Construction method and evaluation system of tumor internal heterogeneity evaluation model and its application

Abstract

The present invention discloses a construction method and an evaluation system for evaluating a model for evaluating internal heterogeneity of a tumor and its application. The construction method includes reading the average nuclear radius, median staining depth and average staining intensity of each tumor cell in a tumor pathology slice image, represented by Xi; sorting Xi to generate a median Xm, and substituting it into the formula MAD=median(|Xi‑Xm|) to obtain MAD; the number of tumor pathology slice images is N, then N MADs are obtained correspondingly, N≥50, and the tumor pathology slice images are from the same tumor cohort; sorting the N MADs to obtain the ranks corresponding to the tumor pathology slice images, dividing the ranks by N to obtain S, so the value range of S is 0‑1, and the tumor internal heterogeneity evaluation model is obtained. The present invention avoids the need for a large amount of sequencing for genomic heterogeneity, reduces costs, and has a high correlation coefficient with genomic heterogeneity.

Description

Technical Field

[0001] The present invention belongs to the field of oncology and diagnosis, and in particular relates to a construction method and an evaluation system for evaluating the internal heterogeneity of a tumor and applications thereof. Background Art

[0002] The discovery of tumor heterogeneity is contrary to conventional theories about tumor development. In the 1950s, it was generally believed that the final clinically detectable tumor was the product of a specific subpopulation of sequentially selected tumor "stems". In this theory, the majority of tumors are dominated by a single "stem" that is superior to other "stems" due to natural selection. The TNM staging system was developed during the same period as the concept of tumor "stems", in which most tumors will be composed of a single "type" of tumor. See Peter Nowel, Science (1976) 194: 23-28.

[0003] In the context of cancer, the term "tumor heterogeneity" describes the observation that different tumor cells may exhibit different morphological, phenotypic and genetic profiles, including cell morphology, gene expression, gene mutation, metabolism, motility, proliferation and metastatic potential. Tumor heterogeneity can occur between tumors (inter-tumor heterogeneity) and within tumors (intra-tumor heterogeneity). Tumor heterogeneity has been observed in a variety of cancers, including but not limited to lung cancer, leukemia, breast cancer, kidney cancer, prostate cancer, colon cancer, brain cancer, esophageal cancer, head and neck cancer, bladder cancer, gynecological cancer, liposarcoma and multiple myeloma. Two models have been proposed to explain heterogeneity in tumor cells: the cancer stem cell model and the clonal evolution model. The cancer stem cell model provides that the heterogeneity observed between tumor cells is due to differences in cancer stem cells from which the tumor cells originate. The clonal evolution model provides that the tumor comes from a single mutant cell but accumulates additional mutations (this produces additional subpopulations, each with the ability to further divide and mutate), which explains the diversity observed in cancer cells from the same tumor. These models are not considered mutually exclusive, and therefore both may result in different amounts of heterogeneity between different tumor types. Tumor heterogeneity includes global variance (overall variance) and the spatial structure of this variation (population spatial stratification), and therefore both elements of variation should be considered in sample design. Tumor heterogeneity may arise from genetic heterogeneity (e.g., by exogenous factors, genomic instability, treatment, etc.), other heterogeneity (e.g., epigenetics), and / or the tumor microenvironment (e.g., regional differences in tumors, such as oxygen utilization or immune surveillance, exert different selective pressures on tumor cells).

[0004] In terms of composition, solid tumors are actually heterogeneous rather than homogeneous. Some solid tumors are reported to be composed of multiple genetically distinct, spatially separated cancer cell populations. See Gerlinger et al., NEJM (2012) 366: 883-92; and Yachida et al. Nature (2010) 467 (7319): 1114-1117.

[0005] Here, we take breast cancer as an example to illustrate the importance of internal tumor heterogeneity to tumor diagnosis and treatment, but it cannot be understood that internal tumor heterogeneity only has a significant impact on breast cancer diagnosis and treatment. As we all know, breast cancer is the malignant tumor with the highest incidence rate among women in China and even in the world. Among them, triple-negative breast cancer (estrogen receptor, progesterone receptor and HER2 expression are all negative) accounts for about 15% of the total number of breast cancers. It is characterized by strong tumor invasiveness, easy distant metastasis and treatment resistance in the early stage of the disease, lack of effective treatment targets, and poor prognosis for patients. It is a major problem in clinical treatment.

