Gastric Cancer Pathology Image Analysis System

Abstract

The present invention relates to the field of image recognition technology, specifically to a gastric cancer pathology image analysis system, the system comprising: a spatiotemporal feature extraction module extracts spatial features of gastric cancer pathology images based on cell morphology and tissue structure, analyzes time series data in the lesion area, calculates statistical features of cell morphology in multiple regions, compares differences between cell nuclear features and tissue boundaries, obtains spatial feature time series, and generates key morphological features. In the present invention, by deeply analyzing cell morphology and tissue structure in gastric cancer pathology images, fine tracking of the evolution of the lesion area is achieved. The processing method extracts the time series of spatial features, which not only reveals subtle changes in cell nuclear features and tissue boundaries, but also maps these changes in high-dimensional manifold space, capturing continuous changes in cell nuclear features at different stages. This dynamic description can fully display the evolution of gastric cancer lesions, further enhancing the dynamic monitoring capability of pathological analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a gastric cancer pathology image analysis system. Background Art

[0002] Image recognition technology is a core part of the field of computer vision, which involves enabling computers to recognize and process objects in images or videos. This technology uses machine learning algorithms, especially deep learning, to train computers to recognize patterns and features in images. Image recognition has a wide range of applications, from simple object recognition to complex scene understanding. In recent years, with the development of technologies such as convolutional neural networks (CNN), the accuracy and efficiency of image recognition have been greatly improved, and it has been widely used in many fields such as medical image analysis, autonomous vehicles, security monitoring, and industrial vision systems.

[0003] Among them, the gastric cancer pathology image analysis system refers to the use of image recognition technology to analyze gastric pathology slice images to assist in the diagnosis of gastric cancer. The system automatically identifies possible cancerous areas by analyzing the cell morphology and tissue structure in the pathology image. This technology can help pathologists identify and analyze cancer faster and more accurately during the diagnosis process, especially when the number of pathology samples is large, improving work efficiency. In addition, the gastric cancer pathology image analysis system may also be used to monitor changes during cancer treatment, providing doctors with more information to make better treatment decisions.

[0004] Although current image recognition technology can identify specific patterns and features in the processing of gastric cancer pathology images, it lacks continuous tracking and analysis of dynamic changes in cells during the pathological process. Especially in terms of time series data processing and continuous change tracking, traditional methods rely more on static images, making it difficult to effectively capture and analyze the evolution of the lesion area. This limitation may result in pathological analysis failing to reflect the continuous changes in cell morphology in a timely manner, affecting diagnostic accuracy. For example, in gastric cancer cases, due to the lack of dynamic monitoring, diagnosis and treatment plans may not be adjusted in a timely manner, thus affecting the patient's final treatment effect and survival rate. In addition, traditional technologies are also insufficient in the deep analysis of complex interactions between cells, limiting the comprehensive understanding of complex pathological conditions, which is particularly prominent in scenarios requiring high-precision and personalized treatment. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a gastric cancer pathology image analysis system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: The gastric cancer pathology image analysis system comprises:

[0007] The spatiotemporal feature extraction module extracts the spatial features of gastric cancer pathological images based on cell morphology and tissue structure, analyzes the time series data in the lesion area, calculates the statistical features of cell morphology in multiple regions, compares the differences between cell nuclear features and tissue boundaries, obtains spatial feature time series, and generates key morphological features;

[0008] The dynamic change tracking module tracks the dynamic changes in the tissue structure based on the key morphological features, calculates the change rules of the intercellular matrix, analyzes the evolution of the lesion area in the time series, obtains the dynamic description of the lesion process, and obtains the deviation dynamic prediction analysis results;

[0009] The high-dimensional manifold mapping module maps the cell morphology and tissue structure in the high-dimensional manifold space based on the dynamic prediction analysis results of the displacement, tracks the continuous changes of the cell nuclear characteristics in the differentiation stage, obtains the high-dimensional dynamic characteristics of the gastric cancer lesions, and generates an enhanced displacement signal;

[0010] The evolutionary pattern recognition module identifies and classifies multiple evolutionary stages of gastric cancer lesions based on the enhanced offset signal, analyzes the characteristics of the lesion area, obtains the fusion result of the local characteristics of the intercellular matrix and the global characteristics of the tissue structure, and generates a cell evolution analysis result.

[0011] As a further solution of the present invention, the step of acquiring the key morphological features is specifically as follows:

[0012] Based on cell morphology and tissue structure, cell morphological features are extracted using the formula:

[0013]

[0014]

[0015] Calculate the size, shape and density characteristics of the cell nucleus to obtain statistical data of the cell nucleus characteristics;

[0016] Among them, Size is the size characteristic of the cell nucleus, Shape is the shape characteristic of the cell nucleus, N represents the number of cell nuclei, and A i represents the area of ​​the ith cell nucleus, P represents the perimeter of the cell nucleus, A represents the area of ​​the cell nucleus, and λ i represents the area attenuation coefficient, γ is the shape adjustment coefficient;

[0017] The statistical data of the cell nuclear characteristics are analyzed, and the differences between the cell nuclear characteristics and the tissue demarcation line are compared, using the formula:

[0018]

[0019] Generates the difference between the cell nucleus and tissue boundaries;

[0020] Among them, Xcell An array of eigenvalues ​​representing the cell nucleus, X boundary represents the array of eigenvalues ​​associated with organizational boundaries, σ represents the standard deviation of the eigenvalues, α is the difference adjustment coefficient, and Δ represents the difference;

