A prognostic prediction method and system for triple-negative breast cancer after neoadjuvant chemotherapy based on tertiary lymphatic structure

Abstract

This invention discloses a method and system for predicting the prognosis of neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure. The method involves acquiring paraffin-embedded breast cancer tissue data corresponding to neoadjuvant-adjuvant triple-negative breast cancer and processing the corresponding tissue sections. Based on the three-tiered lymphatic structure and combined with the paraffin-embedded breast cancer tissue data, tiered data corresponding to TLS density and TLS maturity are generated, along with corresponding estimated data. A prediction model corresponding to neoadjuvant-adjuvant triple-negative breast cancer is constructed, and combined with the estimated data, real-time prediction data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer is generated. This method can accurately assess the application value of TLS maturity and spatial location characteristics in the prognostic evaluation of triple-negative breast cancer, and simultaneously construct a more accurate prognostic prediction model, aiming to provide a strong basis for the formulation of personalized treatment plans, thereby promoting the improvement of survival rate and quality of life for triple-negative breast cancer patients.

Description

Technical Field

[0001] This invention belongs to the field of data prediction and processing technology, specifically relating to a method and system for predicting the prognosis of neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure. Background Technology

[0002] In various types of solid malignancies, the differences in the location of tertiary lymphoid structures (TLS) are closely related to changes in their maturity and have distinctly different prognostic values. However, in breast cancer, especially triple-negative breast cancer, little is known about the prognostic role of TLS with different locational differences. With neoadjuvant therapy becoming the mainstream treatment strategy for triple-negative breast cancer patients, TLS assessment, which relies on surgical specimens, must consider the impact of neoadjuvant therapy. This may lead to discrepancies between the prognostic value of TLS in neoadjuvant-treated triple-negative breast cancer and conclusions drawn from other tumors. Previous studies on the prognostic value of TLS in triple-negative breast cancer have largely ignored the differences in the spatial location of TLS or failed to consider the crucial factor of neoadjuvant therapy.

[0003] Furthermore, there are currently no large-scale clinical studies on the prognostic value of TLS in neoadjuvant-adjuvant triple-negative breast cancer with different heterogeneities. Moreover, it is impossible to accurately assess the application value of TLS maturity and spatial localization characteristics in the prognostic assessment of triple-negative breast cancer, and it is impossible to construct an accurate prognostic prediction model to provide a strong basis for the formulation of personalized treatment plans, thereby promoting the improvement of survival rate and quality of life of triple-negative breast cancer patients.

[0004] Therefore, given the shortcomings of previous studies on the prognostic value of TLS in triple-negative breast cancer, such as neglecting the differences in TLS with different spatial locations, failing to consider the key factor of neoadjuvant therapy, being unable to accurately assess the maturity of TLS and the application value of spatial location characteristics in the prognostic assessment of triple-negative breast cancer, and lacking large-sample clinical studies on the prognostic value of heterogeneous TLS in neoadjuvant-adjuvant triple-negative breast cancer, there is an urgent need to design and develop a prognostic prediction method and system for neoadjuvant-adjuvant triple-negative breast cancer based on the three-level lymphatic structure. Summary of the Invention

[0005] To overcome the shortcomings and difficulties of the existing technologies, the present invention aims to provide a prognostic prediction method, system and platform for neoadjuvant-adjuvant triple-negative breast cancer based on the three-level lymphoid structure. This method can accurately assess the maturity and spatial localization characteristics of TLS in the prognostic evaluation of triple-negative breast cancer, and at the same time construct a more accurate prognostic prediction model, so as to provide a strong basis for the formulation of personalized treatment plans, thereby promoting the improvement of survival rate and quality of life of triple-negative breast cancer patients.

[0006] The first objective of this invention is to provide a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphoid structure; the second objective of this invention is to provide a prognostic prediction system for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphoid structure; and the third objective of this invention is to provide a prognostic prediction platform for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphoid structure.

[0007] The first objective of this invention is achieved as follows: the method comprises the following steps:

[0008] Obtain paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer, and process the corresponding paraffin block of breast cancer tissue into sections based on the paraffin block data of breast cancer tissue.

[0009] Based on the three-level lymphoid structure and combined with the paraffin block data of the breast cancer tissue, first-level data corresponding to TLS density and second-level data corresponding to TLS maturity are generated, and estimated data corresponding to the level data are generated; wherein, the first-level data is a semi-quantitative abundance level; and the second-level data is a maturity level.

[0010] A predictive model corresponding to neoadjuvant triple-negative breast cancer is constructed, and predictive data corresponding to the prognosis of neoadjuvant triple-negative breast cancer is generated in real time by combining the estimated data.

[0011] Furthermore, the step of acquiring paraffin-embedded breast cancer tissue data corresponding to neoadjuvant-adjuvant-adjuvant-adjuvant-negative breast cancer, and sectioning and processing the corresponding paraffin-embedded breast cancer tissue based on the paraffin-embedded breast cancer tissue data, further includes:

[0012] H&E staining and Ki67 immunohistochemical staining were performed on paraffin blocks of breast cancer tissue, and corresponding KFB format files were generated.

[0013] Based on the KFB format file, corresponding pathological image data in SVS format is generated;

[0014] The SVS format WSI file was processed by region delineation, area measurement, and ROI region cell count calculation.

[0015] Furthermore, the construction of a predictive model corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and the generation of predictive data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time, in conjunction with the estimated data, further includes:

[0016] To obtain sample data corresponding to neoadjuvant triple-negative breast cancer, and to determine and estimate whether there are significant prognostic differences between different groups of samples;

[0017] Each covariate was individually subjected to Cox proportional hazards regression analysis, and covariate data that were significantly associated with survival time were generated.

[0018] A multi-factor Cox model is constructed, and continuous variables are converted into restricted cubic splines and added to the multi-factor Cox model to generate corresponding node count data.

[0019] Furthermore, the construction of a predictive model corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and the generation of predictive data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time, in conjunction with the estimated data, further includes:

[0020] Based on the prediction model corresponding to neoadjuvant triple-negative breast cancer, data on the relationship between tertiary lymphoid structure and disease-free survival and overall survival were generated.

[0021] Acquire clinicopathological feature data corresponding to neoadjuvant triple-negative breast cancer, and generate correlation data between the clinicopathological feature data and the tertiary lymphoid structure feature data;

[0022] Linear correlation data is created between the three-level lymphatic structure feature data and prognostic data, and corresponding correction data is generated based on the linear correlation data; wherein, the linear correlation data includes linear correlation data and non-linear correlation data.

[0023] Furthermore, the construction of a predictive model corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and the generation of predictive data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time, in conjunction with the estimated data, further includes:

[0024] Based on the three-level lymphoid structure feature data, a DFS prediction model for triple-negative breast cancer was constructed, and the performance of each prediction model was compared to generate model performance evaluation index data corresponding to the performance.

[0025] The second objective of this invention is achieved as follows: the system is applied to a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure, the system comprising:

[0026] The data acquisition and processing unit is used to acquire paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer, and to slice and process the corresponding paraffin block of breast cancer tissue according to the paraffin block data of breast cancer tissue.

[0027] The first data generation unit is used to generate first-level data corresponding to TLS density and second-level data corresponding to TLS maturity based on the three-level lymphoid structure and the paraffin block data of the breast cancer tissue, and to generate estimated data corresponding to the level data; wherein, the first-level data is a semi-quantitative abundance level; and the second-level data is a maturity level.

[0028] The second data generation unit is used to construct a prediction model corresponding to neoadjuvant triple-negative breast cancer, and combine the estimated data to generate prediction data corresponding to the prognosis of neoadjuvant triple-negative breast cancer in real time.

[0029] Furthermore, the data acquisition and processing unit further includes:

[0030] The first processing module is used to perform H&E staining and Ki67 immunohistochemical staining on paraffin blocks of breast cancer tissue, and generate corresponding KFB format files.

[0031] The second processing module is used to convert the KFB format file into corresponding pathological image data in SVS format.

