Method for identifying hepatocellular carcinoma cell EMT and interstitial collagen characteristics through AI

Abstract

The invention discloses a method for identifying hepatocellular carcinoma cell EMT and interstitial collagen characteristics through AI, and the method comprises the steps: obtaining HCC morphological characteristics based on an immunohistochemical staining picture obtained in advance; the HCC morphological characteristics comprise capillary density and capillary area; wherein the morphological characteristics of the HCC are obtained on the basis of an immunohistochemical staining picture obtained in advance, and the method comprises the following steps: analyzing CD34-labeled HCC interstitial microvessels in the immunohistochemical staining picture in an HE-DAB state; adjusting a blood vessel threshold value in the immunohistochemical staining picture, so that the blood vessel is red, and the background of the blood vessel is white; filling up vascular cavities and eliminating edge effects; and calculating to obtain the microvessel density and microvessel area in the immunohistochemical staining picture, so as to accurately identify the occurrence of the HCC accompanied by the EMT.

Description

Technical Field

[0001] The present invention relates to a method for AI to identify EMT and interstitial collagen characteristics of hepatocellular carcinoma cancer cells, and belongs to the technical field of clinical medicine. Background Art

[0002] Traditional pathological diagnosis is done with the help of a microscope, through visual assessment and manual counting, but it is easily affected by subjective factors. Especially with the rise of personalized treatment, the accuracy of biomarker evaluation and the demand for precise pathological diagnostic information are increasing.

[0003] Hepatocellular carcinoma is the third leading cause of cancer death. Epithelial-mesenchymal transition (EMT) promotes cell proliferation, invasion and metastasis during the progression of hepatocellular carcinoma (HCC), which can lead to chemotherapy resistance and poor patient prognosis. 80% of HCC are accompanied by a background of liver fibrosis or cirrhosis, and HCC interstitial sclerosis can promote cancer progression. The proteins of the interstitial matrix are mainly composed of different types of collagen, fibronectin and elastin. With the development and progression of fibrosis, the amount and distribution of collagen will change. Therefore, evaluating the protein collagen component in the cancer stroma can well predict HCC cell proliferation, invasion and metastasis. However, the prior art lacks a method for evaluating the protein collagen component in the cancer stroma.

[0004] Immunohistochemistry (IHC) is widely used in clinical practice. When HCC undergoes EMT, cancer cells have low expression of E-cadherin, and TGF-β, Ki-67, Vimentin, and MMP7 are all significantly overexpressed. Cell atypia is obvious, differentiation becomes poor, and infiltration or metastasis occurs. Analysis of clinical pathological factors related to EMT showed that poor differentiation, AFP expression, Ki-67, MMP7, vascular metastasis, and recurrence are closely related to cancer EMT. The IHC method also has certain limitations. The predictive ability of single proteomic parameters is limited, and the potential biological information is easily affected by factors such as specimen fixation, sampling, reagent quality, and human factors.

[0005] Artificial Intelligence (AI) can provide intelligent models with image recognition capabilities that can rival or even surpass human experts, which provides strong support for accurate diagnosis and rapid development of pathology. Pathological images produced after AI training can distinguish histological details such as tumor grading and vascular invasion under high-resolution conditions; they can also reveal cell growth patterns, complex spatial structures and arrangements, and the mechanisms of action between different cells. This information is fused and integrated through machine learning for model construction of different tasks. The integration of traditional and reliable pathological diagnosis with efficient and accurate AI provides an important direction for the research of various diseases including HCC.

[0006] Ridge regression has been widely developed in terms of parameter estimation and model prediction. At present, AI is mainly used for intelligent analysis of exfoliated cytology, such as cervical liquid-based cytology, thyroid puncture cytology, pleural and peritoneal effusion cytology, and respiratory exfoliated cytology. Solid tumors similar to HCC are difficult to identify directly because of the complex and varied cell structure morphology and interstitial distribution. Ridge regression is widely used in many fields of machine learning and pattern recognition because of its mathematical tractability and efficiency. Different types of collagen play different roles in the occurrence and progression of HCC EMT, and the detailed roles of various types of collagen need to be further deepened. Multi-organ and multi-omics models related to HCC need to be further developed. There is still a lot of room for improvement in image analysis technology in the field of AI; the preparation and digital scanning of complete pathological large sections to obtain more information will require more work. Summary of the invention