[0021] According to the difference between the cell nucleus and the tissue boundary, the statistical characteristics of cell morphology in multiple regions are calculated using the formula:

[0022]

[0023] Get the spatial feature time series;

[0024] Among them, x k represents the difference between the cell nucleus and the tissue boundary of the kth region, w k represents the relative importance of the region in the overall analysis, δ is the smoothing parameter, and F(x) represents the weighted spatial feature time series;

[0025] According to the spatial feature time series, the morphological features are calculated using the formula:

[0026]

[0027] Obtain key morphological features;

[0028] Where T represents the length of the time series, f t represents the spatial feature value at time point t, β is the time decay factor, and KeyFeatures represents the key morphological features.

[0029] As a further solution of the present invention, the method for tracking dynamic changes in the organizational structure is specifically:

[0030] Based on the key morphological features, the changes in tissue structure at each time point were analyzed using the formula:

[0031]

[0032] Generate an assessment of organizational structural changes at time point t;

[0033] Among them, C t is the assessment of tissue structural changes at time point t, n is the number of key morphological features, and f i (t) is the i-th morphological feature at time point t, λ is the time attenuation coefficient, μ i is the weight coefficient of the i-th feature;

[0034] The evaluation of the organizational structure change at the time point t is weighted averaged using the formula:

[0035]

[0036] Generates average organizational structure changes over the time period;

[0037] Among them, S t is the average organizational structure change in the time period, k is the time window length, ω j is the weighting factor at time point j;

[0038] To analyze the average tissue structure changes during the stated time period, the formula was used:

[0039]

[0040] Generate tracking results of dynamic changes in organizational structure;

[0041] Among them, T t To track the dynamic changes in organizational structure, S t-1 is the average tissue structure change at time point t-1.

[0042] As a further solution of the present invention, the step of obtaining the deviation dynamic prediction analysis result is specifically as follows:

[0043] Based on the results of tracking the dynamic changes of the tissue structure, the interstitial changes in each region were calculated using the formula:

[0044]

[0045] An assessment of intercellular stroma changes in each region was generated;

[0046] Among them, M k For each region, the cell matrix changes were evaluated. T represents the total number of time points, S represents the total number of time points, t (k) represents the organizational structure change of region k at time t, α t represents the weight at time point t, ω k is the target weight of region k;

[0047] The intercellular matrix changes assessed for each region were integrated using the formula:

[0048]

[0049] Calculate the evolution analysis results of the lesion area;

[0050] Among them, E k is the evolution analysis result of the lesion area, K is the total number of areas, and μ is the mean of the regional evaluation;

[0051] Based on the evolution analysis results of the lesion area, dynamic description and prediction are performed using the formula:

[0052]

[0053] Generate dynamic prediction analysis results of the migration;

[0054] Among them, P is the result of dynamic prediction analysis of migration, β k is the prediction weight of region k.

[0055] As a further solution of the present invention, the step of acquiring the enhanced offset signal is specifically:

[0056] Based on the results of the dynamic prediction analysis of the displacement, high-dimensional manifold mapping of cell morphology and tissue structure is performed using the formula:

[0057]

[0058] Generate high-dimensional mapping data;

[0059] Among them, M s is the high-dimensional mapping data, n represents the number of cells in the analysis, and F i represents the morphological characteristics of the cell, λ is the attenuation parameter, and d i is the Euclidean distance from the cell to the center of the high-dimensional manifold;

[0060] The continuous changes of nuclear features during differentiation stages are tracked from the high-dimensional mapping data using the formula:

[0061]

[0062] Generate continuous change tracking results;

[0063] Among them, C t is the result of continuous change tracking, m is the length of the time series, M sj represents the high-dimensional mapping data at time point j, ω j is the weight at time point j;

[0064] Based on the continuous change tracking results, the high-dimensional dynamic characteristics of gastric cancer lesions are analyzed using the formula:

[0065]

[0066] Generate high-dimensional dynamic features;

[0067] Among them, D f is a high-dimensional dynamic feature, m is the number of reference time points, is the continuous change tracking result at time point k, and μ is the mean of the tracking result;

[0068] Based on the high-dimensional dynamic characteristics, the formula is adopted:

[0069]

[0070] generating an enhanced excursion signal;

[0071] Among them, S e To enhance the offset signal, q is the total number of dynamic features, is the high-dimensional dynamic eigenvalue of feature p, β p is the feature weight.

[0072] As a further solution of the present invention, the steps of obtaining the cell evolution analysis results are specifically as follows:

[0073] Based on the enhanced offset signal, the formula is adopted:

[0074]

[0075] Identify multiple evolutionary stages of gastric cancer lesions and generate identification data of multiple evolutionary stages;

[0076] Among them, S p is the identification data of multiple evolution stages, n is the number of evolution stages, s i The enhanced offset signal for each stage, S j is the strength of the signal at each stage, ∈ is a small constant;

[0077] Based on the recognition data of the multiple evolution stages, the characteristics of the lesion area are analyzed, and the formula is used to combine the local and global characteristics:

[0078]

[0079] Generate feature fusion data;

[0080] Among them, F l is the feature fusion data, m is the number of regions analyzed, and f k,local The local intercellular matrix characteristics, f k,global To correspond to the global characteristics of the organizational structure, σ is a normalization constant;

[0081] The feature fusion data of multiple regions are integrated to perform dynamic analysis of cell evolution using the formula:

[0082]

[0083] Generate cell evolution analysis results;

[0084] Among them, E d is the result of cell evolution analysis, q is the number of reference feature fusion regions, is the feature fusion data of region p, β p is the weight of region p in the analysis.