[0032] The third processing module is used to perform region delineation processing, area measurement processing, and ROI region cell number calculation processing on the SVS format WSI file respectively.

[0033] Furthermore, the second data generation unit further includes:

[0034] The first judgment module is used to acquire sample data corresponding to neoadjuvant triple-negative breast cancer, and to determine and estimate whether there are significant prognostic differences between different groups of samples.

[0035] The first data generation module is used to perform Cox proportional hazards regression analysis on each covariate individually and generate covariate data that are significantly associated with survival time.

[0036] The second data generation module is used to construct a multi-factor Cox model, and at the same time convert continuous variables into restricted cubic splines and add them into the multi-factor Cox model to generate corresponding node number data.

[0037] Furthermore, the second data generation unit further includes:

[0038] The third data generation module is used to generate data on the relationship between the three-level lymphoid structure and disease-free survival and overall survival based on the prediction model corresponding to neoadjuvant triple-negative breast cancer.

[0039] The fourth data generation module is used to acquire clinicopathological feature data corresponding to neoadjuvant triple-negative breast cancer, and generate correlation data between the clinicopathological feature data and the tertiary lymphoid structure feature data.

[0040] The fifth data generation module is used to create linear correlation data between the three-level lymphoid structure feature data and prognostic data, and to generate corresponding correction data based on the linear correlation data; wherein, the linear correlation data includes linear correlation data and non-linear correlation data;

[0041] The second data generation unit further includes:

[0042] The sixth data generation module is used to construct a DFS prediction model for triple-negative breast cancer based on the three-level lymphoid structure feature data, compare the performance of each prediction model, and generate model performance evaluation index data corresponding to the performance.

[0043] The third objective of this invention is achieved as follows: it includes a processor, a memory, and a control program for a post-adjuvant triple-negative breast cancer prognostic prediction platform based on a tertiary lymphoid structure; wherein the processor executes the control program, which is stored in the memory, and the control program implements the post-adjuvant triple-negative breast cancer prognostic prediction method based on a tertiary lymphoid structure.

[0044] The present invention obtains paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer, and processes the corresponding paraffin block of breast cancer tissue into sections based on the paraffin block data.

[0045] Based on the three-level lymphoid structure and combined with the paraffin block data of the breast cancer tissue, first-level data corresponding to TLS density and second-level data corresponding to TLS maturity are generated, and estimated data corresponding to the level data are generated; wherein, the first-level data is a semi-quantitative abundance level; and the second-level data is a maturity level.

[0046] A predictive model corresponding to neoadjuvant triple-negative breast cancer is constructed, and predictive data corresponding to the prognosis of neoadjuvant triple-negative breast cancer is generated in real time by combining the estimated data; and a system and platform corresponding to the method are also provided, which can accurately evaluate the maturity and spatial positioning characteristics of TLS in the prognostic assessment of triple-negative breast cancer, and at the same time construct a more accurate prognostic predictive model, so as to provide a strong basis for the formulation of personalized treatment plans, thereby promoting the improvement of survival rate and quality of life of triple-negative breast cancer patients.

[0047] In other words, this solution is a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure. By constructing a prediction model corresponding to neoadjuvant-adjuvant triple-negative breast cancer and combining it with the estimated data, it generates high-precision prediction data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure according to the present invention.

[0050] Figure 2 This is a schematic diagram illustrating a pathological section evaluation example of an embodiment of the present invention, a method for predicting the prognosis of neoadjuvant-adjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure.

[0051] Figure 3 This is an example of the present invention, a method for predicting the prognosis of neoadjuvant-adjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure, showing a comparison of different maturation stages of TLS and SLO (tumor-infiltrated axillary lymph nodes);

[0052] Figure 4 This is a schematic diagram of the research process for an embodiment of the present invention, which is a method for predicting the prognosis of neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure.

[0053] Figure 5 This is an example of a prognostic prediction method for neoadjuvant- ... triple-negative breast cancer based on a three-level lymphatic structure, illustrated by a schematic diagram of the TLS region and maturity level of mulberries in the SYSU cohort of the present invention.

[0054] Figure 6 This is a schematic diagram illustrating the disease-free survival and overall survival estimation (Kaplan-Meier) of the SYSU cohort and TCGA cohort, as an embodiment of the prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure according to the present invention.

[0055] Figure 7 This is a thermogram illustrating the correlation between clinicopathological features and tertiary lymphoid structure features in an embodiment of a prognostic prediction method for neoadjuvant-adjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure according to the present invention.

[0056] Figure 8This is a schematic diagram of a forest from a univariate Cox regression analysis, representing an embodiment of a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphoid structure according to the present invention.

[0057] Figure 9 This is a schematic diagram of a forest in multivariate Cox regression analysis of DFS and OS in Model 1 of an embodiment of the present invention, which is a method for predicting the prognosis of neoadjuvant-adjuvant triple-negative breast cancer based on a three-level lymphoid structure.

[0058] Figure 10 This is a schematic diagram of a forest in multivariate Cox regression analysis of DFS and OS in Model 2 of an embodiment of the present invention, which is a method for predicting the prognosis of neoadjuvant-adjuvant triple-negative breast cancer based on a three-level lymphoid structure.

[0059] Figure 11 This is a schematic diagram of a forest in multivariate Cox regression analysis of DFS and OS in Model 3 of an embodiment of a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-level lymphoid structure according to the present invention.

[0060] Figure 12 This is a schematic diagram of a forest in multivariate Cox regression analysis of DFS and OS in Model 4 of an embodiment of a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-level lymphoid structure according to the present invention.

[0061] Figure 13 This is a schematic diagram illustrating the nonlinear relationship between adjusted TLS features and DFS hazard ratio (HR) in an embodiment of a prognostic prediction method for neoadjuvant-adjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphoid structure according to the present invention.

[0062] Figure 14 The ROC curve and DCA decision curve of Model 1-Model are shown in the embodiment of the present invention, which is a method for predicting the prognosis of neoadjuvant-adjuvant triple-negative breast cancer based on a three-level lymphoid structure.

[0063] Figure 15 This is an example of a prognostic prediction method for neoadjuvant- ...verse triple-negative breast cancer based on a three-level lymphoid structure according to the present invention, showing the three-year and five-year calibration curves of Model 1-Model 5 models.

[0064] Figure 16 This is a schematic diagram of a DFS prognostic model constructed based on TLS features, as an embodiment of a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure according to the present invention.

[0065] Figure 17 This is an example of the prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphoid structure according to the present invention. The diagram shows the DFS and OS stratification (Kaplan-Meier) of the SYSU cohort and TCGA cohort based on TLS immune risk scores.

[0066] Figure 18 This is a schematic diagram of spatial typing based on TLS, representing an embodiment of a prognostic prediction method for neoadjuvant- ... triple-negative breast cancer based on a three-tiered lymphoid structure according to the present invention.

[0067] Figure 19 This is an example of a prognostic prediction method for neoadjuvant-adjuvant-adjuvant triple-negative breast cancer based on a three-level lymphoid structure, illustrating the DFS and OS stratification (Kaplan-Meier) of the SYSU and TCGA cohorts based on TLS spatial typing in accordance with the present invention.

[0068] Figure 20 This is a schematic diagram of the sensitivity analysis of the TCGA cohort, an embodiment of the present invention, which is a method for predicting the prognosis of neoadjuvant-adjuvant triple-negative breast cancer based on a three-level lymphoid structure.

[0069] Figure 21 This is a schematic diagram of the prognostic prediction system architecture for neoadjuvant- ... triple-negative breast cancer based on a three-tiered lymphatic structure, according to the present invention.

[0070] Figure 22 This is a schematic diagram of the prognostic prediction platform architecture for neoadjuvant- ... triple-negative breast cancer based on a three-tiered lymphatic structure, according to the present invention.

[0071] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0072] To facilitate a clearer understanding of the objectives, technical solutions, and advantages of this invention, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Those skilled in the art can easily understand other advantages and effects of this invention from the content disclosed in this specification.

