A liver cancer prognosis evaluation method and system based on cell pixel density

Abstract

The application discloses a hepatocellular carcinoma prognosis evaluation method and system based on cell pixel density, and relates to the technical field of hepatocellular carcinoma prognosis evaluation. The method comprises the following steps: acquiring whole slide imaging and performing image preprocessing to determine a critical value of cell pixel level segmentation result; performing cell pixel density calculation according to the critical value of the cell pixel level segmentation result to determine the CD8+TILs density of a tumor area; performing standardization and average calculation processing on the CD8+TILs density of the tumor area to obtain a critical value of ATLS-8; and performing prognosis evaluation on hepatocellular carcinoma of a patient according to the critical value of ATLS-8. By using the application, the cell density on the whole slide can be accurately quantified, the accuracy of cell density measurement is improved, and the accuracy of prognosis evaluation on hepatocellular carcinoma of a patient is improved. The application can be widely applied to the technical field of hepatocellular carcinoma prognosis evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hepatocellular carcinoma prognosis evaluation, and in particular to a hepatocellular carcinoma prognosis evaluation method and system based on cell pixel density. BACKGROUND

[0002] For the complexity of tumor heterogeneity, the related technology is committed to determine new prognostic parameters and indicators as comprehensive biomarkers for a wider range of HCC patients. At present, a variety of biomarkers have been discovered and used for the description of tumor biological characteristics and prognosis evaluation. The tumor microenvironment (TME) is composed of tumor cells, stromal cells, immune cells and epithelial tissues, which is an important component affecting tumor survival and progression. Tumor heterogeneity plays a crucial role in limiting the effectiveness of drugs and radiotherapy widely accepted by cancer patients, and is closely related to TME. TILs are widely considered to be immune cells related to the prognosis of patients with solid tumors. This includes CD3, CD8 and CD45 lymphocytes and other subtypes, among which CD8+ T cells are confirmed to be key members of tumor cell elimination. The proportion of T lymphocytes in the tumor core and infiltrating edge has been verified for its effectiveness and versatility in various cancer types. Various scoring methods centered on TILs have been widely studied and used for prognosis evaluation of various tumor types. However, for liver cancer, there is currently no widespread consensus. In previous studies, some pathologists reported that the immune response of tissues and tumor regions was manually scored using traditional pathological methods to evaluate prognosis. These methods have limitations and subjectivity, leading to conflicting results in different studies. Due to the large number of cells in WSI, traditional pathological quantification methods are almost impossible to accurately quantify the cell density on the whole slide. Traditional methods often have bias when selecting ROI, and are heavily dependent on the subjective expertise of pathologists, which affects the accuracy of cell density measurement and limits it to qualitative or semi-quantitative evaluation. SUMMARY

[0003] In order to solve the above technical problems, the purpose of the present application is to provide a hepatocellular carcinoma prognosis evaluation method and system based on cell pixel density, which can accurately quantify the cell density on the whole slide, thereby improving the accuracy of cell density measurement and improving the accuracy of hepatocellular carcinoma prognosis evaluation for patients.

[0004] The first technical solution adopted by the present application is: a hepatocellular carcinoma prognosis evaluation method based on cell pixel density, comprising the following steps:

[0005] Obtaining whole slide imaging and performing image preprocessing to determine the critical value of cell pixel-level segmentation results;

[0006] According to the critical value of the cell pixel level segmentation result, the cell pixel density is calculated to determine the CD8+TILs density of the tumor area;

[0007] The CD8+TILs density of the tumor area is standardized and averaged to obtain the critical value of ATLS-8;

[0008] According to the critical value of ATLS-8, the prognosis of hepatocellular carcinoma of the patient is evaluated.

[0009] Further, the step of obtaining whole slide imaging and image preprocessing to determine the critical value of the cell pixel level segmentation result specifically includes:

[0010] Image preprocessing is performed on the whole slide imaging to obtain a tumor area mask image;

[0011] The whole slide imaging and the tumor area mask image are subjected to positive cell recognition and segmentation to determine the critical value of the cell pixel level segmentation result.

[0012] Further, the step of image preprocessing the whole slide imaging to obtain a tumor area mask image specifically includes:

[0013] Obtain whole slide imaging and perform staining to obtain H&E stained section images;

[0014] Based on the quality control instrument, non-overlapping patches are extracted under 10 times magnification of the H&E stained section images, and patches smaller than a preset threshold are removed to obtain tumor area segmentation images;

[0015] The tumor area segmentation image is segmented by nnU-Net deep neural network to obtain a tumor area mask image.