[0085] Compared with the prior art, the advantages and positive effects of the present invention are:

[0086] In the present invention, by deeply analyzing the cell morphology and tissue structure in gastric cancer pathology images, the evolution of the lesion area is accurately tracked. This processing method extracts the time series of spatial features, which not only reveals the subtle changes in the characteristics of the cell nucleus and the tissue boundary, but also maps these changes in the high-dimensional manifold space, capturing the continuous changes of the cell nucleus characteristics at different stages. This dynamic description can fully display the evolution process of gastric cancer lesions and further enhance the dynamic monitoring capability of pathological analysis. This highly dynamic tracking and mapping enhances the recognition of the evolutionary stage of gastric cancer, thereby enabling pathologists to analyze the condition based on richer data, improve the accuracy of formulating treatment strategies, and significantly improve the efficiency of medical diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 is a system flow chart of the present invention;

[0088] Figure 2 is a flow chart of the steps for obtaining the key morphological features of the present invention;

[0089] Figure 3 A flow chart of a method for tracking dynamic changes in an organizational structure according to the present invention;

[0090] Figure 4 A flowchart of the steps for obtaining the results of the dynamic prediction analysis of the displacement of the present invention;

[0091] Figure 5 A flowchart of the steps for obtaining the enhanced offset signal of the present invention;

[0092] Figure 6 The figure is a flow chart of the steps for obtaining the cell evolution analysis results of the present invention. DETAILED DESCRIPTION

[0093] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0094] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0095] Embodiment 1

[0096] See also Figure 1 , Gastric cancer pathology image analysis system includes:

[0097] The spatiotemporal feature extraction module extracts the spatial features of gastric cancer pathological images based on cell morphology and tissue structure, analyzes the time series data in the lesion area, calculates the statistical features of cell morphology in multiple regions, compares the differences between cell nuclear features and tissue boundaries, obtains spatial feature time series, and generates key morphological features;

[0098] The dynamic change tracking module tracks the dynamic changes in tissue structure based on key morphological features, calculates the change rules of intercellular matrix, analyzes the evolution of the lesion area in the time series, obtains the dynamic description of the lesion process, and obtains the deviation dynamic prediction analysis results;

[0099] The high-dimensional manifold mapping module maps the cell morphology and tissue structure in the high-dimensional manifold space based on the results of the dynamic prediction analysis of the offset, tracks the continuous changes of the cell nuclear characteristics in the differentiation stage, obtains the high-dimensional dynamic characteristics of gastric cancer lesions, and generates enhanced offset signals;

[0100] The evolutionary pattern recognition module identifies and classifies the multiple evolutionary stages of gastric cancer lesions based on the enhanced offset signal, analyzes the characteristics of the lesion area, obtains the fusion result of the local characteristics of the intercellular matrix and the global characteristics of the tissue structure, and generates the cell evolution analysis results.

[0101] The key morphological features include cell nuclear variability, tissue hierarchy, and cell arrangement symmetry. The results of the dynamic prediction analysis of the offset include the dynamic pattern index, the change speed index, and the time dependence index. The enhanced offset signal includes the signal intensity level, signal contrast, and signal spatial resolution. The results of the cell evolution analysis are specifically the cell differentiation level, evolution rate, and evolution direction.

[0102] See also Figure 2 , the steps to obtain key morphological features are as follows:

[0103] Based on cell morphology and tissue structure, cell morphological features are extracted using the formula:

[0104]

[0105] Calculate the size, shape and density characteristics of the cell nucleus to obtain statistical data of the cell nucleus characteristics;

[0106] Among them, Size is the size characteristic of the cell nucleus, Shape is the shape characteristic of the cell nucleus, N represents the number of cell nuclei, and A i represents the area of ​​the ith cell nucleus, P represents the perimeter of the cell nucleus, A represents the area of ​​the cell nucleus, and λ i represents the area attenuation coefficient, γ is the shape adjustment coefficient;

[0107] The statistical data of cell nuclear characteristics were analyzed and the differences between cell nuclear characteristics and tissue demarcation lines were compared using the formula:

[0108]

[0109] Generates the difference between the cell nucleus and tissue boundaries;

[0110] Among them, X cell An array of eigenvalues ​​representing the cell nucleus, X boundary represents the array of eigenvalues ​​associated with organizational boundaries, σ represents the standard deviation of the eigenvalues, α is the difference adjustment coefficient, and Δ represents the difference;

[0111] According to the difference between the cell nucleus and the tissue boundary, the statistical characteristics of cell morphology in multiple regions are calculated using the formula:

[0112]

[0113] Get the spatial feature time series;

[0114] Among them, x k represents the difference between the cell nucleus and the tissue boundary of the kth region, w k represents the relative importance of the region in the overall analysis, δ is the smoothing parameter, and F(x) represents the weighted spatial feature time series;

[0115] According to the spatial feature time series, the morphological features are calculated using the formula:

[0116]

[0117] Obtain key morphological features;

[0118] Where T represents the length of the time series, f t represents the spatial feature value at time point t, β is the time decay factor, and KeyFeatures represents the key morphological features.