[0006] The main reason for the invasion, metastasis and treatment failure of triple-negative breast cancer is the existence of internal tumor heterogeneity, that is, the differences between different cell subpopulations within the same tumor. Patients with high internal tumor heterogeneity often have poor prognosis and are resistant to most chemotherapy, targeted therapy and immunotherapy. Therefore, internal tumor heterogeneity has become an important marker for clinical prediction of efficacy and patient prognosis. At present, the evaluation of internal tumor heterogeneity is mainly based on sequencing methods, including whole genome sequencing or whole exome sequencing after multi-site sampling, single cell genome and transcriptome sequencing, etc. Based on the above sequencing technology, the degree of internal tumor heterogeneity can be accurately calculated, however, it is difficult to promote in daily clinical work. On the one hand, due to the high platform requirements of these sequencing technologies, they cannot be carried out in most clinical centers; on the other hand, sequencing is expensive and most patients cannot afford it. Therefore, it is urgent to develop simple and economical technologies that can be used clinically to accurately evaluate internal tumor heterogeneity and provide clues for precise treatment of patients. Summary of the invention

[0007] In order to evaluate the internal heterogeneity of solid tumors, on the one hand, the present invention discloses a method for constructing an internal heterogeneity evaluation model for tumors, including reading the average nuclear radius, median staining depth and average staining intensity of each tumor cell in a tumor pathology section image, represented by Xi; sorting Xi to generate a median Xm, and substituting it into the formula MAD=median(|Xi-Xm|) to obtain MAD (Median Absolute Deviation); the number of the tumor pathology section images is N, then N MADs are obtained correspondingly, N≥50, and the tumor pathology section images are from the same tumor cohort; sorting the N MADs to obtain the rank corresponding to each of the tumor pathology section images, dividing the rank by N to obtain S, so the value range of S is 0-1, and the internal heterogeneity evaluation model for tumors is obtained.

[0008] For the same cell nucleus, multiple cell nucleus radius values ​​can be measured from different angles on the tumor pathology section image, and the average cell nucleus radius is taken for evaluation in the present invention.

[0009] In some embodiments, the evaluation criteria of the tumor internal heterogeneity evaluation model are: taking 0.5 as the evaluation threshold, if S>0.5, the corresponding tumor pathology section image is a high heterogeneity tumor; if S<0.5, the corresponding tumor pathology section image is a low heterogeneity tumor;

[0010] The tumor is a solid tumor; the solid tumor is selected from one of Hodgkin's lymphoma, non-Hodgkin's lymphoma, lung cancer, breast cancer, ovarian cancer, gastric cancer, colon cancer, rectal cancer, liver cancer, pancreatic cancer, head and neck malignancies, urinary system malignancies, endometrial cancer, cervical cancer, osteosarcoma, chondrosarcoma, Ewing's sarcoma, thyroid cancer, hepatoblastoma, and Wilms' tumor.

[0011] In some embodiments, N≥134 (preferably N≥260, more preferably N=260); each of the tumor pathology slice images corresponds to one and only one patient; and the breast cancer is triple-negative breast cancer.

[0012] In some embodiments, the tumor pathology section image is stained with H&E and digitally scanned by a digital pathology section scanning system to form a tumor H&E digital pathology section image, which is then used to read and obtain Xi;

[0013] Before reading, the image analysis software (such as CellProfiler 4.2.1) needs to be used to set parameters, including the following steps:

[0014] Step a1: converting color to gray, so that the tumor pathology slice image is converted from an RGB image to a grayscale image;

[0015] Step a2, identifying the main object and setting the parameters of the image mode that can be identified;

[0016] Step a3, measuring the size and shape of the object, selecting the cell nucleus and its parameters in the measurement image;

[0017] Step a4, measuring the image staining depth and identifying the parameter settings of the image staining depth.

[0018] In some embodiments, the digital scan uses a magnifying glass of the same magnification (such as a 40x objective lens) to scan the H&E-stained tumor pathological sections used to construct and / or use the tumor internal heterogeneity assessment model.

[0019] In some embodiments, the method further comprises performing a correlation analysis between the tumor internal heterogeneity assessment results of the N tumor pathological section images obtained by the tumor internal heterogeneity assessment model and the tumor internal heterogeneity results obtained by the gold standard, confirming that the correlation coefficient r≥0.3 and P<0.05 are statistically significant, proving that the tumor internal heterogeneity assessment model can be used to reflect the degree of internal heterogeneity of the tumor; the gold standard refers to the genomic heterogeneity calculated by the PyClone algorithm based on the whole exome sequencing data and the copy number chip data.

[0020] In some embodiments, P<0.001.