[0016] Further, the step of performing positive cell recognition and segmentation on the whole slide imaging and the tumor area mask image to determine the critical value of the cell pixel level segmentation result specifically includes:

[0017] According to the H&E stained section images, the whole slide imaging with the same tissue staining but different whole slide imaging is registered to obtain aligned whole slide imaging;

[0018] The aligned whole slide imaging is imported into the Qupath platform for observation, and a tiled image is exported;

[0019] The tiled image and the tumor area mask image are subjected to registration and pixel segmentation processing by a random forest algorithm to determine the critical value of the cell pixel level segmentation result.

[0020] Further, the step of determining the CD8+ TILs density of the tumor region according to the critical value of the cell pixel-level segmentation result specifically comprises:

[0021] Determining the tumor region area and the CD8+ pixel area according to the critical value of the cell pixel-level segmentation result;

[0022] Determining the area proportion of CD8+ cells in the tumor tissue according to the tumor region area and the CD8+ pixel area;

[0023] Unit standardization is performed on the area proportion of CD8+ cells in the tumor tissue, pixel density is converted into cell number per unit area, and the CD8+ TILs density of the tumor region is obtained.

[0024] Further, the step of standardizing and average calculating the CD8+ TILs density of the tumor region to obtain the critical value of ATLS-8 specifically comprises:

[0025] The CD8+ TILs density of the tumor region is standardized by a normalization factor to obtain the ratio of normalized patch area to unit area;

[0026] The non-normalized cell density is obtained and multiplied by the ratio of normalized patch area to unit area to obtain the cell density per unit area of the patch;

[0027] The cell density per unit area of the patch is average calculated to obtain the critical value of ATLS-8.

[0028] The second technical solution adopted by the present application is: a hepatocellular carcinoma prognosis evaluation system based on cell pixel density, comprising:

[0029] The first module is used for acquiring whole slide imaging and performing image preprocessing to determine the critical value of the cell pixel-level segmentation result;

[0030] The second module is used for calculating the cell pixel density according to the critical value of the cell pixel-level segmentation result to determine the CD8+ TILs density of the tumor region;

[0031] The third module is used for standardizing and average calculating the CD8+ TILs density of the tumor region to obtain the critical value of ATLS-8;

[0032] The fourth module is used for performing prognosis evaluation on hepatocellular carcinoma of a patient according to the critical value of ATLS-8.

[0033] The method and system have the advantages that the application determines the critical value of the cell pixel segmentation result, and then determines the CD8+TILs density of the tumor area, standardizes and averages the CD8+TILs density of the tumor area, obtains the critical value of ATLS-8, and finally evaluates the prognosis of the patient's hepatocellular carcinoma according to the critical value of ATLS-8, which can accurately quantify the cell density on the whole slide, and further improve the accuracy of cell density measurement and the accuracy of prognosis evaluation of the patient's hepatocellular carcinoma. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 is a step flow chart of a hepatocellular carcinoma prognosis evaluation method based on cell pixel density according to the application;

[0035] Figure 2 is a structural block diagram of a hepatocellular carcinoma prognosis evaluation system based on cell pixel density according to the application;

[0036] Figure 3 is a result schematic diagram of selecting 40 patches for Bland-Altman consistency correlation analysis according to the specific embodiment of the application;

[0037] Figure 4 is a schematic diagram of Kaplan-Meier survival curve analysis of the discovery cohort according to the specific embodiment of the application;

[0038] Figure 5 is a schematic diagram of Kaplan-Meier survival curve analysis of the validation cohort 1 according to the specific embodiment of the application;

[0039] Figure 6 is a schematic diagram of Kaplan-Meier survival curve analysis of the validation cohort 2 according to the specific embodiment of the application. DETAILED DESCRIPTION

[0040] The application will be further described in detail below in combination with the drawings and specific embodiments. For the step numbers in the following embodiments, only the setting is for the convenience of description, and the order between the steps is not limited in any way, and the execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0041] The application introduces a new fully automatic process, which uses artificial intelligence algorithm to quantify CD8 tumor infiltrating lymphocytes, i.e. ATLS. In addition, it also uses a scoring system based on CD8+ lymphocytes (called ATLS-8) to evaluate the prognosis of HCC patients.