[0073] This invention can also be implemented or applied through other different specific examples, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of this invention.

[0074] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0075] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Secondly, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0076] Preferably, the prognostic prediction method for neoadjuvant-adjuvant-adjuvant-mediated triple-negative breast cancer based on a three-tiered lymphatic structure is applied in one or more terminals or servers. The terminal is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0077] The terminal can be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal can interact with the customer via a keyboard, mouse, remote control, touchpad, or voice control device.

[0078] This invention provides a method, system, and platform for predicting the prognosis of neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure.

[0079] like Figure 1 The diagram shown is a flowchart of a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure, provided by an embodiment of the present invention.

[0080] In this embodiment, the prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on the three-level lymphatic structure can be applied to a terminal or fixed terminal with display function. The terminal is not limited to personal computers, smartphones, tablets, desktop computers or all-in-one computers with cameras, etc.

[0081] The prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on the three-level lymphatic structure can also be applied to a hardware environment consisting of a terminal and a server connected to the terminal via a network. The network includes, but is not limited to, a wide area network (WAN), a metropolitan area network (MAN), or a local area network (LAN). The prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on the three-level lymphatic structure in this embodiment of the invention can be executed by a server, by a terminal, or by both a server and a terminal.

[0082] For example, for a terminal requiring prognostic prediction of neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure, the prognostic prediction function based on tertiary lymphoid structure provided by the method of this invention can be directly integrated into the terminal, or a client for implementing the method of this invention can be installed. Alternatively, the method provided by this invention can also run on servers or other devices in the form of a Software Development Kit (SDK), providing an interface for the prognostic prediction function of neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure in the form of an SDK. Terminals or other devices can then implement the prognostic prediction function of neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure through the provided interface. The invention will be further described below with reference to the accompanying drawings.

[0083] like Figure 1 As shown, this invention provides a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure. The method includes the following steps:

[0084] S01. Obtain paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer, and process the corresponding paraffin block of breast cancer tissue into sections based on the paraffin block data of breast cancer tissue.

[0085] S02. Based on the three-level lymphoid structure and combined with the paraffin block data of the breast cancer tissue, first-level data corresponding to TLS density and second-level data corresponding to TLS maturity are generated, and estimated data corresponding to the level data are generated; wherein, the first-level data is a semi-quantitative abundance level; the second-level data is a maturity level.

[0086] S03. Construct a prediction model corresponding to neoadjuvant triple-negative breast cancer, and combine the estimated data to generate prediction data corresponding to the prognosis of neoadjuvant triple-negative breast cancer in real time.

[0087] The step of obtaining paraffin-embedded breast cancer tissue data corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and sectioning and processing the corresponding paraffin-embedded breast cancer tissue based on the paraffin-embedded breast cancer tissue data, further includes:

[0088] S011. Perform H&E staining and Ki67 immunohistochemical staining on paraffin blocks of breast cancer tissue, respectively, and generate corresponding KFB format files.

[0089] S012. Based on the KFB format file, convert and generate corresponding pathological image data in SVS format;

[0090] S013. Perform region delineation processing, area measurement processing, and ROI region cell number calculation processing on the SVS format WSI file respectively.

[0091] The construction of a predictive model corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and the generation of predictive data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time by combining the estimated data, further includes:

[0092] S031. Obtain sample data corresponding to neoadjuvant triple-negative breast cancer, determine and estimate whether there are significant prognostic differences between different groups of samples;

[0093] S032. Perform Cox proportional hazards regression analysis on each covariate separately and generate covariate data that are significantly associated with survival time.

[0094] S033. Construct a multi-factor Cox model, and convert continuous variables into restricted cubic splines, and add them to the multi-factor Cox model to generate corresponding node count data.

[0095] The construction of a predictive model corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and the generation of predictive data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time by combining the estimated data, further includes:

[0096] S034. Based on the prediction model corresponding to triple-negative breast cancer after neoadjuvant therapy, generate data on the relationship between tertiary lymphoid structure and disease-free survival and overall survival.

[0097] S035. Obtain clinicopathological feature data corresponding to neoadjuvant triple-negative breast cancer, and generate correlation data between the clinicopathological feature data and the tertiary lymphoid structure feature data.

[0098] S036. Create linear correlation data between the three-level lymphatic structure feature data and prognostic data, and generate corresponding correction data based on the linear correlation data; wherein, the linear correlation data includes linear correlation data and non-linear correlation data.

[0099] The construction of a predictive model corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and the generation of predictive data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time by combining the estimated data, further includes:

[0100] S037. Based on the three-level lymphoid structure feature data, construct a DFS prediction model for triple-negative breast cancer, compare the performance of each prediction model, and generate model performance evaluation index data corresponding to the performance.

[0101] Specifically, in this embodiment of the invention, mature TLS and TLS in the proximal (T) region indicate a better prognosis for patients with triple-negative breast cancer, while TLS in the distal (P) region is associated with a worse prognosis for patients with triple-negative breast cancer.

[0102] This retrospective analysis reviewed an independent long-term follow-up cohort (SYSU cohort) from Sun Yat-sen Memorial Hospital, Sun Yat-sen University. The inclusion criteria for the study population are as follows:

[0103] (1) Female, aged ≥20 years and <75 years at the time of first diagnosis.

[0104] (2) Breast cancer patients who underwent surgery after receiving standard neoadjuvant therapy at Sun Yat-sen Memorial Hospital of Sun Yat-sen University between 2011 and 2021.

[0105] (3) The pathological molecular classification is triple negative breast cancer, that is, both ER and PR are negative (IHC ER and PR positive percentage <1%), and Her2 is negative or low expression (IHC Her2- / + or IHC Her2++ but FISH-).

[0106] (4) The pathological type is invasive carcinoma.

[0107] (5) Sign the informed consent form for the reuse of discarded specimens.

[0108] The exclusion criteria are as follows:

[0109] (1) The patient’s electronic medical record is incomplete, the neoadjuvant therapy course has not been completed, or follow-up data is difficult to obtain.

[0110] (2) The pathological sections of the surgically removed parts of the patient are missing or difficult to obtain.

[0111] (3) The patient had already experienced recurrence or distant metastasis at the time of diagnosis.

[0112] (4) The pathological diagnosis was inflammatory breast cancer.

[0113] (5) History of autoimmune disease or any active autoimmune disease.

[0114] The study cohort ultimately included 235 patients with early- to mid-stage triple-negative breast cancer who underwent surgery at the Breast Cancer Center of Sun Yat-sen Memorial Hospital, Sun Yat-sen University, between August 2011 and September 2021.

[0115] The primary outcome measure of this study was disease-free survival (DFS), defined as the time from the completion of surgical treatment to cancer recurrence, metastasis, or all-cause mortality. The primary outcome events were disease progression and death, i.e., the occurrence of any one of cancer recurrence, metastasis, or all-cause mortality. The assessment methods primarily involved clinical follow-up, including imaging examinations (such as breast ultrasound, mammography, CT, MRI, etc.) and blood test results performed at our hospital, outpatient records, and inpatient records. For patients lacking routine postoperative follow-up records, telephone follow-ups were conducted with the patients or their contact persons.

[0116] The secondary outcome measure was overall survival (OS), defined as the time from enrollment to death from any cause. The same clinical follow-up and data recording methods as for disease-free survival (DFS) were used. The secondary outcome event was all-cause mortality.

[0117] Collect all patients' preoperative clinical baseline data, including age, sex, height, weight at initial diagnosis, tumor size on preoperative ultrasound (to determine clinical T stage), preoperative axillary lymph node examination and ultrasound results (clinical N stage), clinical TNM stage according to the American Joint Committee on Cancer (AJCC) standard, and operation time.

[0118] Postoperative pathological baseline data were collected for all patients, including pathology number, number of positive lymph nodes during surgery, Ki67 level, and endovascular invasion (LVI).