[0021] On the other hand, the present invention also discloses a system for evaluating the internal heterogeneity of a tumor, comprising:

[0022] A tumor internal heterogeneity assessment model establishment module is used to read the average nuclear radius, median staining depth and average staining intensity of each tumor cell in the tumor pathology section image, which is represented by Xi; it is also used to sort Xi to generate the median Xm, and substitute it into the formula MAD=median(|Xi-Xm|) to obtain MAD; the number of the tumor pathology section images is N, then N MADs are obtained correspondingly, N≥50, and the tumor pathology section images are from the same tumor cohort; it is also used to sort the N MADs to obtain the rank corresponding to each of the tumor pathology section images, and the rank is divided by N to obtain S, so the value range of S is 0-1;

[0023] A pathological heterogeneity calculation module is used to calculate the S of other tumor pathological slice images outside the tumor internal heterogeneity assessment model establishment module, specifically: reading the average nuclear radius, median staining depth and average staining intensity of each tumor cell in the other tumor pathological slice images, represented by Xi; and also used to substitute Xi into the formula MAD=median(|Xi-Xm|) to obtain the MAD of the other tumor pathological slice images; and also used to sort the MAD of the other tumor pathological slice images with the N MADs to obtain the rank corresponding to the other tumor pathological slice images, and divide the rank corresponding to the other tumor pathological slice images by N to obtain the S of the other tumor pathological slice images;

[0024] An evaluation standard module is used to record the evaluation standard and compare the S obtained by the calculation pathological heterogeneity module and / or the S of the other tumor pathological slice image with 0.5 to obtain the tumor internal heterogeneity evaluation result of the corresponding tumor pathological slice image; the evaluation standard is: taking 0.5 as the evaluation threshold, if S>0.5, the corresponding tumor pathological slice image is a high heterogeneity tumor; if S<0.5, the corresponding tumor pathological slice image is a low heterogeneity tumor.

[0025] In some embodiments, it also includes:

[0026] A digital pathology section scanning system, used for digitally scanning the H&E stained tumor pathology section image to form a tumor H&E digital pathology section image, so as to read and obtain Xi;

[0027] Image analysis module, used for parameter setting, including:

[0028] A color-to-gray submodule, used to convert the tumor pathology slice image from an RGB image to a grayscale image;

[0029] The main object recognition submodule is used to set the parameters of the image mode that can be recognized;

[0030] The object size and shape measurement submodule is used to select the cell nuclei and their parameters in the measurement image;

[0031] The submodule for measuring image staining depth is used to identify the parameter settings of image staining depth.

[0032] In some embodiments, N≥134 (preferably N≥260, more preferably N=260); each of the tumor pathology section images corresponds to only one patient; the tumor is a solid tumor; the solid tumor is selected from one of Hodgkin's lymphoma, non-Hodgkin's lymphoma, lung cancer, breast cancer (preferably, the breast cancer is triple-negative breast cancer), ovarian cancer, gastric cancer, colon cancer, rectal cancer, liver cancer, pancreatic cancer, head and neck malignancies, urinary system malignancies, endometrial cancer, cervical cancer, osteosarcoma, chondrosarcoma, Ewing's sarcoma, thyroid cancer, hepatoblastoma, and nephroblastoma.

[0033] In a third aspect, the present invention further discloses a computer-readable storage medium, wherein the storage medium stores executable instructions, wherein the executable instructions execute the method for constructing a tumor internal heterogeneity assessment model as described above; and / or

[0034] When the executable instructions are executed, the functions of the evaluation system for internal tumor heterogeneity as described above are realized.

[0035] In a fourth aspect, the present invention further discloses a terminal, comprising:

[0036] A memory for storing executable instructions;

[0037] A processor, configured to execute executable instructions stored in the memory, wherein the executable instructions execute the method for constructing a tumor internal heterogeneity assessment model as described above; and / or

[0038] When used to execute the executable instructions stored in the memory, the functions of the evaluation system for internal tumor heterogeneity as described above are realized.

[0039] At present, there is no simple technology for evaluating internal tumor heterogeneity in clinical practice, especially for triple-negative breast cancer. Based on the largest multi-omics database of triple-negative breast cancer, the present invention creates a method and system for calculating internal tumor heterogeneity based on tumor H&E stained sections, and implements and verifies it in an external cancer genome atlas cohort and 4 independent triple-negative breast cancer immunotherapy cohorts. The results of this method are highly consistent with the gold standard-the genomic heterogeneity calculated using the PyClone algorithm based on whole exome sequencing and copy number chip results. The internal tumor heterogeneity calculated based on H&E digital pathology section images of triple-negative breast cancer samples can accurately predict patient prognosis and also indicate sensitivity to immune checkpoint inhibitors. The present invention provides a reference for more accurate prognostic diagnosis of tumors, especially triple-negative breast cancer, and for specifying more targeted treatment strategies.

[0040] The present invention achieves the evaluation of internal tumor heterogeneity through pathological heterogeneity, avoids the need for large-scale sequencing of genomic heterogeneity, reduces costs, and has a high correlation coefficient with genomic heterogeneity, which can well reflect the degree of internal tumor heterogeneity of patients.

[0041] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flowchart of a method for constructing a tumor internal heterogeneity assessment model according to an embodiment of the present invention.

[0043] Figure 2 Schematic diagram of the calculated pathological H&E scans and software circled nuclear ranges of high and low pathological heterogeneity.