[0119] formula

[0120]

[0121] Parameter Explanation

[0122] N: number of nuclei.

[0123] A i : The area of ​​the ith cell nucleus.

[0124] λ i : Area attenuation coefficient, which affects the calculation of the cell nucleus size and usually depends on the density or other biological properties of the cell nucleus.

[0125] Derivation process and examples

[0126] Assume there are three cell nuclei with areas of A1 = 10 μm 2 , A2=15μm 2 , A3=12μm 2 Assume that the attenuation coefficients are λ1=0.1, λ2=0.2, λ3=0.15. Calculate the average size:

[0127]

[0128] Calculate each term:

[0129]

[0130] formula

[0131]

[0132] Parameter Explanation

[0133] α: Difference adjustment coefficient, which adjusts the difference sensitivity.

[0134] X cell : array of cell nucleus eigenvalues.

[0135] X boundary : Array of tissue boundary eigenvalues.

[0136] σ: standard deviation of the eigenvalues.

[0137] Derivation process and examples

[0138] Assume X cell =[3,3.5,3.2],X boundary =[2.5, 2.8, 2.7], the mean and standard deviation are:

[0139] avg(X cell )=3.23

[0140] avg(X boundary )=2.67

[0141] σ=0.3

[0142] α=1.5

[0143]

[0144] formula

[0145]

[0146] Parameter Explanation

[0147] x k : The difference between the cell nucleus and the tissue boundary of the kth region.

[0148] w k : The relative importance of the region in the overall analysis.

[0149] δ: Smoothing parameter that prevents the denominator from reaching zero and reduces the impact of high variance values.

[0150] Derivation process and examples

[0151] Suppose there are three regions, the difference x k =[1.2,1.5,1.1], weight w k =[0.5,0.3,0.2], smoothing parameter δ = 0.05:

[0152]

[0153] Calculate each term:

[0154] Item 1:

[0155] Item 2:

[0156] Item 3:

[0157]

[0158] formula

[0159]

[0160] Parameter Explanation

[0161] T: length of the time series.

[0162] f t : The spatial eigenvalue at time point t.

[0163] β: time decay factor.

[0164] Derivation process and examples

[0165] Assume T = 3, f t =[0.4,0.42,0.38], β=0.9:

[0166]

[0167] Calculate each term:

[0168] First item: 0.4 0.9 = 0.36

[0169] Second term: 0.42 0.81 = 0.3402

[0170] The third term: 0.38 0.729 = 0.27702

[0171]

[0172] See also Figure 3 , the specific method for tracking dynamic changes in organizational structure is:

[0173] Based on key morphological features, the changes in tissue structure at each time point were analyzed using the formula:

[0174]

[0175] Generate an assessment of organizational structural changes at time point t;

[0176] Among them, C t is the assessment of tissue structural changes at time point t, n is the number of key morphological features, and f i (t) is the i-th morphological feature at time point t, λ is the time attenuation coefficient, μ i is the weight coefficient of the i-th feature;

[0177] The evaluation of organizational structure changes at time point t is weighted averaged using the formula:

[0178]

[0179] Generates average organizational structure changes over the time period;

[0180] Among them, S t is the average organizational structure change in the time period, k is the time window length, ω j is the weighting factor at time point j;

[0181] The average organizational structure change during the analysis period is calculated using the formula:

[0182]

[0183] Generate tracking results of dynamic changes in organizational structure;

[0184] Among them, T t To track the dynamic changes in organizational structure, S t-1 is the average tissue structure change at time point t-1.

[0185] formula:

[0186]

[0187] Parameter explanation:

[0188] n: number of key morphological features.

[0189] f i (t): the i-th morphological feature value at time point t.

[0190] λ: time decay coefficient, used to reduce the impact of old data.

[0191] μ i : The weight coefficient of the i-th feature.

[0192] Calculation process and examples:

[0193] Assume that at a certain time point t, there are three morphological features (n = 3), with values ​​of f1(t) = 2, f2(t) = 3, and f3(t) = 4. The time decay coefficient λ = 0.1, and the weight coefficients of each feature are μ1 = 0.5, μ2 = 0.3, and μ3 = 0.2.

[0194] Calculation process:

[0195]

[0196] Assume t = 5,

[0197] e -0.1·5 =e -0.5 ≈0.606

[0198]

[0199] Result interpretation:

[0200] The result 0.5672 represents the evaluation value of tissue structure change at time point t=5, and a smaller value indicates that the tissue structure change is not significant at this time point.

[0201] formula:

[0202]

[0203] Parameter explanation:

[0204] C j : Assessment of organizational structure changes at time point j.

[0205] k: time window length, assuming k=3.

[0206] ω j : Weighting factor at time point j.

[0207] Calculation process and examples:

[0208] Assume C3=0.5672, C4=0.6201, C5=0.6500.

[0209] The weighting factors are ω3=0.2, ω4=0.5, ω5=0.3.

[0210] Calculation process:

[0211]

[0212] S5=0.61845

[0213] Result interpretation:

[0214] The result 0.61845 represents the mean organizational structure change value, the average performance in the three time points considered.

[0215] formula:

[0216]

[0217] Parameter explanation:

[0218] S t : Average tissue structure change at time point t.

[0219] S t-1 : Average tissue structure change at time point t-1.