[0007] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a method for AI to identify the EMT and interstitial collagen characteristics of hepatocellular carcinoma cancer cells. This paper uses the IHC method and the reticular fiber staining method to mark HCC cancer cells and local collagen fibers of cancer respectively, and combines AI image information technology to establish a ROI prediction model, refine the HCC pathological morphological characteristics related to the occurrence of EMT, and use the Ridge regression model to screen features and model. Studies have shown that EMT is closely related to the clinical pathological factors and prognosis of HCC; the ROI prediction model can accurately identify the occurrence of HCC with EMT; the pathological morphological characteristics of AI image analysis provide an objective basis for the accurate and quantitative diagnosis and prognosis of HCC with EMT in pathology.

[0008] Preferably, the present invention provides a method for AI to identify EMT and interstitial collagen characteristics of hepatocellular carcinoma cancer cells, comprising: obtaining HCC morphological characteristics based on pre-acquired immunohistochemical staining images;

[0009] The morphological characteristics of HCC include microvessel density and microvessel area;

[0010] Among them, based on the pre-acquired immunohistochemical staining images, the morphological characteristics of HCC were obtained, including:

[0011] Analyze the HCC stromal microvessels marked by CD34 in immunohistochemical staining images under HE-DAB status;

[0012] Adjust the blood vessel threshold in the immunohistochemical staining image so that the blood vessels appear red and the blood vessel background appears white;

[0013] Fill vascular holes and eliminate edge effects;

[0014] The microvessel density and microvessel area in the immunohistochemical staining images were calculated.

[0015] Preferably, the morphological characteristics of HCC are obtained based on pre-acquired immunohistochemical staining images, including:

[0016] Perform optical density correction on the immunohistochemical staining pictures, randomly select several areas in the immunohistochemical staining pictures to measure the optical density, and take the average value of the optical density measurement as the optical density value of the immunohistochemical staining pictures;

[0017] Describe the measurement area in the immunohistochemical staining image and segment to obtain the collagen staining area;

[0018] Calculate the total area of ​​the measurement area, the total area of ​​the collagen staining area within the measurement area, and the cumulative optical density to obtain the average optical density value, collagen value, and collagen thickness;

[0019] Select the collagen arrangement direction, take points along the monomer collagen walking trajectory to generate a curve and perform curve fitting to obtain the real-time curvature of each point on the curve and the average collagen curvature;

[0020] HCC morphological features also include mean optical density value, collagen value, collagen thickness, real-time curvature and mean collagen curvature.

[0021] Preferably, the morphological characteristics of HCC are obtained based on pre-acquired immunohistochemical staining images, including:

[0022] Marking α-SMA-positive CAFs and α-SMA-positive background areas in the immunohistochemical staining images, separating CAFs and background in the immunohistochemical staining images, and generating binary images;

[0023] The threshold was adjusted and the CAF proportion was counted to obtain the α-SMA-positive CAF intensity;

[0024] The morphological features of HCC also include the intensity of α-SMA-positive CAFs.

[0025] Preferably, the morphological features of HCC are input into the pre-trained ROI prediction model, and the result of HCC with EMT is output; wherein the result of HCC with EMT includes HCC with EMT and HCC without EMT.

[0026] Preferably, the pre-trained ROI prediction model includes:

[0027] Through a preset number of cross-validations, the previously acquired historical immunohistochemical staining images are divided into a training set and a test set in a preset ratio according to the random number table method;

[0028] The HCC morphological features were extracted from the historical immunohistochemical staining images previously acquired in the training set, and the best morphological features were screened among the HCC morphological features using the Ridge regression model;

[0029] The constructed ROI prediction model is trained using the optimal morphological features to obtain a trained ROI prediction model;

[0030] The trained ROI prediction model is tested using the test set, and a qualified ROI prediction model is output.

[0031] Prioritize the use of the test set to test the trained ROI prediction model, and output a qualified ROI prediction model, including:

[0032] The receiver operating characteristic curve was used to evaluate the prediction efficiency of the model. If the significance level of the ROI prediction model was P < 0.05 and the confidence interval was 95%, the ROI prediction model was considered qualified.

[0033] Preferably, the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the methods described in the first aspect when executing the program.

[0034] Preferably, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described in the first aspect.