[0042] REFERENCE Figure 1 The application provides a hepatocellular carcinoma prognosis evaluation method based on cell pixel density, which comprises the following steps:

[0043] S100, acquiring whole slide imaging and performing image preprocessing to determine the critical value of cell pixel-level segmentation results;

[0044] S110, performing image preprocessing on whole slide imaging to obtain a tumor region mask image;

[0045] Specifically, whole slide imaging is acquired and staining processing is performed to obtain H&E stained section images; based on a quality control instrument, non-overlapping patches are extracted under 10 times magnification of the H&E stained section images, and patches smaller than a preset threshold are removed to obtain a tumor region to-be-segmented image; the tumor region to-be-segmented image is segmented by a nnU-Net deep neural network to obtain a tumor region mask image.

[0046] In this embodiment, in order to remove unnecessary white background, artifacts and other types of noise in WSI, professional pathologists use a quality control instrument to identify the available ROI in WSI. In order to obtain more accurate tumor segmentation results and higher efficiency, non-overlapping patches (1000x 1000 pixels) are extracted under 10 times magnification. Patches with a tissue proportion less than 0.05 will be discarded. Then, a pre-trained nnU-Net is used to segment the tumor region. Finally, pathologists are invited to help improve the tumor mask, which is obtained by aggregating patch-level masks.

[0047] S120, performing positive cell recognition and segmentation on whole slide imaging and the tumor region mask image to determine the critical value of cell pixel-level segmentation results.

[0048] Specifically, the whole slide imaging with the same tissue staining but different whole slide imaging is registered according to the H&E stained section images to obtain aligned whole slide imaging; the aligned whole slide imaging is imported into a Qupath platform for observation, and a tiled image is exported; the tiled image and the tumor region mask image are registered and pixel segmented by a random forest algorithm to determine the critical value of cell pixel-level segmentation results.

[0049] In this embodiment, to align adjacent tissue WSI stained with H&E and IHC, a method combining rigid registration and twice non-rigid registration was adopted. This method can ensure that the two slices are as similar as possible in spatial position and morphology. In this step, all WSI images of the same tissue but different staining were registered according to the H&E staining image as a reference for multiple IHC staining registration. Then the aligned image was imported into the Qupath platform to observe the registration effect. A 1024x1024 pixel tiled image was generated and exported. Subsequently, the mask was registered with the calibrated IHC slice using a Python program. The tumor and non-tumor regions were segmented into 1024x1024 pixels, and the blocks with more than 50% tissue in the mask could be used for further analysis.

[0050] Further, a random forest algorithm model was used to segment the tiles. Initially, a small amount of accurate annotation was made by a pathologist to observe the segmentation results. Then, according to these observations, the model was trained on a larger scale. The probability generated for each pixel represents the likelihood of the pixel becoming a target cell pixel. The critical value for determining the final cell pixel-level segmentation result is P = 0.25.

[0051] S200, according to the critical value of the cell pixel-level segmentation result, the cell pixel density is calculated to determine the CD8+ TILs density of the tumor area;

[0052] S210, according to the critical value of the cell pixel-level segmentation result, the area of the tumor area and the area of the CD8+ pixel are determined;

[0053] S220, according to the area of the tumor area and the area of the CD8+ pixel, the area proportion of CD8+ cells in the tumor tissue is determined;

[0054] In this embodiment, after cell segmentation is completed, the tiles are divided into tumor regions according to a predetermined strategy. By determining the area occupied by the pixels labeled as CD8+ in the entire segment, the pixel density of each segment is calculated. This measurement provides a WSI-level CD8+ cell pixel density, which represents the area proportion of CD8+ cells in the tumor tissue, and its expression is:

[0055]

[0056] wherein, Pixel CD8label counting patch means that the random forest algorithm is used to segment the pixels at the tile level. Therefore, Pixel total label countiong patchIt includes all pixels within the plaque, including CD8+ cells (1) and non-CD8+ cells (0). This method allows for the calculation of CD8+ cell density within the tumor region at the WSI level. The calculation formula is summarized below:

[0057]

[0058] Where n represents the number of available tumor plaques.

[0059] S230. The area ratio of CD8+ cells in the tumor tissue is standardized by unit, and the pixel density is converted into the number of cells per unit area to obtain the CD8+TILs density in the tumor region.