[0044] Figure 3 It is a curve chart showing the correlation between pathological heterogeneity and genomic heterogeneity.

[0045] Figure 4 This is a correlation calculation curve diagram of the TCGA cohort to verify pathological heterogeneity.

[0046] Figure 5 This is the prognostic survival analysis curve of patients with high and low pathological heterogeneity in the TCGA cohort.

[0047] Figure 6 This is a flow chart of Example 1 of the present invention applied to four independent clinical trials of immunotherapy for triple-negative breast cancer.

[0048] Figure 7 Three independent triple-negative breast cancer immunotherapy cohorts verified the predictive value of pathological heterogeneity for the efficacy of immunotherapy. A shows that in the NCT04613674 clinical trial, the pathological remission rate of 11 patients with high heterogeneity tumors after immunotherapy was significantly reduced; B shows that in the NCT04418154 clinical trial, the pathological remission rate of 9 patients with high heterogeneity tumors after immunotherapy was significantly reduced; C shows that in the NCT03805399 clinical trial, the objective remission rate of 9 patients with high heterogeneity tumors after immunotherapy was significantly reduced.

[0049] Figure 8 The CT04129996 clinical trial verified the predictive value of pathological heterogeneity for the efficacy of immunotherapy. A shows that the objective response rate of 11 patients with highly heterogeneous tumors decreased significantly after immunotherapy; B shows that the progression-free survival time (left) and overall survival time (right) of 11 patients with highly heterogeneous tumors were shortened after immunotherapy. Figure 7 and 8 In the table, * indicates P < 0.05; ** indicates P < 0.01. DETAILED DESCRIPTION

[0050] In order to make the technical means, creative features, objectives and effects of the invention easier to understand, the invention is further described below with reference to specific diagrams. However, the invention is not limited to the following implementation cases.

[0051] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0052] Since the present invention divides high heterogeneity tumors and low heterogeneity tumors by analyzing tumor pathological section images, in order to distinguish them from the genomic heterogeneity analysis method, the high heterogeneity tumors and low heterogeneity tumors divided by the present invention are respectively referred to as high pathological heterogeneity tumors and low pathological heterogeneity tumors, or simply referred to as high pathological heterogeneity and low pathological heterogeneity, or even more simply referred to as high heterogeneity and low heterogeneity.

[0053] The term "rank" refers to an ordinal number arranged according to numerical size.

[0054] Example 1: Constructing a tumor internal heterogeneity assessment model using 260 tumor pathology slice images

[0055] The general process is as follows Figure 1 As shown:

[0056] 1. Generate tumor H&E digital pathology slice images

[0057] Based on the triple-negative breast cancer cohort constructed by the Affiliated Cancer Hospital of Fudan University, hematoxylin and eosin (H&E) stained sections of the tumor were selected and digitally scanned (40x objective) using the Hamamatsu NanoZoomer digital pathology section scanning system to form H&E digital pathology section images of the tumor. H&E stained sections with poor image quality were not included in the subsequent analysis, and a total of 260 H&E digital pathology section images of the tumor were generated. The tumors corresponding to these H&E stained sections also contain exome sequencing, copy number chips, transcriptome sequencing, and complete clinical pathology and treatment information of the patients, which are used to calculate correlation coefficients with the gold standard - genomic heterogeneity calculated by PyClone.

[0058] 2. Read pathological parameters

[0059] Use the image analysis software CellProfiler 4.2.1 to set parameters. The key steps are: Step a1, color to gray, so that the pathological slice image is converted from RGB image to grayscale image; Step a2, identify the main object and set the parameters of the image mode that can be identified; Step a3, measure the size and shape of the object, select the cell nucleus and its parameters in the measurement image; Step a4, measure the image staining depth, and identify the parameter setting of the image staining depth. The following is a detailed introduction to the specific parameter setting process using a tumor H&E digital pathological slice image as an example:

[0060] 1) Import the image

[0061] Import a tumor H&E digital pathology slide image from the local computer, check "Show files excluded by filters" in the software interface, filter the image to "Image only" and select "Apply all filters to file list".

[0062] 2) Interpret the data

[0063] Select "No" to extract interpretation data.

[0064] 3) Name and type

[0065] Select "All images" for the given name, select "No" for 3D processing, select "Color image" for the image type, select "DNA" for the name assigned to the image, and select "Image metadata" for setting the range of staining depth.

[0066] 4) Grouping

[0067] Select "No" if you want to group the pictures.

[0068] 5) Color turns to gray

[0069] Select "DNA" for the imported image, select "Segmentation" for the transposition method, select "RGB" for the image type, select "Yes" for converting red to gray, select "OrigRed" for the exported image name, select "Yes" for converting green to gray, select "OrigGreen" for the exported image name, select "Yes" for converting blue to gray, and select "OrigBlue" for the exported image name.