[0119] The latest follow-up data was updated in January 2023. If the outcome event has occurred, the final follow-up time is recorded as the date of the outcome event; if the outcome event has not occurred, the final follow-up time is recorded as the date of the last follow-up. The median follow-up time was 26.9 months.

[0120] TCGA Cohort: TCGA (The Cancer Genome Atlas) is a large-scale cancer genomics project jointly launched by the National Cancer Institute (NCI) and the National Human Genome Research Institute (NHGRI) of the National Institutes of Health (NIH). TCGA-BRCA refers to a subset of the TCGA project specifically targeting breast cancer, including clinical information, follow-up information, transcriptome analysis, epigenetics, and pathological slide data for most participants. Regarding pathological slides, TCGA provides two types of data: frozen sections and formalin-fixed paraffin-embedded (FFEP) sections. Files marked with "TS#" or "BS#" are frozen sections, while those marked with "DX#" are formalin-fixed paraffin-embedded sections. Frozen sections are smaller in area, and their TLS (transferable tissue tract markers) are difficult to identify; therefore, this invention only includes formalin-fixed paraffin-embedded sections marked with "DX#".

[0121] Based on the clinical baseline data downloaded from the TCGA official website (https: / / portal.gdc.com) and the definition of triple-negative breast cancer, patients in the cohort who were negative for ER, PR, and Her2 were considered triple-negative breast cancer patients. The inclusion criteria were as follows: (1) ER negative, i.e., “breast carcinoma progesterone receptor status” was “Negative”. (2) PR negative, i.e., “breast carcinoma estrogen receptor status” was “Negative”. (3) Her2 negative, i.e., “lab procedure her2 neuimmunohistochemistry receptor status” or “lab procedure her2 neu in situ hybrid outcome type” was “Negative”. (4) The pathological type was invasive ductal carcinoma, i.e., “histological type” was “Infiltrating Ductal Carcinoma”.

[0122] The exclusion criteria are as follows: (1) Distant metastasis was present at the time of diagnosis, i.e., TNM stage M1. (2) Paraffin pathology data was missing or the pathology section area was too small. (3) The paraffin pathology specimen lacked the peritumoral area, making it impossible to assess the peritumoral area.

[0123] The study cohort ultimately included 147 patients with triple-negative breast cancer. RNA-seq data (level 3), corresponding clinical baseline information, and SVS format files with complete paraffin-embedded pathological section images from these 147 patients were downloaded from the TCGA official website (https: / / portal.gdc.com).

[0124] The primary endpoint was disease-free survival (DFS). The secondary endpoint was overall survival (OS). The definitions of DFS and OS were the same as in section 2.1.1. Survival follow-up data for these 147 triple-negative breast cancer patients were downloaded from the officially recognized UCSC Xena website (https: / / xena.ucsc.edu / ). The median follow-up time was 26.3 months. The assessment of the primary and secondary endpoints was the same as in the SYSU cohort.

[0125] Experimental Materials: All breast cancer tissue specimens used in this study were obtained from patients in the SYSU cohort who met the inclusion and exclusion criteria. They were collected by the pathology department after surgical resection, and preserved as paraffin blocks after fixation, dehydration, clearing, and paraffin impregnation. All samples were obtained from the pathology department of Sun Yat-sen Memorial Hospital to ensure consistency in sample processing.

[0126] Submit a list of patients including their pathology numbers to the pathology department and request H&E stained sections and Ki67 immunohistochemical stained sections from patients who meet the inclusion and exclusion criteria.

[0127] For patients with the following conditions, additional tissue blocks should be requested:

[0128] (1) H&E staining and Ki67 immunohistochemical staining sections showed defects, contamination, and fading.

[0129] (2) The area of ​​the peritumoral region in H&E staining and Ki67 immunohistochemical staining sections was less than 30%, making it impossible to assess the peritumoral region.

[0130] The tumor boundary area sections of the paraffin-embedded tissue block were stained to ensure that the assessment of the TLS of the sections could cover the tumor area and the peritumoral area.

[0131] This study was approved by the hospital's ethics committee (approval number: [SYSKY-2024-058-01]), and all patients signed informed consent forms for the reuse of discarded specimens.

[0132] The experimental reagents are shown in the table below:

[0133] Table 1: Experimental Reagents

[0134] Reagent Name Manufacturers Eosin hematoxylin staining solution (G1005-500ML) Wuhan Sewell Company Hematoxylin Blue Reversion Solution (G1040-500ML) Wuhan Sewell Company Hematoxylin differentiation solution (G1039-500ML) Wuhan Sewell Company Ki67 antibody (monoclonal, clone ID: D2H10) CST (Cell Signaling Technology) DAB colorimetric reagent Wuhan Sewell Company PBS phosphate buffer powder Beijing Solarbio Company xylene Sigma-Aldrich 100× Sodium Citrate Buffer Shanghai Maclin Company Anhydrous ethanol Guangzhou Reagent Factory 3% hydrogen peroxide Guangzhou Reagent Factory 5% bovine serum albumin Gibco, USA 4% paraformaldehyde tissue fixative Wuhan Sewell Company

[0135] Experimental methods: tissue processing

[0136] All additional tissue specimens requested were paraffin-embedded blocks of postoperative breast cancer tissue from patients meeting inclusion and exclusion criteria, obtained from the pathology department. After paraffin fixation, the tissues were serially sectioned to a thickness of approximately 4 μm.

[0137] H&E staining: (1) Preheat paraffin sections in a 58°C oven for 60 minutes, then dewax them. (2) Clear and dewax them with xylene, followed by gradient dehydration with 100%, 95%, and 70% ethanol. (3) Add hematoxylin staining solution for 5 minutes, then wash quickly with water, and use aluminum acetate solution for identification staining for 1 minute. (4) Add eosin staining solution for 2 minutes, then dehydrate and clear with 95% and 100% ethanol and xylene. (5) Mount the sections with paraffin to complete the slide preparation process.

[0138] Ki67 Immunohistochemical Staining: (1) The dewaxing and hydration steps are the same as those for H&E staining. (2) Add citrate buffer at pH 6.0 for antigen retrieval, heat to 121°C in an autoclave, and maintain at this temperature for 2 minutes. (3) Cool the sample to ambient temperature and rinse 3 times with PBS buffer, each rinse lasting about 5 minutes. (4) To prevent interference from endogenous peroxidase, incubate the sample in a solution containing 3% hydrogen peroxide for 20 minutes. (5) Block non-specific binding sites with 5% bovine serum albumin for 30 minutes. (6) Add primary antibody Ki67 (1:150 dilution) and incubate overnight at 4°C. (7) Incubate the secondary antibody with the sample at ambient temperature for 1 hour. (8) DAB staining lasts for 1-2 minutes, and the reaction should be closely monitored to prevent over-staining. (9) After light counterstaining with eosin, dehydrate and clear the sample, and mount with paraffin.

[0139] Microscopic image acquisition: All H&E stained specimens and Ki67 immunohistochemically stained specimens were whole-slide imaging (WSI) using a digital scanner (model KF-PRO-040-HI, Jiangfeng Biotechnology) and stored as KFB format files. These were then converted to SVS format pathological image data using conversion software. The publicly available downloadable QuPath software (version 0.3.2) was used to delineate regions of interest (ROIs), measure their area, and calculate the number of cells within each ROI in the SVS format WSI files.

[0140] The format conversion software is open source (https: / / github.com / wilmerwang / SLFCD / releases).

[0141] Interpretation and Evaluation of Pathological Sections: Overall Evaluation Principles: SVS format digital pathological sections of H&E-stained breast cancer tissue or adjacent normal breast tissue from the SYSU and TCGA cohorts were independently evaluated by the authors and another breast surgeon with pathology training. When disagreements arose, a pathologist was involved in the evaluation process to reach a final assessment. The interpretation and evaluation of these pathological sections followed a clearly defined methodology. Throughout the process, the other breast surgeon and pathologist were completely isolated from any clinical information or follow-up results to ensure unbiased evaluation. Key evaluation components included tertiary lymphoid structures (TLS), tumor-infiltrating lymphocytes (TIL), and vascular invasive tumor thrombi (LVI). Ki67-stained sections from the SYSU cohort were used to assist in the evaluation of H&E-stained sections. Due to the lack of Ki67 staining data in the TCGA cohort, the interpretation of the H&E-stained portion in this cohort relied on extensive evaluation experience gained from the analysis of H&E-stained sections from the SYSU cohort.