[0035] The beneficial effects achieved by the present invention are:

[0036] This paper uses IHC method and reticular fiber staining method to mark HCC cancer cells and local collagen fibers respectively, combines AI image information technology to establish ROI prediction model, refines HCC pathological morphological characteristics related to EMT, and uses Ridge regression model to screen features and build models. EMT is closely related to clinical pathological factors and prognosis of HCC; the ROI prediction model provided by the present invention can accurately identify the occurrence of HCC with EMT; the pathological morphological characteristics of AI image analysis provide an objective basis for accurate and quantitative diagnosis and prognosis of HCC with EMT in pathology. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0038] Figure 1 is a flow chart of some embodiments of the present application;

[0039] Figure 2 is a flow chart for obtaining microvascular characteristics in some embodiments of the present application;

[0040] Figure 3 is a flow chart for obtaining average optical density value, collagen value and collagen thickness in some embodiments of the present application;

[0041] Figure 4 is a flow chart of obtaining collagen curvature in some embodiments of the present application;

[0042] Figure 5 is a flow chart for obtaining the strength of α-SMA-positive CAFs in some embodiments of the present application. DETAILED DESCRIPTION

[0043] Embodiment 1

[0044] HE in HE-DAB state: (Hematoxylin and eosin staining) This is a staining method commonly used in histology to observe cell and tissue structure.

[0045] DAB in HE-DAB status: 3,3'-Diaminobenzidine, used as a chromogen in immunohistochemistry, produces a brown color when it reacts with peroxidase, helping to visualize antigen-antibody reactions.

[0046] HE-DAB staining refers to a staining method that combines hematoxylin-eosin staining with DAB immunohistochemical staining.

[0047] CD34 is an endothelial cell marker that can mark the proliferating capillaries in HCC.

[0048] See also Figure 1 , the present application discloses a method for AI to identify EMT and interstitial collagen characteristics of hepatocellular carcinoma cancer cells, comprising:

[0049] ① Immunohistochemical staining: Each paraffin section was 4 μm thick, and routine HE staining, Envision two-step immunohistochemical method, Gordon-Sweets staining and sealing were performed. Main antibodies: E-Cadherin, Vimentin, CD34 and α-SMA; TGF-β was purchased from Xiamen Tongling Company; MMP7 was purchased from Abcam Company; Ki-67 was purchased from Maixin Biotechnology Co., Ltd.; Reticular fiber staining was purchased from Anbiping Company. Results were interpreted according to the (5th edition) Digestive System Tumor Classification Guidelines: Two highly qualified diagnostic doctors in the Department of Pathology used a double-blind method to interpret the results. If the results were the same, they would proceed to the next step; if the results were inconsistent, they would discuss and reach a consensus.

[0050] ② Image acquisition and software application: Strictly ensure the same microscope conditions, use OLYMPUSBX53 microscope under the same lighting environment to complete all immunohistochemical staining pictures, and obtain immunohistochemical staining pictures. The format of immunohistochemical staining pictures is 4912×3684 pixels, with a magnification of ×100 to ×400. The format and magnification of the same batch of immunohistochemical staining pictures are exactly the same. ImageJ software is used for immunohistochemical staining picture analysis, and Ridge (ridge regression) has been widely developed in parameter estimation and model prediction.

[0051] ③ Microvascular feature learning: Analyze the HCC interstitial microvessels marked by CD34 in the immunohistochemical staining images under the HE-DAB state; adjust the vascular threshold so that the blood vessels appear red and the background appears white; fill the vascular holes and eliminate the edge effect; automatically count to obtain the microvascular density and microvascular area in the immunohistochemical staining images.

[0052] ④Collagen feature learning: 1. First, perform optical density correction and set measurement parameters. Randomly select several areas in the immunohistochemical staining picture to measure optical density (OD), and take the average value of the optical density measurement as the optical density value of the immunohistochemical staining picture; then draw the measurement area, manually segment the collagen staining area, and calculate the total area of ​​the measurement area, the total area of ​​the collagen staining area in the measurement area, and the cumulative optical density (IOD) to obtain the average optical density value (Average optical density value, AOD), collagen value and collagen thickness. 2. Select the collagen arrangement direction, manually select points along the monomer collagen walking trajectory to generate a curve, and then perform curve fitting to finally obtain the real-time curvature of each point of the curve and the average collagen curvature.