[0070] 6) Threshold setting

[0071] Select "OrigBlue" for the imported image, name the exported image "ThreshBlue", select "Overall" for the threshold strategy, set the threshold method to "Ostu", select "Two Categories" for the two-category or three-category threshold, select "1.0" for the threshold smoothing coefficient, select "1.0" for the threshold correlation coefficient, select "0.0 to 1.0" for the upper and lower boundaries of the threshold, and select "No" for the log transformation before the threshold.

[0072] 7) Do not mix colors

[0073] Select the imported color image as "DNA", the exported image name as "Hematoxylin", the stain as "Hematoxylin", the exported image name as "Eosin", the stain as "Eosin", the exported image name as "OrangeG", the stain as "OrangeG".

[0074] 8) Identify the main objects

[0075] Select "Yes" for using advanced settings, select "Eosin" for importing images, name the main object "Tissue", select the diameter size of the object to be "10 to 40 pixels", select "No" for discarding objects outside the diameter range, select "No" for discarding objects touching the image boundary, select "Overall" for threshold range, set threshold method to "Ostu", select "Two Classes" for two or three class thresholds, select "1.0" for threshold smoothing coefficient, select "1.0" for threshold correlation coefficient, select "0.0 to 1.0" for upper and lower threshold boundaries, select "No" for log transformation before thresholding, select "None" for method of distinguishing clustered objects, select "Never" for filling holes on identified objects, and select "Continue" for object processing when too many identified objects are present.

[0076] 9) Mask Image

[0077] Select "Hematoxylin" for the imported image, name the exported image "MaskedRed", select "Object" for the object or image as mask, select "Tissue" as the object mask, and select "No" for the transpose mask.

[0078] Select "Eosin" for the imported image, name the exported image "MaskedBlue", select "Object" for the object or image as a mask, select "Organize" as the object mask, and select "No" for the transpose mask.

[0079] Select "OrigenG" for the imported image, name the exported image "MaskedOrange", select "Object" for the object or image as a mask, select "Organize" as the object mask, and select "No" for the transpose mask.

[0080] 10) Identify the main object

[0081] Select "Yes" for using advanced settings, select "MaskedRed" for importing images, name the original object "Nuclei", select "5 to 40 pixels" for diameter size of objects, select "Yes" for discarding objects outside the diameter range, select "No" for discarding objects touching the image borders, select "Overall" for threshold range, select "Ostu" for threshold method, select "Three Classes" for two or three class thresholds, assign pixels with medium intensity levels to "Background", select "1.0" for threshold smoothing factor, select "1.0" for threshold correlation factor, select "0.0 to 1.0" for upper and lower threshold boundaries, select "No" for log transformation before thresholding, select "Darkness of staining" for identification method of clumped objects, select "Darkness of staining" for method of drawing dividing lines between clumped objects, select "Yes" for automatic calculation of smoothing filter size, select "Yes" for automatic calculation of minimum allowed distance between local maxima, select "Yes" for speeding up by using a low-resolution image to find local maxima, select "No" for displaying accepted local maxima, select "After thresholding and subtraction" for filling holes in identified objects, and select "Continue" for processing objects if too many objects are identified.

[0082] 11) Measure the size and shape of objects

[0083] Select "Nuclei" as the object to be measured, select "Yes" for calculating Zernike features, and select "No" for calculating advanced features.

[0084] 12) Measure the staining depth of the object

[0085] Select the image measurement parameters as "OrigBlue" and "OrigRed", and select the measurement object as "Nuclei".

[0086] 13) Measure the image staining depth

[0087] Select the image measurement parameter as "OrigBlue", select "No" for measuring intensity only from the area surrounded by the object, and select "No" for custom calculation of percentiles.

[0088] 14) Measuring image quality

[0089] Select "OrigBlue", "OrigGreen" and "OrigRed" for calculating photo parameters, select "Yes" for including image scaling value, fill in 20 for the spatial scale of blur measurement, select "Yes" for calculating staining intensity parameters, select "Yes" for calculating saturation index, select "Yes" for calculating thresholds, select "No" for using all threshold methods, select "Otsu" for threshold method, select "Weighted variance" for minimizing weighted variance or entropy, and select "Two-class" for two-class or three-class threshold.

[0090] 15) Measuring Occupancy Images

[0091] To measure occupied area, select "Binary Image" and "ThreshBlue" to measure binary image.

[0092] 16) Measurement structure

[0093] Select "OrigBlue", "OrigGreen" and "OrigRed" to measure the image, measure the image and the object at the same time, select the object to be measured as "Nuclei", enter 256 grayscale values ​​to measure the image texture, and the texture scale to be measured is 3.

[0094] 17) Export result table

[0095] Select the column delimiter as "Tab", select the location of the output file, and select to export all measurement types.