[0220] Calculation process and examples:

[0221] Assume S4 ​​= 0.5900,

[0222]

[0223] T5=0.04822

[0224] Result interpretation: The result 0.04822 represents the rate of tissue structure change, which indicates that the tissue structure changed significantly from time point t-1 to t.

[0225] See also Figure 4 , the specific steps for obtaining the results of the offset dynamic prediction analysis are:

[0226] Based on the results of tracking the dynamic changes in tissue structure, the interstitial changes in each region were calculated using the formula:

[0227]

[0228] An assessment of intercellular stroma changes in each region was generated;

[0229] Among them, M k For each region, the cell matrix changes were evaluated. T represents the total number of time points, S represents the total number of time points, t (k) represents the organizational structure change of region k at time t, α t represents the weight at time point t, ω k is the target weight of region k;

[0230] The evaluation of interstitial changes in each region was integrated using the formula:

[0231]

[0232] Calculate the evolution analysis results of the lesion area;

[0233] Among them, E k is the evolution analysis result of the lesion area, K is the total number of areas, and μ is the mean of the regional evaluation;

[0234] The evolution analysis results of the lesion area are integrated to perform dynamic description and prediction using the formula:

[0235]

[0236] Generate dynamic prediction analysis results of the migration;

[0237] Among them, P is the result of dynamic prediction analysis of migration, β k is the prediction weight of region k.

[0238] formula:

[0239]

[0240] Parameter interpretation and derivation process:

[0241] S t (k): The organizational structure change value of region k at time t. The value is assumed to be randomly generated.

[0242] α t : The weight at time point t, assumed to decrease linearly over time, for example, from 1.0 to 0.1 at 10 time points.

[0243] ω k : The specific weight of region k, assumed to be set according to the activity of the region.

[0244] T: total time points, assumed to be 10.

[0245] K: The total number of regions, assumed to be 5.

[0246] Assumptions and examples:

[0247] S t (k) is the change value of region k at 10 time points, which is randomly generated, for example

[0248] [0.5,0.6,0.55,0.59,0.62,0.65,0.66,0.68,0.7,0.72]. α t is [1.0, 0.9, 0.8, 0.7, 0.6, 0.5, 0.4, 0.3, 0.2, 0.1].

[0249] ω k Assume that is the weight of region k, and the weight of k=1 is 0.5.

[0250] Calculation process:

[0251]

[0252] formula:

[0253]

[0254] Parameter explanation:

[0255] M k : The evaluation value of intercellular matrix changes in the kth region, obtained from step 1. μ: The mean of the evaluation values ​​of all regions, calculated from M k value.

[0256] K: The total number of regions, assumed to be 5.

[0257] Assumptions and examples:

[0258] Assume that the M calculated in step 1 is k The values ​​are [0.5, 0.6, 0.55, 0.59, 0.62].

[0259] Calculate the mean μ:

[0260]

[0261] Calculation process:

[0262]

[0263] formula:

[0264]

[0265] Parameter explanation:

[0266] E k : Evolution analysis results of region k, obtained from step 2.

[0267] β k : The prediction weight of region k, assuming that the weight varies across regions.

[0268] K: The total number of regions, assumed to be 5.

[0269] Assumptions and examples:

[0270] Assume β k is [1,2,1,1,2].

[0271] Calculation process:

[0272]

[0273] P = 0.0426

[0274] Result interpretation:

[0275] This value P = 0.0426 indicates a balanced dynamic description result of the overall prediction, and the numerical value reflects a quantitative assessment of the average lesion area change trend. The numbers used in each step are hypothetical values ​​used to demonstrate the calculation process and the derivation of the results.

[0276] See also Figure 5 , the specific steps for obtaining the enhanced offset signal are:

[0277] Based on the results of the offset dynamic prediction analysis, high-dimensional manifold mapping of cell morphology and tissue structure is performed using the formula:

[0278]

[0279] Generate high-dimensional mapping data;

[0280] Among them, M s is the high-dimensional mapping data, n represents the number of cells in the analysis, and F i represents the morphological characteristics of the cell, λ is the attenuation parameter, and d i is the Euclidean distance from the cell to the center of the high-dimensional manifold;

[0281] To track the continuous changes of nuclear features during differentiation stages from high-dimensional mapping data, the formula is used:

[0282]

[0283] Generate continuous change tracking results;

[0284] Among them, Ct is the result of continuous change tracking, m is the length of the time series, represents the high-dimensional mapping data at time point j, ω j is the weight at time point j;

[0285] Based on the continuous change tracking results, the high-dimensional dynamic characteristics of gastric cancer lesions were analyzed using the formula:

[0286]

[0287] Generate high-dimensional dynamic features;

[0288] Among them, D f is a high-dimensional dynamic feature, m is the number of reference time points, is the continuous change tracking result at time point k, and μ is the mean of the tracking result;

[0289] Comprehensive high-dimensional dynamic features, using the formula:

[0290]

[0291] generating an enhanced excursion signal;

[0292] Among them, S e To enhance the offset signal, q is the total number of dynamic features, is the high-dimensional dynamic eigenvalue of feature p, β p is the feature weight.

[0293] formula:

[0294]

[0295] Parameter explanation:

[0296] F i : The morphological characteristic value of the i-th cell, such as the area or perimeter of the cell nucleus.

