generate a fluorescence intensity distribution curve. The fluorescence intensity distribution curve was smoothed, the trend of fluorescence intensity with radial distance was calculated, and the peak position of fluorescence intensity was extracted. The rate of change of fluorescence intensity from the peak position of fluorescence intensity to the sampling boundary is calculated along the radial direction to generate a fluorescence intensity attenuation gradient; The points of maximum and minimum rate of change are identified on the fluorescence intensity decay gradient, and the radial distance between the points of maximum and minimum rate of change is determined as the transition band width. The fluorescence intensity attenuation gradient and the transition band width are combined to generate boundary feature parameters; The boundary feature parameters are matched with the preset pathogen type discrimination rules, the matching scores of the boundary feature parameters under each pathogen type discrimination rule are calculated and normalized, a pathogen type probability distribution is generated, and the pathogen type is determined based on the maximum probability value in the pathogen type probability distribution.

4. The method according to claim 1, characterized in that, The temporal fluorescence intensity of candidate pathogen sites is collected, the rate of intensity change between adjacent time points is calculated, and candidate pathogen sites with an intensity change rate exceeding the activity threshold are selected as active pathogen sites, including: Fluorescence images are continuously acquired at preset time intervals, and the fluorescence intensity values ​​of pathogen candidate locations at continuous time points are extracted to construct the temporal fluorescence intensity of pathogen candidate locations. Adaptive filtering is applied to the temporal fluorescence intensity at candidate pathogen sites to generate filtered temporal fluorescence intensity. Calculate the rate of intensity change between adjacent time points in the filtered time-series fluorescence intensity to construct a sequence of intensity change rates at candidate pathogen locations; Dynamic change features are extracted from the intensity change rate sequence of pathogen candidate sites to form activity feature indicators of pathogen candidate sites; The activity characteristic indicators of pathogen candidate sites are compared with preset activity thresholds to screen out pathogen candidate sites whose activity characteristic indicators exceed the preset activity thresholds. Spatial correlation analysis was performed on the screened pathogen candidate locations to identify pathogen candidate locations with spatial clustering as active pathogen locations.

5. The method according to claim 1, characterized in that, Based on the spatial distribution of active pathogen locations, the main diffusion axis is identified. The spatial density gradient is calculated by statistically analyzing the density changes of active pathogen locations along the main diffusion axis. The rate of intensity change is coupled with the spatial density gradient to calculate the pathogen diffusion rate and determine the warning area, including: Based on the spatial distribution of the location of the active pathogen, the spatial coordinates of the location of the active pathogen are extracted, an adjacency graph is constructed, the connected link with the most connected nodes in the adjacency graph is identified, and the extension direction of the connected link is determined as the diffusion axis. A continuous measurement segment is set up along the main diffusion axis. The number of active pathogen locations in each measurement segment is counted to form a location density sequence. The inflection point of the sign conversion of the density difference between adjacent measurement segments in the location density sequence is identified. The measurement segment between the inflection point and the end is extracted as the diffusion wavefront segment. The density difference between adjacent measurement segments in the diffusion wavefront segment is calculated as the spatial density gradient. The intensity change rate of the active pathogen location within the diffusion wavefront is obtained, and the intensity change rate sequence is formed by sorting the active pathogen locations according to their projection positions on the diffusion principal axis. The increasing trend coefficient of the intensity change rate sequence is then calculated. Calculate the consistency between the increasing trend coefficient and the spatial density gradient, determine the coupling weight based on the consistency, and couple the intensity change rate with the spatial density gradient to generate the pathogen diffusion rate. Starting from the boundary position of the diffusion wavefront, the propagation distance is calculated along the main diffusion axis according to the pathogen diffusion rate, and the spatial area covered by the propagation distance is determined as the warning area.

6. The method according to claim 1, characterized in that, Within the warning area, identify anomalous pixels that deviate from the background baseline. Perform time-series tracking to verify the continuous growth characteristics of these anomalous pixels. Identify anomalous pixels that meet the continuous growth characteristics as early infection areas and mark them accordingly. Extract the fluorescence intensity of each pixel within the warning area, select a reference area outside the boundary of the warning area where no active pathogens appear, statistically analyze the fluorescence intensity distribution characteristics of pixels within the reference area, and construct a background baseline; Calculate the degree of deviation between the fluorescence intensity of each pixel within the warning area and the background baseline, and filter out pixels whose deviation exceeds the preset fluctuation range as abnormal pixels; Fluorescence intensity is continuously collected for abnormal pixels at preset time intervals to construct a time-series fluorescence intensity trajectory for abnormal pixels. Continuous rising segments are extracted from the time-series fluorescence intensity trajectory, and the duration and cumulative increase of the continuous rising segments are calculated. Abnormal pixels whose duration and cumulative increase simultaneously meet the growth determination criteria are identified as abnormal pixels that meet the continuous growth characteristics. Spatial clustering is performed on abnormal pixels that exhibit continuous growth characteristics. Connected regions composed of spatially adjacent abnormal pixels that exhibit continuous growth characteristics are identified, and these connected regions are confirmed as early infection regions. Contour annotation information is generated on the boundaries of the early infection regions.

7. A conjunctival sac pathogen detection system based on fluorescence staining image analysis, used to implement the method described in any one of claims 1-6, characterized in that, The system includes: The image processing module is used to acquire fluorescence images of conjunctival sac samples at multiple wavelengths, perform inter-channel difference operations on fluorescence images at different wavelengths to construct spectral response surfaces, and identify curvature abrupt change regions on the spectral response surfaces to determine candidate locations of pathogens. The type discrimination module is used to extract the fluorescence intensity distribution curve radially with the pathogen candidate location as the center, calculate the attenuation gradient and transition band width to form boundary feature parameters, and match the boundary feature parameters with the preset pathogen type discrimination rules to determine the pathogen type; The activity recognition module is used to collect the time-series fluorescence intensity of pathogen candidate sites, calculate the rate of intensity change between adjacent time points, and screen pathogen candidate sites whose rate of intensity change exceeds the activity threshold as active pathogen sites. The diffusion analysis module is used to identify the main diffusion axis based on the spatial distribution of the location of active pathogens, calculate the spatial density gradient by statistically analyzing the density changes of active pathogens along the direction of the main diffusion axis, and couple the intensity change rate with the spatial density gradient to calculate the pathogen diffusion rate and determine the warning area. The region labeling module is used to identify abnormal pixels that deviate from the background baseline within the warning area, perform time-series tracking to verify the continuous growth characteristics of the abnormal pixels, and identify and label the abnormal pixels that meet the continuous growth characteristics as early infection areas.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 6.

Images