[0096] 3. Calculation of pathological heterogeneity

[0097] For each tumor H&E digital pathology section image, the diversity of cell nuclei inside the tumor is first calculated. In the above-mentioned pathology parameter result file, the results of the average cell nucleus radius, median staining depth and average staining intensity of each cell nucleus on the section (i.e., pathology section image) are extracted. Then the absolute median deviation (MAD) of all cell nucleus features on a tumor pathology section image is calculated to characterize the degree of heterogeneity of the cell nuclei in the tumor pathology section image. The formula involved is MAD = median (|Xi-Xm|) (Formula 1), where Xi represents the average cell nucleus radius, median staining depth and average staining intensity of all cells on a tumor pathology section image, and Xm represents the median of the set composed of all Xi on the tumor pathology section image. Here, there is no need to classify the average cell nucleus radius, median staining depth and average staining intensity separately, but to merge them into one set, so that a tumor pathology section image corresponds to an Xm, which is substituted into Formula 1 to obtain a MAD. Thus, the diversity of the nuclei of a tumor pathology section image is reflected in the absolute median difference (MAD) of the average nucleus radius, median staining depth and average staining intensity of all identifiable nuclei on this tumor pathology section image. Then, each tumor pathology section image is sorted according to the heterogeneity value (i.e., MAD) calculated based on the nucleus morphology to obtain the corresponding rank. Finally, the final tumor internal heterogeneity score (S) of the tumor corresponding to this tumor pathology section image is calculated, that is, the rank corresponding to the tumor pathology section image is divided by the number of tumor pathology section images (N) involved in the sorting, ranging from 0 to 1. Through the dichotomy, 0.5 is set as the threshold. The final calculated tumor internal heterogeneity score of the tumor corresponding to the digital pathology section is greater than 0.5, which is considered to be a high heterogeneous tumor; and the score is less than 0.5, which is considered to be a low heterogeneous tumor. Since tumor pathology section images are used to evaluate the internal heterogeneity of tumors, they are referred to as pathological heterogeneity in this article.

[0098] Take a tumor pathology section image as an example to calculate the degree of heterogeneity: 9872 tumor cells were read in a single tumor pathology section image by CellProfiler software (for example, version 4.2.1), with average nuclear radius 1, average nuclear radius 2...average nuclear radius 9872, median staining depth 1, median staining depth 2...median staining depth 9872, average staining intensity 1, average staining intensity 2...average staining intensity 9872. First, the median (Xm) of the 29616 data (represented by Xi, i=1-29616) generated by these three parameters was calculated, and the median (MAD) of the absolute value (|Xi-Xm|) of the new data obtained by subtracting the median (Xm) from all data (Xi, i=1-29616) was the nuclear heterogeneity value of this pathology section. For example, the rank of the tumor pathology section image that reads 9872 tumor cells ranks 26th among the 260 tumor pathology section images used in this embodiment, then the S of the tumor pathology section image that reads 9872 tumor cells is 0.1, which is less than 0.5. The tumor pathology section image that reads 9872 tumor cells is a low heterogeneity tumor. Figure 2 The two figures above show two tumor H&E digital pathology slide images with S values ​​of 0.16 and 0.89 ( Figure 2 Pathological H&E scan film); Figure 2 The following two figures show the schematic diagram of the circle cell nucleus range of two tumor H&E digital pathology slide images with S of 0.16 and 0.89 during the analysis process of the image analysis software CellProfiler 4.2.1 ( Figure 2 (referred to as software core scope in Chinese).

[0099] 4. Calculate the correlation between pathological heterogeneity and genomic heterogeneity

[0100] Based on whole exome sequencing data and copy number chip data, the genomic heterogeneity calculated using the PyClone algorithm is a recognized method for evaluating intratumor heterogeneity. The calculated genomic heterogeneity and pathological heterogeneity were analyzed for correlation (or correlation calculation) in 260 triple-negative breast cancers at the Affiliated Cancer Hospital of Fudan University. It was found that the correlation coefficient between the two could reach 0.47, and the statistical test showed that it was significant (P<0.001), see Figure 3 Therefore, it can be considered that the pathological heterogeneity calculated based on the tumor pathological slice image (furthermore, the tumor digital pathological slice image, and more preferably the tumor H&E digital pathological slice image) can well reflect the degree of internal heterogeneity of the patient's tumor. This embodiment successfully constructs a tumor internal heterogeneity assessment model.

[0101] Example 2: TCGA cohort verifies the relevance of pathological heterogeneity and its prognostic value

[0102] This pathological heterogeneity calculation method has been implemented in patients with triple-negative breast cancer in The Cancer Genome Atlas (TCGA), verifying that pathological heterogeneity has a good correlation with genomic heterogeneity, and verifying the predictive value of pathological heterogeneity for survival in the TCGA cohort.