[0297] λ: Attenuation parameter, controlling the effect of distance on mapping.

[0298] d i : The Euclidean distance from the ith cell to the center of the high-dimensional manifold.

[0299] n: number of cells involved in the analysis.

[0300] Calculation process:

[0301] The impact of calculating the distance in high-dimensional space: For each cell, calculate where λ is assumed to be 0.5, d i is the distance from the cell to the center, assuming the distances are 1, 2, and 3 respectively.

[0302] Normalization: Normalize the result of the previous step by dividing it by the The sum of .

[0303] Weighted sum: morphological feature value F of each cell i Multiply it by its normalized distance effect, and then sum it up assuming F i They are 100, 150 and 200 respectively.

[0304] Calculation example:

[0305] Calculate e -0.5×1 =0.607, e -0.5×2 =0.368, e -0.5×3 =0.223.

[0306] Normalizing these values: 0.607 / (0.607+0.368+0.223)=0.475, 0.368 / 1.198=0.307, 0.223 / 1.198=0.186.

[0307] Calculate the weighted sum: 100×0.475+150×0.307+200×0.186=47.5+46.05+37.2=130.75.

[0308] Result description:

[0309] Final result s = 130.75 represents the comprehensive value of high-dimensional mapping data, which combines the effects of morphological features and spatial positions.

[0310] formula:

[0311]

[0312] Parameter explanation:

[0313] High-dimensional mapping data at time point j.

[0314] ω j : The weight of time point j, reflecting the importance of the time point.

[0315] m: length of the time series.

[0316] Calculation process:

[0317] Calculate the sum of squares of weights: weight ω for each time point j Square it, then sum it.

[0318] Calculate weighted mapping data: mapping data for each time point Multiply by the corresponding weight ω j, and then sum.

[0319] Calculate the ratio: Divide the sum of the weighted map data by the sum of the squares of the weights.

[0320] Calculation example:

[0321] Assume m = 3, ω j They are 1, 2, and 3 respectively.

[0322] Assumptions They are 130.75 (from the above example), 140, and 120 respectively.

[0323] Calculate the sum of the squares of the weights: 1 2 +2 2 +3 2 =1+4+9=14.

[0324] Calculate the sum of weighted mapping data: 130.75×1+140×2+120×3=130.75+

[0325] 280+360=770.75.

[0326] Calculation ratio: 770.75 / 14≈55.05.

[0327] Result description:

[0328] Final result C t =55.05 represents the average value of the weighted high-dimensional mapping data of the time series, which combines the influence of time weight and mapping data.

[0329] formula:

[0330]

[0331] Parameter explanation:

[0332] Continuous change tracking results at time point k.

[0333] μ: mean of the tracking results.

[0334] m: The number of time points considered.

[0335] Calculation process:

[0336] Calculate the mean of the tracking results: First calculate the tracking results of all time points The average value of .

[0337] Calculate the sum of squared deviations: For each tracking result at each time point, calculate its deviation from the mean μ, then square it and sum it.

[0338] To find the RMS: divide the sum of the squared deviations by the number of time points m, and then take the square root of the result.

[0339] Calculation example:

[0340] Continue using the result from step 2 C t =55.05 as an example value for all time points.

[0341] Assume m = 3, other time points Assume it is 50 or 60.

[0342] Calculate the mean μ=(55.05+50+60) / 3=55.02.

[0343] Calculate the sum of squared deviations: (55.05-55.02) 2 +(50-55.02) 2 +(60-

[0344] 55.02) 2 =0.0009+25.2004+24.8004=50.0017.

[0345] Find the RMS:

[0346] Result description:

[0347] Final result D f = 4.08 represents the consistency and abnormality of changes in the time series and is used to quantify the volatility of high-dimensional dynamic features.

[0348] formula:

[0349]

[0350] Parameter explanation:

[0351] The high-dimensional dynamic eigenvalue of feature p.

[0352] β p : Feature weight, assigned according to the predictive value of the feature.

[0353] q: total number of dynamic features.

[0354] Calculation process:

[0355] Calculate the square root sum of weights: Calculate all feature weights β p The square root of the sum of the squares of .

[0356] Calculate the weighted feature sum: Each dynamic feature Multiply it by its weight β p , and find the sum.

[0357] Calculate the ratio: divide the sum of the weighted features by the square root of the sum of the weights.

[0358] Calculation example:

[0359] Assume q=3, They are the result of the previous step, 4.08, and the other two assumed values, 4.00 and 4.20.

[0360] β p Assume 1, 2, 3.

[0361] Calculate the square root sum of weights:

[0362] Calculate the weighted feature sum: 4.08×1+4.00×2+4.20×3=4.08+8.00+

[0363] 12.60=24.68.

[0364] Calculation ratio: 24.68 / 3.74≈6.60.

[0365] Result description:

[0366] Final result e = 6.60 represents the strength of the enhanced offset signal calculated by the weighted and normalized method. This value represents the comprehensive evaluation of the dynamic features, taking into account the importance and contribution of each feature.