[0142] TLS assessment method: TLS, or tertiary lymphoid structures, can be broadly defined as organized lymphocyte aggregates

[16] , which can be observed and analyzed on H&E stained sections. However, in a specific context, in order to minimize the potential confusion between the presence of small TLS and TIL, this protocol specifies that structures formed by aggregates of 100 or more organized lymphocytes are TLS

[47] .

[0143] Therefore, based on the research objective, the samples were divided into two different groups based on the presence or absence of TLS (e.g., Figure 2 ):(in, Figure 2 A shows a comparison between the TLS(-) and TLS(+) groups. The yellow solid line delineates the TLS in the TLS(+) group as the ROI. B shows examples of near-tumor (T) and far-tumor (P) regions in TLS distribution. The area within the blue solid line is the T region, and the area outside the area is the P region. Invasive cancer areas, areas with residual degenerative cancer cells, areas where cancer cells have died or completely disappeared and been replaced by granulation tissue or fibrosis, and areas within 0.5 mm of their edges are all considered near-tumor (T) regions. Far-tumor (P) regions are areas on the slide that do not meet the above criteria. C shows an example of intravascular invasion (LVI). The red arrow points to the intravascular tumor thrombus. D shows an example of TIL grouping based on TIL guidelines: low-invasive group (0-10%), intermediate-invasive group (11%-40%), and high-invasive group (41%-90%).

[0144] (1) The TLS(+) group is characterized by the presence of at least one aggregate containing at least 100 lymphocytes;

[0145] (2) The TLS(-) group includes cases with no TLS at all or with only small clusters of fewer than 100 cells.

[0146] The Qupath software (version 0.3.2) is used to calculate cell count, label regions of interest (ROIs) in TLS(+) groups, measure area, and save ROI labeling data as geojson format files.

[0147] To study TLS heterogeneity in greater depth, a comprehensive analysis of various factors was conducted, including the spatial distribution, abundance, and maturity stage of TLS in the TLS(+) group.

[0148] To determine the spatial distribution of TLS, this protocol referenced previously published criteria for TLS studies in different tumor types: the distribution area of ​​TLS was subdivided into two subregions: the proximal T region and the distal P region. The T region consisted of the marginal area less than 0.5 mm from the tumor invasive margin plus the invasive carcinoma area, while the P region consisted of the stromal area greater than or equal to 0.5 mm from the tumor invasive margin. It is noteworthy that, unlike other tumor types, most triple-negative breast cancer patients receive neoadjuvant chemotherapy before surgery, followed by pathological evaluation of TLS using surgically removed tissue samples. According to the hypothesis of this protocol, after cancerous tissue necrosis and replacement by granulation tissue or fibrosis, TLS located near the tumor area before neoadjuvant therapy may not be directly adjacent to the periphery of the invasive cancer. Nevertheless, these TLS must still be classified as part of the T region. Therefore, it is necessary to adjust the criteria commonly used in previous studies when differentiating the localization subregions of TLS.

[0149] Therefore, in this study, referring to the pathological evaluation criteria after neoadjuvant therapy for breast cancer

[61] for determining the tumor bed, a new distribution area standard for TLS was established: invasive carcinoma areas, areas with residual regressive cancer cells, areas where cancer cells have died or completely disappeared and have been replaced by granulation tissue or fibrosis, and areas within 0.5 mm of their edges should all be considered as proximal (T) areas. Distal (P) areas are areas on the slide that do not meet the above criteria.

[0150] For the abundance assessment of TLS, three methods were used for evaluation: rating level, area ratio per (‰), and count. Referring to the standards set by published studies

[47] and the research practice of this scheme, the quantity of TLS is divided into four levels:

[0151] (1) None (Level 0) - No TLS observed or only small aggregates of less than 100 cells; (2) Low density (Level 1) - 1 to 3 TLS observed; (3) Medium density (Level 2) - 4 or more TLS observed, but not meeting the high density standard; (4) High density (Level 3) - Some TLS are connected in sheets.

[0152] Semi-quantitative abundance levels (T_level and P_level) were obtained by assessing TLS density in the T and P regions, respectively. In addition to the semi-quantitative quantitative grading, a quantitative method was also used to assess TLS abundance. The TLS area (S-TLS) on each H&E-stained slide of the patient was measured, including the TLS area in the T and P regions (TS-TLS & PS-TLS) and the total WSI area of ​​the tissue section (S-WSI). For cases with multiple sections, the measured areas were averaged. T_per(‰) = TS-TLS / S-WSI*1000 and P_per(‰) = PS-TLS / S-WSI*1000 were used as one of the indicators for assessing TLS abundance. The number of ROIs marked in different regions was recorded as T_count and P_count.

[0153] In assessing the maturity of TLS, the currently widely accepted assessment criteria [30-32,47] classify TLS based on lymphoid follicles (Fol) and germinal centers (GC): Early TLS (E-TLS) is a dense aggregate of mixed immature B cells and T cells, lacking lymphoid follicles and germinal centers (Fol-&GC-); Primary follicle-like TLS (PFL-TLS) is an aggregate of lymphocytes with lymphoid follicles but lacking germinal centers (Fol+&GC-); Secondary follicle-like TLS (SFL-TLS) is characterized by having both lymphoid follicles and germinal centers, and is the most mature TLS (Fol+&GC+). Lymphoid follicles are mainly composed of rapidly proliferating B lymphocytes and FDCs, etc.

[62] Previous studies [30-32,47] have typically used CD21 and CD23 to label FDCs to help identify lymphoid follicles and thus assess the maturity of TLS. Ki67 is a marker of cell proliferation and can be used to label highly proliferating B cells in TLS

[21] . Previous studies have also demonstrated the feasibility of Ki67 labeling of lymphoid follicles

[63] . Ki67 is also one of the most commonly used immunohistochemical staining methods for breast cancer; therefore, Ki67 immunohistochemical staining was chosen to be analogous to SLO, i.e., axillary lymph nodes, to help identify lymphoid follicles and germinal centers, thereby assessing the maturity of TLS. Figure 3 ).in, Figure 3A. Three maturity stages of TLS: E-TLS (Early TLS): No lymphoid follicles, no Ki67 staining; PFL-TLS (Primaryfollicle-like TLS): Lymphoid follicles present, but no germinal centers, Ki67 staining present but not aggregated; SFL-TLS (Secondary follicle-like TLS): Germinal centers present. Ki67 aggregates in the germinal centers. B. Three maturity stages of SLO. The H&E staining and Ki67 staining in SLO are similar to those in TLS, proving the reliability of the Ki67-based method for evaluating TLS maturity.

[0154] Therefore, similar to lymph nodes, the absence or minimal presence of Ki67-positive TLS indicates almost no B-cell proliferation, classifying it as E-TLS; scattered Ki67 positivity indicates that proliferating B cells are not aggregated and are distributed in primary lymphoid follicles, classifying it as PFL-TLS; aggregated Ki67 positivity indicates the presence of germinal centers, classifying it as SFL-TLS. Care should be taken to differentiate between Ki67 positivity in tumor cells and Ki67 positivity in B cells, primarily through comparison with the nuclei of tumor cells in the tumor area.

[0155] The maturity of TLS was assessed in the T and P regions respectively, and the maturity of TLS was divided into four levels: (1) cases in which no TLS was observed were classified as level 0; (2) cases in which only E-TLS was observed in all slices were classified as level 1; (3) cases in which PFL-TLS was observed in some slices but no SFL-TLS was found in any slice were classified as level 2; (4) cases in which SFL-TLS was observed were classified as level 3.