[0103] Download the H&E digital pathology slide images of triple-negative breast cancer in the TCGA cohort from https: / / tcga-data.nci.nih.gov / tcga / , and remove the slides with poor quality to obtain a total of 134 tumor H&E digital pathology slide images that can be analyzed in the CellProfiler software. Using the same method as in Example 1, read the pathological parameters of all tumor H&E digital pathology slide images, and calculate the pathological heterogeneity score S. First, the pathological heterogeneity was correlated with the gold standard-the genomic heterogeneity calculated by PyClone. It was found that the two had a good correlation (correlation coefficient r = 0.38, statistical test showed a difference, P < 0.001), see Figure 4 The triple-negative breast cancer patients in the TCGA cohort were divided into 67 patients with high heterogeneity and 67 patients with low heterogeneity based on the pathological heterogeneity score of 0.5. Subsequent survival analysis found that the prognosis of patients with high pathological heterogeneity was worse than that of patients with low pathological heterogeneity. Figure 5 .

[0104] Example 3: Four independent triple-negative breast cancer immunotherapy cohorts validate the predictive value of pathological heterogeneity for immunotherapy efficacy

[0105] This pathological heterogeneity calculation method was implemented in four independent triple-negative breast cancer immunotherapy cohorts to verify the predictive value of pathological heterogeneity for immunotherapy efficacy and patient survival.

[0106] Four independent triple-negative breast cancer immunotherapy clinical trials were used to digitally scan the H&E digital pathology slice images of the tumors before treatment, and calculate the pathological heterogeneity score S. Specifically, the ranking of the 260 MADs of the tumor internal heterogeneity assessment model constructed in Example 1 was used to calculate the rank of each H&E digital pathology slice image of the tumor in the four independent triple-negative breast cancer immunotherapy clinical trials, and the rank was divided by 260 to obtain the corresponding S.

[0107] Taking 0.5 as the cutoff value for high and low heterogeneity patients, it was found that patients with high heterogeneity tumors with pathological heterogeneity scores S>0.5 before treatment had poor efficacy after immunotherapy, mainly manifested as follows: In the clinical trials of NCT04613674 and NCT04418154, the pathological remission rate of patients with high heterogeneity tumors was low (see Figure 7 A and 7B); in the NCT03805399 and NCT04129996 clinical trials, patients with highly heterogeneous tumors had low objective response rates (see Figure 7 C and Figure 8 A); In the NCT04129996 clinical trial, it was also found that patients with highly heterogeneous tumors had shorter progression-free survival and overall survival after immunotherapy (see Figure 8 B). The above four clinical trials further confirmed that the invention of the pathological heterogeneity algorithm is convenient for clinical use and has strong immunotherapy efficacy and prognostic prediction value.

[0108] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for constructing a tumor internal heterogeneity assessment model, characterized in that: The method includes reading the average nuclear radius, median staining depth and average staining intensity of each tumor cell in the tumor pathology section image, which are represented by Xi; merging the average nuclear radius, median staining depth and average staining intensity into one set; sorting Xi to generate the median Xm, and substituting it into the formula MAD=median(|Xi-Xm|) to obtain MAD; the number of the tumor pathology section images is N, then N MADs are obtained correspondingly, N≥50, and the tumor pathology section images are from the same tumor cohort; sorting the N MADs to obtain the rank corresponding to each of the tumor pathology section images, dividing the rank by N to obtain S, so the value range of S is 0-1, and the tumor internal heterogeneity evaluation model is obtained; The tumor pathology section image is stained with H&E and digitally scanned by a digital pathology section scanning system to form a tumor H&E digital pathology section image, which is then used to read and obtain Xi; Before reading, the image analysis software needs to be used to set parameters, including the following steps: Step a1: converting color to gray, so that the tumor pathology slice image is converted from an RGB image to a grayscale image; Step a2, identifying the main object and setting the parameters of the image mode that can be identified; Step a3, measuring the size and shape of the object, selecting the cell nucleus and its parameters in the measurement image; Step a4, measuring the image staining depth and identifying the parameter settings of the image staining depth.

2. The method for constructing a tumor internal heterogeneity assessment model according to claim 1, characterized in that: The evaluation criteria of the tumor internal heterogeneity evaluation model are: taking 0.5 as the evaluation threshold, if S>0.5, the corresponding tumor pathology slice image is a high heterogeneity tumor; if S<0.5, the corresponding tumor pathology slice image is a low heterogeneity tumor; The tumor is a solid tumor; the solid tumor is selected from one of Hodgkin's lymphoma, non-Hodgkin's lymphoma, lung cancer, breast cancer, ovarian cancer, gastric cancer, colon cancer, rectal cancer, liver cancer, pancreatic cancer, head and neck malignancies, urinary system malignancies, endometrial cancer, cervical cancer, osteosarcoma, chondrosarcoma, Ewing's sarcoma, thyroid cancer, hepatoblastoma, and Wilms' tumor.