[0053] ⑤ α-SMA-positive CAF intensity: Use the multi-point tool to mark the α-SMA-positive CAF and background areas in the immunohistochemical staining image, separate the CAF and background in the immunohistochemical staining image, generate a binary image, adjust the threshold and count the CAF ratio.

[0054] ⑥Model construction and analysis: ImageJ software was used to extract HCC morphological features from immunohistochemical staining images. HCC morphological features include microvascular features, collagen features, and α-SMA-positive CAF intensity. The Ridge regression model was used to screen the best morphological features in HCC morphological features, and the corresponding kernel functions and parameters were normalized and configured for ROI prediction model development. Through 10 cross-validations, the historical immunohistochemical staining and the corresponding HCC with EMT results were divided into training set and test set in a ratio of 7:3 according to the random number table method. Ridge regression was used to establish the ROI prediction model of HCC EMT based on the selected morphological features, and the receiver operating characteristic (ROC) curve was used to evaluate the prediction efficiency of the model.

[0055] Technical principle: Data analysis was performed using SPSS 26.0 and Prism 8.0 software. 2 The significance of the differences between the groups was tested and analyzed; Spearman correlation analysis was used for correlation comparison; ImageJ software was used for image analysis; Ridge regression model was completed based on the "sklearn" machine learning package in "Python". Kaplan-Meier survival curves were drawn to analyze the relationship with cancer prognosis. ROC curves were drawn to compare the efficiency of training and test sets in predicting HCC EMT, with a significance level of P < 0.05 and a confidence interval (CI) of 95% level.

[0056] In an embodiment of the present application, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above methods when executing the program.

[0057] In an embodiment of the present application, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0058] The biggest problem with AI analysis of cancer blood vessels and stroma is solid tumors like HCC, because the cell structure morphology and stroma distribution are complex and changeable, making direct identification difficult.

[0059] Based on IHC and special staining technology, the present invention extracted a series of HCC pathological morphological features with the help of AI image analysis function, screened out the six best morphological features related to HCC EMT using the Ridge regression model, and established a ROI prediction model based on the best morphological features.

[0060] The study showed that the ROI prediction model had a high predictive value for HCC EMT in the test set after training with a training set, with an area under the curve (AUC) of 0.868 in the training set and 0.819 in the test set. Morphology reflects the characteristics and quality of HCC cancer cells and cancer stroma.

[0061] The present invention shows from the perspective of genetics and morphology that when HCC progresses, under the stimulation of migration-promoting factors such as TGF-β, cancer cells have a low degree of differentiation, active proliferation, high expression of anaplastic embryonic protein (AFP), and the initiation of EMT. The loss of E-cadherin reduces cell-cell adhesion, leading to changes in the cytoskeleton and cell polarity, and the function of the cells also changes accordingly. The cells are endowed with the characteristics of mesenchymal cells, and the expression of Vimentin and MMP7 makes them more invasive, produces high MVD, transforms into CAFs, or stimulates the increase of cancer-associated fibroblasts (CAFs) originally present in the cancer stroma, and obtains more interstitial collagen deposition and composition changes.

[0062] Among the 125 cases of HCC in the present invention, there are 107 cases with liver cirrhosis, accounting for 86% of the total cases, which proves that the occurrence of HCC is closely related to liver cirrhosis. The morphological characteristics of cancer stroma collagen detected by AI image analysis are also one of the innovative points of the present invention. The present invention shows that the average collagen thickness, the average curvature of monomeric collagen fibers, the collagen density and the collagen value of the EMT group are significantly higher than those of the non-EMT group, proving that when HCC undergoes EMT, TGF-β can promote the differentiation of mesenchymal progenitor cells into fibroblasts and myofibroblast lineages, and promote the background of cancer sclerosis. The present invention found that α-SMA positive CAFs detected by IHC often form dense connective tissue proliferation stroma around cancer cells. It was observed by AI image technology that α-SMA positive CAFs in cancer stroma were significantly more in the EMT group than in the non-EMT group, indicating that CAFs in EMT proliferate actively, can promote the synthesis and deposition of collagen components (type I collagen), reshape the extracellular matrix ECM, and promote EMT. The survival curve study of this group of cases showed that EMT was the main risk factor for OS and PFS in HCC patients (P < 0.05). The larger the EMT-related cancer stroma collagen curvature, collagen density, MVD and tumor diameter obtained by AI imaging technology, the worse the prognosis. This result proves that EMT is the main risk factor for HCC, and accurate differentiation of EMT through IHC combined with AI can ultimately better predict the prognosis of HCC.