[0156] LVI Assessment Method: Lymphovascular invasion (LVI) refers to the infiltration of cancer cells into blood vessels or lymphatic vessels during pathological examination. Traditionally, LVI is one of the most important pathological features of malignant tumors and is associated with poor prognosis in various cancers, including breast cancer

[64] . Therefore, LVI is also included in the pathological variables for assessment. In the peritumoral region of the H&E stained area, look for lymphatic vessels and capillary lumens composed of endothelial cells. Assess and record the presence of tumor emboli composed of cancer cells within the lumen.

[0157] TIL assessment method: TILs in SVS format digital pathological sections were assessed according to the standards developed in the 2017 International Working Group on Immunomodulatory Biomarkers for Breast Cancer

[38] . Specific precautions were as follows: (1) Grossly scan under low magnification to determine the area to be assessed for TILs, including the central tumor and the invasive margin. The invasive margin was defined as a 1 mm wide area centered on the boundary between the cancer nest and normal tissue; (2) The mean TIL level of the tumor area was assessed, rather than being limited to hotspot areas; (3) TILs were reported as a percentage of a continuous variable during assessment. According to the guidelines, TILs were divided into three groups: low invasive group (0-10%), intermediate invasive group (11%-40%), and high invasive group (41%-90%).

[0158] Ki67 assessment method: The Ki67 expression level in the SYSU cohort was judged using the standardized "typewriter" visual assessment method

[65] , and Ki67≤30% was classified as low expression group, and Ki67>30% was classified as high expression group. However, the TCGA cohort lacked Ki67 immunohistochemical results, so this method used RNAseq of the MIB1 gene (which regulates Ki67 protein expression) instead, and divided the TCGA cohort into low expression group and high expression group with the median expression level as the cutoff point.

[0159] Statistical analysis: All statistics were analyzed using R 4.3.2 software.

[0160] Baseline statistics: Normally distributed continuous data are expressed as Mean (SD), and non-normally distributed data are expressed as Median [IQR]. For normally distributed data, independent samples t-tests were used for comparisons between two groups; for non-normally distributed data, Mann-Whitney U rank-sum tests were used for comparisons between two groups. Count data are described using n (%), and Spearson chi-square test or Fisher's exact test was used to compare differences between groups. Two-tailed tests showed p < 0.05 as statistically significant. Ordinal data in this protocol, such as clinical TNM stage, TLS Mature, T level, P level, and TIL invasion level, were all included in the analysis as quantitative variables.

[0161] Mulberry plot drawing: drawn using the networkD3 package.

[0162] Kaplan-Meier (KM) analysis: The `survfit` function of the `survival` package was used to analyze the Kaplan-Meier survival function curves for disease-free survival and overall survival in different groups of the SYSU and TCGA cohorts. The log-rank test was used to assess whether there were significant prognostic differences between the different groups.

[0163] Correlation analysis: Performed using the pheatmap package, Spearman correlation analysis was used to describe the correlation between quantitative variables that are not normally distributed. A p-value less than 0.05 was considered statistically significant.

[0164] Univariate and multivariate Cox regression analysis: The coxph function from the survival package was used to perform Cox proportional hazards regression analysis on each covariate individually. Covariates significantly associated with survival time were identified. Based on the univariate analysis, statistically significant variables were selected for multivariate Cox regression, and a multivariate Cox regression model was established using the coxph function.

[0165] Nonlinear analysis: The `rcs` function from the `rms` package was used to convert continuous variables into restricted cubic splines and incorporated into a multifactor Cox model. The number of RCS nodes (k-value) was selected based on the AIC value. The AIC values ​​of all models from k=3 to k=8 were compared, and the k-value with the smallest AIC value was selected as the optimal number of nodes for the restricted cubic splines.

[0166] Predictive Model Construction: A predictive model was constructed based on the results of multivariate Cox regression analysis. The `predict` function was used to predict survival probabilities. The `survivalROC` package was used to evaluate the model's predictive accuracy and plot the ROC curve. The `timeROC` package was used to calculate the area under the curve (AUC) value based on the time-dependent ROC curve. Decision curve analysis (DCA) was performed using the `stdca` package to evaluate clinical decision utility. The calibration of the nomogram was evaluated through multiple resampling using the `calibrate` function and the `bootstrap` method, and the calibration curve was plotted using the `plot` function. The `nomogram` function in the `rms` package was used to plot the nomogram based on the Model5 multivariate Cox model.

[0167] Online nomogram creation: Based on the filtered Model 5, a dynamic nomogram is created using DynNom. The model is uploaded and deployed via shinyapp using the rsconnect package. The deployed dynamic nomogram is available as an open-source website: https: / / 4ravewaving.shinyapps.io / TLS_Nomogram / .

[0168] like Figure 4 The diagram shown is a flowchart illustrating the research process for an embodiment of the present invention.

[0169] Results and Analysis: Clinical characteristics of the study population: A total of 235 patients were included in the SYSU cohort, all of whom were Asian. 177 patients (approximately 75%) were in the TLS(+) group, and 58 patients (approximately 25%) were in the TLS(-) group. The differences in baseline characteristics between the TLS(+) and TLS(-) groups are shown in Table 2.1. The results showed significant statistical differences between the two groups in clinical T stage, clinical TNM stage, pathological complete response (pCR) rate, Ki67 expression, and disease-free survival (DFS) (p<0.05). Specifically, compared to the TLS(-) group, the TLS(+) group had lower clinical T stage, lower clinical TNM stage, lower Ki67 expression, and lower DFS. Surprisingly, the TLS(+) group had a lower pCR rate, which may be related to the lack of corresponding inflammatory factors after complete tumor regression following neoadjuvant therapy; however, TLS was still found in the slides of many pCR cases. (As shown in Table 2)

[0170] The TCGA cohort included 147 patients, primarily Europeans (89) and Africans (47), with only 8 Asians. Due to differences in time and treatment advancements, none of the patients in the TCGA cohort received neoadjuvant therapy, unlike the SYSU cohort. Height and weight data for the TCGA cohort were unavailable because the TCGA database does not publicly disclose data on breast cancer cohorts. Immunohistochemistry of Ki67 in breast cancer was not widely used when the TCGA database was established, resulting in a lack of Ki67-related data. Therefore, the expression level of MIB1, a gene regulating Ki67 expression, was used instead, which may introduce some bias. As shown in Table 2, the TLS(+) group had a lower clinical T stage compared to the TLS(-) group.

[0171] Because the TCGA cohort has many missing clinical endpoints and consists entirely of individuals who have not received neoadjuvant chemotherapy, its data structure makes it unsuitable as an external validation set for validating the constructed predictive model. Therefore, the TCGA cohort was chosen as the sensitivity analysis cohort to validate the conclusions and risk stratification in individuals who have not received neoadjuvant chemotherapy, in order to assess the robustness of the conclusions in the context of patients with triple-negative breast cancer who have not received neoadjuvant chemotherapy.

[0172] Table 2. Baseline Clinical Characteristics of the SYSU and TCGA Cohorts

[0173]

[0174]

[0175]

[0176] 1. Mean ± SD; Median [IQR]; n (%).

[0177] 2. Independent samples t-test, Mann-Whitney U rank-sum test, Spearson chi-square test, Fisher exact test; p < 0.05 is statistically significant and is shown in bold.

[0178] Pathological and tertiary lymphoid structure characteristics of the study population: Based on previous studies [31,32], in addition to spatial and maturational features associated with TLS, prognostic TIL and LVI were additionally assessed. All metrics of the SYSU cohort and some metrics of the TCGA cohort were evaluated according to the criteria in Method 2.3.6.