[0367] See also Figure 6 , the specific steps for obtaining the results of cell evolution analysis are:

[0368] Based on the enhanced offset signal, the formula is adopted:

[0369]

[0370] Identify multiple evolutionary stages of gastric cancer lesions and generate identification data of multiple evolutionary stages;

[0371] Among them, S p is the identification data of multiple evolution stages, n is the number of evolution stages, s i The enhanced offset signal for each stage, S j is the strength of the signal at each stage, ∈ is a small constant;

[0372] Based on the recognition data of multiple evolution stages, the characteristics of the lesion area are analyzed, and the local and global characteristics are combined to adopt the formula:

[0373]

[0374] Generate feature fusion data;

[0375] Among them, F l is the feature fusion data, m is the number of regions analyzed, and f k,local The local intercellular matrix characteristics, f k,global To correspond to the global characteristics of the organizational structure, σ is a normalization constant;

[0376] The feature fusion data of multiple regions are integrated to perform dynamic analysis of cell evolution using the formula:

[0377]

[0378] Generate cell evolution analysis results;

[0379] Among them, E d is the result of cell evolution analysis, q is the number of reference feature fusion regions, is the feature fusion data of region p, β p is the weight of region p in the analysis.

[0380] formula:

[0381]

[0382] Parameter Explanation

[0383] n: number of evolutionary stages.

[0384] s i : The strength of the enhanced offset signal in the i-th stage.

[0385] S j : The strength of the signal in the i-th stage.

[0386] ∈: A small constant used to prevent division by zero errors.

[0387] Hypothetical Example

[0388] Assume there are 3 evolutionary stages, among which:

[0389] s1=2,S1=100

[0390] s2=3,S2=150

[0391] s3=1,S3=50

[0392] ∈=0.01

[0393] Calculation process:

[0394] Calculate the total signal strength:

[0395] Calculate for each stage:

[0396]

[0397] Add the results: S p =-2.142-1.589-0.728=-4.459

[0398] Result interpretation:

[0399] This result represents a composite index that quantifies the importance and relative contribution of each stage in the progression of the disease, with lower values ​​indicating smaller contributions.

[0400] formula:

[0401]

[0402] Parameter Explanation

[0403] m: The number of regions analyzed.

[0404] f k，local : Local intercellular matrix characteristics of region k.

[0405] f k，global : Global characteristics of the organizational structure of region k.

[0406] σ: normalization constant.

[0407] Hypothetical Example

[0408] Assume there are 2 regions:

[0409] f 1，local =0.7, f 1，global =30

[0410] f 2，local =0.5, f 2，global =70

[0411] σ=5

[0412] Calculation process:

[0413] Compute the global feature sum:

[0414] For each region, calculate:

[0415]

[0416] Calculate the average:

[0417] Result interpretation:

[0418] The results show that by integrating local and global features, an average cell evolution feature score was obtained, which reflects the typical evolutionary characteristics of gastric cancer lesion areas.

[0419] formula:

[0420]

[0421] Parameter Explanation

[0422] q: The number of feature fusion regions considered.

[0423] Feature fusion data of region p.

[0424] β p : The weight of region p in the analysis.

[0425] Hypothetical Example

[0426] Assume that there are two feature fusion regions and use the F calculated above l data:

[0427]

[0428] β1=2,β2=3

[0429] Calculation process:

[0430] Calculate the sum of squared weights:

[0431] Perform a weighted sum for each region:

[0432]

[0433] Calculate the weighted average:

[0434] Result interpretation:

[0435] This result, E d represents the weighted regional feature fusion index, where the larger weight β p It indicates that a certain region or feature has a higher relative importance in the cell evolution analysis. The final value of 0.325 represents the average evolution feature strength after comprehensive consideration of the weights.

[0436] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. Gastric cancer pathology image analysis system, characterized in that: The system comprises: The spatiotemporal feature extraction module extracts the spatial features of gastric cancer pathological images based on cell morphology and tissue structure, analyzes the time series data in the lesion area, calculates the statistical features of cell morphology in multiple regions, compares the differences between cell nuclear features and tissue boundaries, obtains spatial feature time series, and generates key morphological features; The dynamic change tracking module tracks the dynamic changes in the tissue structure based on the key morphological features, calculates the change rules of the intercellular matrix, analyzes the evolution of the lesion area in the time series, obtains the dynamic description of the lesion process, and obtains the deviation dynamic prediction analysis results; The high-dimensional manifold mapping module maps the cell morphology and tissue structure in the high-dimensional manifold space based on the dynamic prediction analysis results of the displacement, tracks the continuous changes of the cell nuclear characteristics in the differentiation stage, obtains the high-dimensional dynamic characteristics of the gastric cancer lesions, and generates an enhanced displacement signal; The evolution pattern recognition module identifies and classifies multiple evolution stages of gastric cancer lesions based on the enhanced offset signal, analyzes the characteristics of the lesion area, obtains a fusion result of the local characteristics of the intercellular matrix and the global characteristics of the tissue structure, and generates a cell evolution analysis result; The steps for obtaining the key morphological features are specifically as follows: Based on cell morphology and tissue structure, cell morphological features are extracted using the formula: Calculate the size, shape and density characteristics of the cell nucleus to obtain statistical data of the cell nucleus characteristics; Among them, Size is the size characteristic of the cell nucleus, Shape is the shape characteristic of the cell nucleus, N represents the number of cell nuclei, and A i represents the area of ​​the ith cell nucleus, P represents the perimeter of the cell nucleus, A represents the area of ​​the cell nucleus, and λ i represents the area attenuation coefficient, γ is the shape adjustment coefficient; The statistical data of the cell nuclear characteristics are analyzed, and the differences between the cell nuclear characteristics and the tissue demarcation line are compared, using the formula: Generates the difference between the cell nucleus and tissue boundaries; Among them, X cell An array of eigenvalues ​​representing the cell nucleus, X boundary represents the array of eigenvalues ​​associated with organizational boundaries, σ represents the standard deviation of the eigenvalues, α is the difference adjustment coefficient, and Δ represents the difference; According to the difference between the cell nucleus and the tissue boundary, the statistical characteristics of cell morphology in multiple regions are calculated using the formula: Get the spatial feature time series; Among them, x k represents the difference between the cell nucleus and the tissue boundary of the kth region, w k represents the relative importance of the region in the overall analysis, δ is the smoothing parameter, and F(x) represents the weighted spatial feature time series; According to the spatial feature time series, the morphological features are calculated using the formula: Obtain key morphological features; Where T represents the length of the time series, f t represents the spatial feature value at time point t, β is the time decay factor, and KeyFeatures represents the key morphological features.