3. The method for constructing a tumor internal heterogeneity assessment model according to claim 2, characterized in that: N≥134; each of the tumor pathology slice images corresponds to one and only one patient; the breast cancer is triple-negative breast cancer.

4. The method for constructing a tumor internal heterogeneity assessment model according to claim 1, characterized in that: It also includes performing a correlation analysis between the tumor internal heterogeneity assessment results of the N tumor pathological section images obtained by the tumor internal heterogeneity assessment model and the tumor internal heterogeneity results obtained by the gold standard, confirming that the correlation coefficient r≥0.3 and P<0.05 statistical test are significant, proving that the tumor internal heterogeneity assessment model can be used to reflect the degree of internal heterogeneity of the tumor; the gold standard refers to the genomic heterogeneity calculated by the PyClone algorithm based on whole exome sequencing data and copy number chip data.

5. A system for evaluating intra-tumor heterogeneity, characterized in that: include: A tumor internal heterogeneity assessment model establishment module is used to read the average nuclear radius, median staining depth and average staining intensity of each tumor cell in the tumor pathology section image, which is represented by Xi; it is also used to merge the average nuclear radius, median staining depth and average staining intensity into one set, sort Xi to generate the median Xm, and substitute it into the formula MAD=median(|Xi-Xm|) to obtain MAD; the number of tumor pathology section images is N, then N MADs are obtained correspondingly, N≥50, and the tumor pathology section images are from the same tumor cohort; it is also used to sort the N MADs to obtain the corresponding ranks of the tumor pathology section images, and the ranks are divided by N to obtain S, so the value range of S is 0-1; A pathological heterogeneity calculation module is used to calculate the S of other tumor pathological slice images outside the tumor internal heterogeneity assessment model establishment module, specifically: reading the average nuclear radius, median staining depth and average staining intensity of each tumor cell in the other tumor pathological slice images, represented by Xi; and also used to substitute Xi into the formula MAD=median(|Xi-Xm|) to obtain the MAD of the other tumor pathological slice images; and also used to sort the MAD of the other tumor pathological slice images with the N MADs to obtain the rank corresponding to the other tumor pathological slice images, and divide the rank corresponding to the other tumor pathological slice images by N to obtain the S of the other tumor pathological slice images; An evaluation standard module, used to record the evaluation standard and compare the S obtained by the calculation pathological heterogeneity module and / or the S of the other tumor pathological slice image with 0.5 to obtain the tumor internal heterogeneity evaluation result of the corresponding tumor pathological slice image; the evaluation standard is: taking 0.5 as the evaluation threshold, if S>0.5, the corresponding tumor pathological slice image is a high heterogeneity tumor; if S<0.5, the corresponding tumor pathological slice image is a low heterogeneity tumor; The evaluation system for intra-tumor heterogeneity also includes: A digital pathology section scanning system, used for digitally scanning the H&E stained tumor pathology section image to form a tumor H&E digital pathology section image, so as to read and obtain Xi; Image analysis module, used for parameter setting, including: A color-to-gray submodule, used to convert the tumor pathology slice image from an RGB image to a grayscale image; The main object recognition submodule is used to set the parameters of the image mode that can be recognized; The object size and shape measurement submodule is used to select the cell nuclei and their parameters in the measurement image; The submodule for measuring image staining depth is used to identify the parameter settings of image staining depth.

6. The system for evaluating intra-tumor heterogeneity according to claim 5, characterized in that: N≥134; each of the tumor pathology slice images corresponds to only one patient; the tumor is a solid tumor; the solid tumor is selected from one of Hodgkin's lymphoma, non-Hodgkin's lymphoma, lung cancer, breast cancer, ovarian cancer, gastric cancer, colon cancer, rectal cancer, liver cancer, pancreatic cancer, head and neck malignant tumors, urinary system malignant tumors, endometrial cancer, cervical cancer, osteosarcoma, chondrosarcoma, Ewing's sarcoma, thyroid cancer, hepatoblastoma, and Wilms' tumor.

7. A computer-readable storage medium, characterized in that: The storage medium stores executable instructions, and the executable instructions execute the method for constructing a tumor internal heterogeneity assessment model according to any one of claims 1 to 4; and / or The executable instructions, when executed, implement the functions of the system for evaluating internal tumor heterogeneity according to claim 5 or 6.

8. A terminal, characterized in that: The terminal comprises: A memory for storing executable instructions; A processor, configured to execute executable instructions stored in the memory, wherein the executable instructions execute the method for constructing a tumor internal heterogeneity assessment model according to any one of claims 1 to 4; and / or When used to execute the executable instructions stored in the memory, the functions of the tumor internal heterogeneity evaluation system according to claim 5 or 6 are realized.

Images