[0179] In the SYSU cohort, a total of 235 patients underwent pathological section evaluation. Of these, 58 (24.68%) were in the TLS(-) group and 177 (75.32%) were in the TLS(+) group. Regarding the abundance TLS in the near-tumor (T) region, grade 0 (no TLS) was found in 78 cases (33.19%), grade 1 in 43 cases (18.30%), grade 2 in 90 cases (38.30%), and grade 3 in 24 cases (10.21%). The mean TLS count in the T region was 5.12, and the mean percentage of T region TLS to the total section area was 4.73‰. The abundance of TLS in the distant tumor (P) region was graded as follows: 160 cases (68.09%) were grade 0, 36 cases (15.32%) were grade 1, 37 cases (15.74%) were grade 2, and 2 cases (0.85%) were grade 3. The average count of TLS in the P region was 1.52. The average percentage of TLS in the P region to the total slice area was 0.78‰.

[0180] (Table 3)

[0181] In the TCGA cohort, a total of 147 patients underwent pathological section evaluation. Of these, 48 (32.65%) were in the TLS(-) group and 99 (67.35%) were in the TLS(+) group. The primary assessment of this cohort included TLS partition abundance rating, TLS number, and vascular invasion status (LVI). In the near-tumor (T) region, TLS abundance rating was grade 0 in 51 cases (34.69%), grade 1 in 24 cases (16.33%), grade 2 in 58 cases (39.46%), and grade 3 in 14 cases (9.52%); the mean count of T region TLS was 5.54. In the distant-tumor (P) region, TLS abundance rating was grade 0 in 85 cases (57.82%), grade 1 in 22 cases (14.97%), and grade 2 in 40 cases (27.21%). There were no contiguous P region TLS, i.e., 0 cases were grade 3; the mean count of P region TLS was 2.80. Furthermore… Due to the lack of Ki67 staining validation, maturity levels were not scored for the TCGA cohort. (Table 3)

[0182] In both cohorts, the abundance of TLS in the T region was mainly at levels 0 and 2. There were almost no TLS groups at level 3 in the P region. Comparing the two cohorts, there were no significant differences in TLS grouping or TLS abundance in the T region. However, there were significant differences in TLS abundance and vascular invasion in the P region; the TCGA cohort had more TLS abundance and less vascular invasion in the P region, which may be related to the lack of immunohistochemical validation.

[0183] In the SYSU queue, the relationship between TLS spatial distribution and maturity level was preliminarily explored.

[0184] At the TLS level, a total of 1570 TLSs were detected in 235 patient slices from the SYSU cohort. Among different regions, 1213 (77.26%) were found in the T region and 357 (22.74%) in the P region. Among TLSs of different maturity levels, 589 (27.52%) were E-TLS, 944 (60.13%) were PFL-TLS, and 37 (2.36%) were SFL-TLS. The spatial distribution of TLS was significantly correlated with maturity level (r = 0.357, p < 0.001, e.g., ...). Figure 5 (As shown). Among them, SFL-TLS is relatively rare, T-area TLS is mainly PFL-TLS, and P-area TLS is mainly E-TLS.

[0185] At the patient and slide levels, among the 235 patients in the SYSU cohort, 58 (24.68%) had a TLS maturity score of 0, 65 (27.66%) had a score of 1, 98 (41.70%) had a score of 2, and 14 (5.96%) had a score of 3 (Table 2.2). The TLS maturity score was also correlated with the TLS abundance levels in the T and P regions (r = 0.796, p < 0.001; r = 0.279, p < 0.001; ...). Figure 5 (As shown). Among them, Figure 5 A. TLS level rating. The left square represents the distribution area of ​​TLS, and the right square represents the maturity stage of TLS. The white numbers in the squares represent the number of TLS. B. WSI level rating. The middle square represents the maturity rating, and the two sides represent the TLS abundance ratings in the T and P regions. Specific rating criteria are detailed in Method 2.4.5. r is the Spearson correlation coefficient; p < 0.05 is statistically significant.

[0186] Table 3. Baseline Pathological Assessment Characteristics of the SYSU and TCGA Cohorts

[0187]

[0188]

[0189] 1. Mean ± SD; Median [IQR]; n (%).

[0190] 2. Independent samples t-test, Mann-Whitney U rank-sum test, Spearson chi-square test, Fisher exact test; p < 0.05 is statistically significant and is shown in bold.

[0191] Relationship between Grade III lymphoid structure and disease-free survival and overall survival: During a median follow-up of 26.9 months, among the 235 patients in the SYSU cohort who received neoadjuvant chemotherapy, 76 (32.3%) experienced a primary outcome event (disease progression or death), and 41 (17.4%) experienced a secondary outcome event (all-cause mortality). As shown in the KM survival curves, the prognostic advantage of the TLS(+) group in terms of DFS was significant compared to the TLS(-) group (p<0.001). Figure 6 (As shown in Figure A). Although there was no significant difference in survival between the two groups regarding the secondary outcome of all-cause mortality (p = 0.11, as shown in Figure A). Figure 6 (As shown in Table C), the TLS(+) group still showed a better trend. However, since the TLS(+) group had a better clinical TNM stage than the TLS(-) group (p = 0.03, Table 2.1), it may affect the prognosis to some extent. Therefore, multivariate analysis should be used to correct for the influence of clinical TNM stage in subsequent studies.

[0192] Of the 147 patients in the TCGA cohort who did not receive neoadjuvant chemotherapy, 31 (21.0%) experienced disease progression or death during a median follow-up of 26.3 months, and 23 (15.6%) died from all causes. A comparison revealed that the overall survival rates of the two cohorts were roughly the same, while the disease-free survival rates differed significantly (chi-square test p = 0.001). This is presumably because the TCGA-BRCA project began in 2006, and the detection technology and follow-up conditions at that time were relatively underdeveloped, leading to some relapses and metastases in the TCGA cohort not being detected in a timely manner. Furthermore, on the KM survival curves, the TLS(+) group did not show better disease-free survival or overall survival (e.g., ...). Figure 6 B, Figure 6 (As shown in C). This may be related to the various heterogeneous confounding factors of TLS. Figure 6 In the AB.SYSU and TCGA cohorts, the KM survival curves for disease-free survival (DFS) of TLS(+) and TLS(-) groups were analyzed, and differences between groups were tested using log rank. The KM survival curves for overall survival (OS) of TLS(+) and TLS(-) groups in the CD.SYSU and TCGA cohorts were also analyzed, and differences between groups were tested using log rank.

[0193] Correlation between clinicopathological features and tertiary lymphoid structure features: When constructing a Cox regression model, if there is a high or exact correlation between the included explanatory variables, the model estimation will be distorted. Therefore, it is necessary to perform correlation analysis before including explanatory variables. For collinear variables, selective inclusion or merging them into a single variable should be adopted.

[0194] In Spearson correlation analysis, two variables with p < 0.001 are defined as having a sufficiently significant collinear relationship.

[0195] Among clinical characteristics, age was significantly correlated with menopause status, Her2, and height; clinical T stage, clinical N stage, clinical TNM stage, and the number of positive axillary lymph nodes were significantly correlated; and height was significantly correlated with weight. Figure 7 By screening and merging relevant variables, age, clinical TNM stage, Her2, and Ki67 were selected from clinical characteristics for univariate regression analysis, while height and weight data were combined into Body Mass Index (BMI) for inclusion in the analysis.

[0196] In pathological features, for both T-region and P-region TLS, level, area ratio, and count [level, per (‰), count], as three indicators for evaluating TLS abundance, showed significant correlations, which indirectly proves the reliability and reproducibility of the evaluation method. TLS maturity was highly correlated with the TLS abundance indicators in the T region, especially the T level, suggesting a high degree of collinearity between the TLS abundance indicators in the T region and TLS maturity. Furthermore, TIL infiltration level was also highly positively correlated with TLS maturity and the TLS abundance indicators in the T region. Figure 7 ).exist Figure 7 The image shows a heatmap of correlations between multiple clinicopathological features and tertiary lymphoid structure features. Both the x and y axes represent clinicopathological features, and the color and circle size represent correlation coefficients. Red indicates a positive correlation, blue indicates a negative correlation, and darker colors and larger circles indicate a stronger correlation. ***p<0.001, only those p<0.001 are sufficiently significant and are indicated on the graph. TLS Mature: The maturity level of TLS. T_per(‰) and P_per(‰): The percentage of T-region TLS and P-region TLS in the total tissue section area. T_per(‰) = TS-TLS / S-WSI*1000; P_per(‰) = PS-TLS / S-WSI*1000. T count and P count: The number of T-region TLS and P-region TLS. T level and P level: The level of T-region TLS and P-region TLS, see Methods for details.