[0060] In this embodiment, to standardize with units used in most studies, pixel density is converted to cell count per unit area. This involves measuring the number of cells in each sample based on cell segmentation marker images. The cell count / mm is determined by calculating the area ratio and pixel size occupied by each cell. 2 This yields the CD8+ TILs density in the tumor region. The calculation formula is shown below:

[0061]

[0062] In the above formula, Pstch Cell density unnormalised This indicates the density of CD8+ TILs in the tumor region.

[0063] S300. The density of CD8+TILs in the tumor region is standardized and averaged to obtain the critical value of ATLS-8.

[0064] S310. The density of CD8+TILs in the tumor region is standardized by a normalization factor to obtain the ratio of normalized patch area to unit area.

[0065] S320. Obtain the non-normalized cell density and multiply it with the ratio of the normalized patch area to the unit area to calculate the cell density per unit area of ​​the patch.

[0066] S330. The cell density per unit area of ​​the patch is averaged to obtain the critical value of ATLS-8.

[0067] In this embodiment, after calculating the cell density of each patch, we normalize the pixel density at the three centers using a normalization factor to reflect the cell size. The calculation formula is as follows:

[0068]

[0069] Among them, Fmag represents the actual area of each patch, F nor represents the ratio of normalized patch area to unit area (unit: mm 2 ). In this experiment, Patch height height and Patch weight weight are both 1024 pixels. Then, F nor is multiplied by the non-normalized cell density, and the cell density per unit area (mm 2 ) of each patch is obtained. The calculation formula is as follows

[0070] Cell density patch-normalised = Cell density patch-unnormalised *F nor

[0071] Finally, the CD8+ cell density of all patches is averaged to reflect the CD8+ cell density in the tumor area on the WSI. This is called ATLS-8. The calculation formula is as follows

[0072]

[0073] where n represents the number of available tumor area patches.

[0074] S400, according to the critical value of ATLS-8, the prognosis of patients with hepatocellular carcinoma is evaluated.

[0075] In summary, the embodiments of the present application scan the H&E stained sections using a whole slide scanner, apply quality control measures, and use deep learning methods to segment images to obtain tumor area masks annotated by professional pathologists. The adjacent CD8 IHC stained sections are registered with the H&E images, the registered images are segmented into patches, and the annotated cells are identified and segmented using machine learning algorithms, the pixel density of positive cells is calculated and converted into cell density, the ATLS-8 critical value is determined according to the maximum selection rank statistics, the patients are divided into low index group and high index group, and the Kaplan-Meier (KM) curve is drawn to intuitively show the survival difference between the groups, as shown in Figure 4 The prognostic value of ATLS-8 is evaluated in two independent external validation groups.

[0076] Finally, the cell segmentation results of the embodiments of the present application are analyzed by simulation experiments, as shown in Figure 3Forty patches were randomly selected for cell segmentation. The segmentation results were compared with the manual counting results by pathologists using the traditional pathological counting method, and the pathologists ignored the results generated by the algorithm. The comparison results were analyzed using the Bland-Altman method. The consistency between the manual counting and automatic counting using our algorithm was good, with an ICC of 0.98 (95% CI, 0.97-0.99; P < 0.001).

[0077] The prognostic value of ATLS-8 was further analyzed, and the critical value of ATLS-8 was determined using the maximum selection rank statistic method. The critical value of the tumor area was 67.25 cells / mm 2 In the discovery cohort, 90 patients (72.58%) had an ATLS-8 value higher than the critical value and were classified as low risk, and 34 patients (27.42%) had an ATLS-8 value lower than the critical value and were classified as high risk. The 5-year overall survival (OS) rate of the low-risk group was 65.03%, and that of the high-risk group was 34.05%. As shown in FIG. 2A, the 5-year OS rate of the low-risk group was significantly higher than that of the high-risk group (P < 0.001). Figure 5 As shown in FIG. 2B, in validation cohort 1, 74 patients (90.24%) had an ATLS-8 value higher than the critical value and were classified as low risk, and 8 patients (9.76%) had an ATLS-8 value lower than the critical value and were classified as high risk. The 5-year OS rate of the low-risk group was 68.73%, and that of the high-risk group was 28.57%. As shown in FIG. 2B, the 5-year OS rate of the low-risk group was significantly higher than that of the high-risk group (P = 0.001). Figure 6 As shown in FIG. 2C, in validation cohort 2, 27 patients (50.0%) had an ATLS-8 value higher than the critical value and were classified as low risk, and 27 patients (50.0%) had an ATLS-8 value lower than the critical value and were classified as high risk. The 5-year OS rate of the low-risk group was 81.48%, and that of the high-risk group was 59.26%. As shown in FIG. 2C, the 5-year OS rate of the low-risk group was significantly higher than that of the high-risk group (P = 0.001).