2. The gastric cancer pathology image analysis system according to claim 1, characterized in that: The method for tracking dynamic changes in the organizational structure is specifically: Based on the key morphological features, the changes in tissue structure at each time point were analyzed using the formula: Generate an assessment of organizational structural changes at time point t; Among them, C t is the assessment of tissue structural changes at time point t, n is the number of key morphological features, and f i (t) is the i-th morphological feature at time point t, λ is the time attenuation coefficient, μ i is the weight coefficient of the i-th feature; The evaluation of the organizational structure change at the time point t is weighted averaged using the formula: Generates average organizational structure changes over the time period; Among them, S t is the average organizational structure change in the time period, k is the time window length, ω j is the weighting factor at time point j; To analyze the average tissue structure changes during the stated time period, the formula was used: Generate tracking results of dynamic changes in organizational structure; Among them, T t To track the dynamic changes in organizational structure, S t-1 is the average tissue structure change at time point t-1.

3. The gastric cancer pathology image analysis system according to claim 1, characterized in that: The steps for obtaining the deviation dynamic prediction analysis results are specifically as follows: Based on the results of tracking the dynamic changes of the tissue structure, the interstitial changes in each region were calculated using the formula: An assessment of intercellular stroma changes in each region was generated; Among them, M k For each region, the cell matrix changes were evaluated. T represents the total number of time points, S represents the total number of time points, t (k) represents the organizational structure change of region k at time t, α t represents the weight at time point t, ω k is the target weight of region k; The intercellular matrix changes assessed for each region were integrated using the formula: Calculate the evolution analysis results of the lesion area; Among them, E k is the evolution analysis result of the lesion area, K is the total number of areas, and μ is the mean of the regional evaluation; Based on the evolution analysis results of the lesion area, dynamic description and prediction are performed using the formula: Generate dynamic prediction analysis results of the migration; Among them, P is the result of dynamic prediction analysis of migration, β k is the prediction weight of region k.

4. The gastric cancer pathology image analysis system according to claim 1, characterized in that: The steps of acquiring the enhanced offset signal are specifically as follows: Based on the results of the dynamic prediction analysis of the displacement, high-dimensional manifold mapping of cell morphology and tissue structure is performed using the formula: Generate high-dimensional mapping data; Among them, M s is the high-dimensional mapping data, n represents the number of cells in the analysis, and F i represents the morphological characteristics of the cell, λ is the attenuation parameter, and d i is the Euclidean distance from the cell to the center of the high-dimensional manifold; The continuous changes of nuclear features during differentiation stages are tracked from the high-dimensional mapping data using the formula: Generate continuous change tracking results; Among them, C t is the result of continuous change tracking, m is the length of the time series, represents the high-dimensional mapping data at time point j, ω j is the weight at time point j; Based on the continuous change tracking results, the high-dimensional dynamic characteristics of gastric cancer lesions are analyzed using the formula: Generate high-dimensional dynamic features; Among them, D f is a high-dimensional dynamic feature, m is the number of reference time points, is the continuous change tracking result at time point k, and μ is the mean of the tracking result; Based on the high-dimensional dynamic characteristics, the formula is adopted: generating an enhanced excursion signal; Among them, S e To enhance the offset signal, q is the total number of dynamic features, is the high-dimensional dynamic eigenvalue of feature p, β p is the feature weight.

5. The gastric cancer pathology image analysis system according to claim 1, characterized in that: The steps for obtaining the cell evolution analysis results are specifically as follows: Based on the enhanced offset signal, the formula is adopted: Identify multiple evolutionary stages of gastric cancer lesions and generate identification data of multiple evolutionary stages; Among them, S p is the identification data of multiple evolution stages, n is the number of evolution stages, s i The enhanced offset signal for each stage, S j is the strength of the signal at each stage, ∈ is a small constant; Based on the recognition data of the multiple evolution stages, the characteristics of the lesion area are analyzed, and the formula is used to combine the local and global characteristics: Generate feature fusion data; Among them, F l is the feature fusion data, m is the number of regions analyzed, and f k,local The local intercellular matrix characteristics, f k,global To correspond to the global characteristics of the organizational structure, σ is a normalization constant; The feature fusion data of multiple regions are integrated to perform dynamic analysis of cell evolution using the formula: Generate cell evolution analysis results; Among them, E d is the result of cell evolution analysis, q is the number of reference feature fusion regions, is the feature fusion data of region p, β p is the weight of region p in the analysis.

Images