[0063] Embodiment 2

[0064] Manual interpretation of the intensity of classic EMT markers in immunohistochemically stained HCC paraffin sections and their correlation with clinical prognostic factors

[0065] 1. IHC test results showed that among 125 HCC cases, 45 cases had high expression of TGF-β, low expression of E-cadherin and high expression of Vimentin, with significant cell spindle formation and EMT; 80 cases had low expression of TGF-β, high expression of E-cadherin and low expression of Vimentin, with round or quasi-round cells and no EMT.

[0066] 2. Analyze the correlation between HCC with EMT and clinical pathological factors of cancer and the expression of HCC Ki-67 and MMP7.

[0067]

[0068]

[0069] 3. The prepared "lung cancer single cell isolation solution" was stored at -20°C.

[0070] Embodiment 3

[0071] Machine learning was used to analyze the morphological parameters of cancer cells and cancer stroma microvessels in HE-stained HCC paraffin sections and their correlation with EMT.

[0072] ① When the image was acquired and analyzed at 200 times magnification, HCC interstitial microvessels were unevenly distributed in the tumor stroma. The tumor blood vessels were dilated, cystic, tortuous, and interconnected in a chaotic manner.

[0073] ② AI image analysis of cancer interstitial MVA and MVD showed that the mean MVA in this group was 6788.8 pixels (1205.3-20743.1 pixels), and the median vascular area ratio was 7.85% (1.3%-21.2%), with ≥7.85% being considered a high vascular area ratio and <7.85% being considered a low vascular area ratio;

[0074] ③The proportion of cancer stroma MVA in non-EMT cancer was significantly lower than that in EMT group (χ2=6.516, P=0.011);

[0075] ④The mean MVD in HCC was 259 (range, 32-802), which was used as the boundary for high MVD and low MVD;

[0076] ⑤ The 66.7% high MVD in the EMT group was significantly higher than the 32.2% in the non-EMT group (χ2=4.265, P=0.039).

[0077] Embodiment 4

[0078] Machine learning analysis of cancer stroma collagen morphological parameters in collagen-stained HCC paraffin sections, fibroblast marker α-SMA intensity in immunohistochemically stained HCC paraffin sections, and their correlation with EMT

[0079] ① AI measured the collagen thickness of each mesh staining: the results showed that the average collagen thickness of the EMT group was 310.9 μm, which was significantly higher than the average collagen thickness of the non-EMT group (239.1 μm).

[0080] ②AI measured the collagen density and collagen value of each case: The results showed that the median collagen density of HCC in this group was 0.11 (0.02-0.21). The median was used as the threshold to divide the group into a high-density group (≥0.11) and a low-density group (<0.11). High-density collagen accounted for 66.7% in the EMT group and 46.3% in the non-EMT group, while low-density collagen accounted for 53.8% in the non-EMT group and 33.3% in the EMT group. The difference between the two groups was statistically significant (χ 2 =4.827, P=0.028). When calculating the collagen density, the total pixel value of collagen in each case can be obtained, and the collagen value of each case can be calculated. The average collagen value of the EMT group was 14.5%, which was significantly higher than the average collagen value of the non-EMT group, 9.6%.

[0081] ③ Manually track the arrangement trend of HCC cancer stroma monomer collagen fibers and measure the curvature. The average curvature of the monomer collagen fibers in the two groups was 0.47±0.17μm in the EMT group. -1 and non-EMT group 0.35±0.15μm -1 , the collagen fibers in the EMT group had a higher curvature.

[0082] ④ Using a multi-point tool to mark CAFs and background areas, AI image analysis was performed to analyze the proportion of CAFs in each cancer stroma in the collagen background area. In the EMT group, CAFs accounted for 73.3%, which was significantly higher than that in the non-EMT group (43.7%), and the difference between the two groups was significant (P = 0.003).