[0078] KM curves showed that patients with lower ATLS-8 scores had poorer OS compared with patients with higher scores (discovery cohort, unadjusted HR, 2.23 (95% CI, 1.27-3.91); P = 0.0042; validation cohort 1, HR 3.38 (95% CI, 1.27-9.02) P = 0.0096; validation cohort 2, HR 2.74 (95% CI, 1.05-7.15); P = 0.031).

[0079] Table 1. Data table of univariate Cox regression analysis results of three cohorts

[0080]

[0081]

[0082] Table 2. Data table of multivariate Cox regression analysis results of three cohorts

[0083]

[0084]

[0085] Table 1 lists the results of univariate Cox regression analysis of the three cohorts, and Table 2 lists the results of multivariate Cox regression analysis of the three cohorts. In the univariate analysis of the discovery cohort, factors that reached statistical significance (P < 0.05) included age, HBV, tumor size, differentiation grade, MVI, BCLC stage, and ATLS-8, which were all included in the multivariate analysis to assess the robustness of ATLS-8 in the validation cohorts. Univariate and multivariate model analyses excluded data with unknown MVI status. Multivariate analysis showed that ATLS-8 score was independently associated with OS after adjusting for other clinicopathological factors (discovery cohort, low HR vs. high HR, 2.015 (95% CI: 1.115-3.641); P = 0.0203; validation cohort 1, HR, 4.871 (95% CI: 1.753-13.534); P = 0.0024; validation cohort 2, 3.925 ((95% CI: 1.212-12.714; P = 0.0226). Finally, for the development and validation of a prognostic prediction model, in the discovery cohort, age, HBV, ATLS-8, BCLC stage, tumor size, differentiation grade, and MVI were considered independent factors affecting OS, so we established a complete prognostic prediction model based on the above factors (age + HBV + ATLS-8 + BCLC stage + tumor size + differentiation grade + MVI) (hence we refer to this model as the complete model). As shown in Table 3, we further compared the performance of the complete model and the other four models, including the BCLC stage model, the ATLS-8 model, the BCLC stage and ATLS-8 combination model, and the clinicopathological model (combination model of age, HBV, BCLC stage, tumor size, differentiation grade, and MVI).

[0086] Table 3 Performance indicators of ATLS-8 model and multiple reference models

[0087]

[0088]

[0089] In all three cohorts, the complete model outperformed the clinicopathological model in terms of discriminative ability (evaluated by C-index) and calibration ability (evaluated by AIC). Specifically,

[0090] In the discovery cohort, the C-index was 0.770 for the full model vs 0.757 for the clinicopathological model; AIC was 381.6 vs 384.8 for the clinicopathological model. In validation cohort 1, the C-index was 0.769 vs 0.727, AIC was 285.8 vs 291.3. In validation cohort 2, the C-index was 0.712 vs 0.642, AIC was 152.7 vs 156.6. The BCLC staging and ATLS-8 model also showed better discriminative and calibration ability than the BCLC staging and ATLS-8 model.

[0091] In all three cohorts, the BCLC staging and ATLS-8 model also showed better discriminative and calibration ability than the BCLC staging model. In the discovery cohort, the C-index was 0.715 vs 0.655, AIC was 389.4 vs 392.8. In validation cohort 1, the C-index was 0.619 vs 0.674, AIC was 292.6 vs 296. In validation cohort 2, the C-index was 0.643 vs 0.516, AIC was 146.2 vs 160.

[0092] Inclusion of ATLS-8 score into the BCLC staging model improved the prediction of OS (likelihood ratio P=0.0013). In addition, inclusion of ATLS-8 score into the clinicopathological model also improved the prediction of OS (likelihood ratio P=0.0011).