[0083] ⑤Comparison of the results of collagen morphological characteristics and CAF characteristics between the two groups showed that the collagen thickness, collagen value, collagen density, collagen curvature and CAF in the EMT group were significantly higher than those in the non-EMT group (all P<0.05).

[0084] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0085] Those skilled in the art will readily appreciate other embodiments of the invention after considering the specification and practicing the invention invented herein. This application is intended to cover any variations, uses or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art that are not invented by the present invention. The specification and examples are intended to be exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0086] The above specific implementation methods further illustrate the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above are only specific implementation methods of the present application and are not used to limit the protection scope of the present application. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solutions of the present application should be included in the protection scope of the present application.

Claims

1. A method for AI to identify EMT and interstitial collagen characteristics of hepatocellular carcinoma cells, characterized in that: include: Based on the immunohistochemical staining images obtained in advance, the morphological characteristics of HCC were obtained; The morphological characteristics of HCC include microvessel density and microvessel area; Among them, based on the pre-acquired immunohistochemical staining images, the morphological characteristics of HCC were obtained, including: Analyze the HCC stromal microvessels marked by CD34 in immunohistochemical staining images under HE-DAB status; Adjust the blood vessel threshold in the immunohistochemical staining image so that the blood vessels appear red and the blood vessel background appears white; Fill vascular holes and eliminate edge effects; The microvessel density and microvessel area in the immunohistochemical staining images were calculated.

2. The method for identifying EMT and interstitial collagen characteristics of hepatocellular carcinoma cells by AI according to claim 1, characterized in that: in, Based on the pre-acquired immunohistochemical staining images, the morphological characteristics of HCC were obtained, including: Perform optical density correction on the immunohistochemical staining pictures, randomly select several areas in the immunohistochemical staining pictures to measure the optical density, and take the average value of the optical density measurement as the optical density value of the immunohistochemical staining pictures; Describe the measurement area in the immunohistochemical staining image and segment to obtain the collagen staining area; Calculate the total area of ​​the measurement area, the total area of ​​the collagen staining area within the measurement area, and the cumulative optical density to obtain the average optical density value, collagen value, and collagen thickness; Select the collagen arrangement direction, take points along the monomer collagen walking trajectory to generate a curve and perform curve fitting to obtain the real-time curvature of each point on the curve and the average collagen curvature; HCC morphological features also include mean optical density value, collagen value, collagen thickness, real-time curvature and mean collagen curvature.

3. The method for AI identification of EMT and interstitial collagen characteristics of hepatocellular carcinoma cells according to claim 1, characterized in that: in, Based on the pre-acquired immunohistochemical staining images, the morphological characteristics of HCC were obtained, including: Marking α-SMA-positive CAFs and α-SMA-positive background areas in the immunohistochemical staining images, separating CAFs and background in the immunohistochemical staining images, and generating binary images; The threshold was adjusted and the CAF proportion was counted to obtain the α-SMA-positive CAF intensity; The morphological features of HCC also include the intensity of α-SMA-positive CAFs.

4. The method for AI identification of EMT and interstitial collagen characteristics of hepatocellular carcinoma cells according to claim 1, characterized in that: The morphological features of HCC are input into the pre-trained ROI prediction model, and the results of HCC with EMT are output; wherein the results of HCC with EMT include HCC with EMT and HCC without EMT.

5. The method for AI identification of EMT and interstitial collagen characteristics of hepatocellular carcinoma cells according to claim 4, characterized in that: Pre-trained ROI prediction models include: Through a preset number of cross-validations, the previously acquired historical immunohistochemical staining images are divided into a training set and a test set in a preset ratio according to the random number table method; The HCC morphological features were extracted from the historical immunohistochemical staining images previously acquired in the training set, and the best morphological features were screened among the HCC morphological features using the Ridge regression model; The constructed ROI prediction model is trained using the optimal morphological features to obtain a trained ROI prediction model; The trained ROI prediction model is tested using the test set, and a qualified ROI prediction model is output.

6. The method for AI identification of EMT and interstitial collagen characteristics of hepatocellular carcinoma cells according to claim 5, characterized in that: The trained ROI prediction model is tested using the test set, and the qualified ROI prediction model is output, including: The receiver operating characteristic curve was used to evaluate the prediction efficiency of the model. If the significance level of the ROI prediction model was P < 0.05 and the confidence interval was 95%, the ROI prediction model was considered qualified.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 6 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images