[0093] Reference Figure 2 A hepatocellular carcinoma prognosis evaluation system based on cell pixel density, comprising:

[0094] A first module for acquiring whole slide imaging and performing image preprocessing to determine a critical value of cell pixel-level segmentation results;

[0095] A second module for calculating cell pixel density according to the critical value of the cell pixel-level segmentation results to determine the CD8+TILs density of the tumor area;

[0096] A third module for standardizing and averaging the CD8+TILs density of the tumor area to obtain a critical value of ATLS-8;

[0097] A fourth module for performing prognosis evaluation on hepatocellular carcinoma of a patient according to the critical value of ATLS-8.

[0098] The contents in the above method embodiments are all applicable to the system embodiments, the system embodiments specifically implement the same functions as the above method embodiments, and achieve the same beneficial effects as the above method embodiments.

[0099] The above is a specific description of the preferred embodiment of the application, but the application is not limited to the described embodiments, and those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the application, and these equivalent modifications or replacements are all included in the scope defined by the claims of the present application.

Claims

1. A hepatocellular carcinoma prognosis evaluation method based on cell pixel density, characterized by, The method comprises the following steps: full slide imaging is acquired and staining is performed to obtain an H&E stained section image; non-overlapping patches are extracted from the H&E stained section image under 10 times magnification based on a quality control instrument, and patches smaller than a preset threshold are removed to obtain a tumor region image to be segmented; segmentation is performed on the tumor region image to be segmented by an nnU-Net deep neural network to obtain a tumor region mask image; different full slide images with the same tissue staining are registered according to the H&E stained section image to obtain aligned full slide images; the aligned full slide images are imported into a Qupath platform for observation, and tiled images are exported; registration and pixel segmentation are performed on the tiled images and the tumor region mask image by a random forest algorithm to determine a critical value of a cell pixel-level segmentation result; areas of the tumor region and CD8+ pixels are determined according to the critical value of the cell pixel-level segmentation result; an area proportion of CD8+ cells in the tumor tissue is determined according to the areas of the tumor region and the CD8+ pixels; a CD8+ TIL density of the tumor region is obtained by converting a pixel density into a cell number per unit area through unit standardization of the area proportion of the CD8+ cells in the tumor tissue; standardization and average calculation are performed on the CD8+ TIL density of the tumor region to obtain a critical value of ATLS-8; prognosis of hepatocellular carcinoma of a patient is evaluated according to the critical value of ATLS-8.

2. The method of claim 1, wherein the method is based on the pixel density of the cell. The step of standardizing and average calculating the CD8+ TIL density of the tumor region to obtain the critical value of ATLS-8 specifically comprises: standardization is performed on the CD8+ TIL density of the tumor region by a normalization factor to obtain a ratio of a normalized patch area to a unit area; a non-normalized cell density is obtained and multiplied by the ratio of the normalized patch area to the unit area to obtain a cell density per unit area of a patch; average calculation is performed on the cell density per unit area of the patch to obtain the critical value of ATLS-8.

3. A hepatocellular carcinoma prognosis evaluation system based on cell pixel density, characterized by, The method comprises the following modules: a first module is configured to acquire full slide imaging and perform staining to obtain an H&E stained section image; non-overlapping patches are extracted from the H&E stained section image under 10 times magnification based on a quality control instrument, and patches smaller than a preset threshold are removed to obtain a tumor region image to be segmented; segmentation is performed on the tumor region image to be segmented by an nnU-Net deep neural network to obtain a tumor region mask image; different full slide images with the same tissue staining are registered according to the H&E stained section image to obtain aligned full slide images; the aligned full slide images are imported into a Qupath platform for observation, and tiled images are exported; registration and pixel segmentation are performed on the tiled images and the tumor region mask image by a random forest algorithm to determine a critical value of a cell pixel-level segmentation result; a second module is configured to determine areas of a tumor region and CD8+ pixels according to the critical value of the cell pixel-level segmentation result; According to the area of the tumor region and the area of the CD8+ pixels, the area proportion of the CD8+ cells in the tumor tissue is determined; The area proportion of the CD8+ cells in the tumor tissue is unit-normalized to convert the pixel density into the cell number per unit area, so as to obtain the CD8+ TILs density of the tumor region; The third module is configured to standardize and average the CD8+ TILs density of the tumor region to obtain a critical value of ATLS-8; The fourth module is configured to perform prognosis evaluation on the hepatocellular carcinoma of the patient according to the critical value of ATLS-8.

Images