[0197] Linear correlation between tertiary lymphoid structural features and prognosis

[0198] First, we analyzed the relationship between each variable and DFS using univariate Cox regression to initially identify and include indicators related to the primary outcome DFS after multivariate correction.

[0199] In univariate Cox regression analysis, clinical TNM stage, Ki67 level, P-region TLS abundance indices [P_per(‰) and P count], and vascular invasion LVI were significant adverse prognostic factors, while TLS maturity, T-region TLS abundance indices [Tlevel, T_per(‰), and T count], pathological complete remission (pCR), and TIL invasion degree were significant favorable prognostic factors. Figure 8 It can be observed that P level performs worse prognostically than P_per(‰) and P count. This is likely because P level, as a semi-quantitative indicator, loses more information during assessment compared to the other two indicators, thus significantly impacting its prognostic value. In further analysis, to avoid ignoring important variables, a p<0.10 threshold was set, and therefore BMI was also included.

[0200] Subsequently, multivariate Cox regression analysis was used to correct for confounding factors, further exploring the role of TLS characteristics in prognosis. Based on the correlation analysis in section 3.4 and the univariate regression analysis above, TLS maturity was selected as the primary variable in Model 1, the TLS area ratio in region T and region P as the primary variables in Model 2, the number of TLS in region T and region P as the primary variables in Model 3, and the TLS level in region T and the number of TLS in region P as the primary variables in Model 4.

[0201] Five indicators—clinical TNM stage, BMI, pCR, Ki67, and LVI—were included as covariates for adjustment to assess the relationship between TLS characteristics and disease-free survival (DFS) and overall survival (OS).

[0202] In Model 1, TLS maturity was found to be a significant favorable prognostic factor for DFS (HR 0.575, 95% CI: 0.443–0.747, p < 0.001). Figure 9 However, the relationship with OS was not significant (HR 0.758, 95% CI: 0.540-0.747, p = 0.109). Figure 9 ).

[0203] In Model 2, the TLS area ratio [T_per(‰)] in region T was determined to be a significant favorable prognostic factor for DFS (HR 0.851, 95% CI: 0.775–0.934, p = 0.001). Figure 10 It was a significant protective factor against OS (HR 0.905, 95% CI: 0.829-0.987, p = 0.024). Figure 10 The TLS area ratio in the P region [P_per(‰)] was a significant disadvantage for DFS (HR 1.183, 95% CI: 1.044-1.34, p = 0.008). Figure 10 Additionally, it may indicate an unfavorable trend in OS, although this trend is not significant (HR 1.168, 95% CI: 0.998-1.367, p = 0.052). Figure 10 ).

[0204] In Model 3, the number of TLS in region T (T count) was a significant protection factor for DFS (HR 0.903, 95% CI: 0.847–0.962, p = 0.002). Figure 11 However, the relationship with OS was not significant (HR 1.041, 95% CI: 0.978–1.108, p = 0.209). Figure 11 The number of TLS in the P region (P count) has no significant relationship with either DFS or OS.

[0205] In Model 4, for ease of model application, a feature combination of T-region TLS level (T level) and P-region TLS count (P count) was selected. The T-region TLS level (T level) is not only a significantly favorable prognostic factor for DFS (HR 0.52, 95% CI: 0.402–0.671, p < 0.001), Figure 12 It was a significant protective factor against OS (HR 0.681, 95% CI: 0.496-0.934, p = 0.017). Figure 12 The number of TLS in the P region (P count), after adjustment, was a significant adverse factor for DFS (HR 1.09, 95% CI: 1.022-1.163, p = 0.009). Figure 12 ).

[0206] In summary, Model 1 and Model 2 demonstrate that the maturity of TLS and TLS in the proximal (T) region can indicate a more positive disease-free survival (DFS) in triple-negative breast cancer, while TLS in the distal (P) region is associated to some extent with a poor DFS prognosis in triple-negative breast cancer, consistent with the hypothesis. To further verify the effect of TLS region localization in Model 2, TLS abundance was assessed using multiple evaluation methods. Sensitivity analyses were conducted using Model 3 (quantifying TLS by counting) and Model 4 (quantifying TLS abundance by grading), demonstrating that the conclusions remain robust across different evaluation systems.

[0207] Among them, Figure 8 In the diagram, the univariate Cox regression forest plot shows the univariate Cox regression results and p-values ​​for clinical, pathological, and TLS characteristics on DFS. Colors represent different hazard ratios (HRs). Red represents favorable factors (HR>1), blue represents unfavorable factors (HR<1), and bold represents factors included in subsequent analyses.

[0208] exist Figure 9 Five covariates—clinical TNM stage, BMI, pCR, Ki67, and LVI—were included for adjustment, with clinical TNM stage treated as a continuous variable. The relationship between TLS maturity and DFS and OS was analyzed. Forest plots show the results and p-values ​​of multivariate Cox regression on DFS and OS. The colors of TLS features represent different hazard ratios (HRs): red represents favorable TLS features (HR>1), blue represents unfavorable TLS features (HR<1), and gray represents insignificant.

[0209] exist Figure 10 Five covariates—clinical TNM stage, BMI, pCR, Ki67, and LVI—were included for adjustment, with clinical TNM stage treated as a continuous variable. The relationship between the area ratio of TLS in the T region and the area ratio of TLS in the P region and DFS and OS was analyzed. Forest plots were used to display the multivariate Cox regression results and p-values ​​for DFS and OS. The colors of the TLS features represent different hazard ratios (HRs), with red representing favorable TLS features (HR>1) and blue representing unfavorable TLS features (HR<1).

[0210] exist Figure 11 Five covariates—clinical TNM stage, BMI, pCR, Ki67, and LVI—were included for adjustment, with clinical TNM stage treated as a continuous variable. The relationship between T-region TLS counts, P-region TLS counts, and DFS and OS was analyzed. Forest plots were used to display the multivariate Cox regression results and p-values ​​for DFS and OS. The colors of the TLS features represent different hazard ratios (HRs): red represents favorable TLS features (HR>1), blue represents unfavorable TLS features (HR<1), and gray represents insignificant.

[0211] exist Figure 12 Five covariates—clinical TNM stage, BMI, pCR, Ki67, and LVI—were included for adjustment, with clinical TNM stage treated as a continuous variable. The relationship between T-region TLS grading, P-region TLS count, and DFS and OS was analyzed. Forest plots were used to display the multivariate Cox regression results and p-values ​​for DFS and OS. The colors of the TLS features represent different hazard ratios (HR): red represents favorable TLS features (HR>1), blue represents unfavorable TLS features (HR<1), and gray represents insignificant.

[0212] Nonlinear Association Between Tertiary Lymphoid Structure Characteristics and Prognosis: Among the characteristics of tertiary lymphoid structures (TLS), area ratio [per (‰)] and count are continuous variables, while a key assumption of Cox regression analysis is that the independent variables and survival are linearly associated. Further investigation into whether the association between T_per (‰), P_per (‰), T count, P count, and prognosis is nonlinear may contribute to a deeper understanding and analysis of the prognostic impact of TLS location heterogeneity.

[0213] Restricted Cubic Splines (RCS) are a regression analysis method for fitting nonlinear data. They primarily involve segmenting the independent variable by nodes during the test, transforming it into several linear segments, and then modeling each segment using a cubic function. RCS is used to conduct a nonlinear study of the continuous characteristics of disease-free survival (DFS) and DFS, including the area ratio and count of TLS. The model used in the multivariate Cox regression analysis is followed for adjusting for confounding factors. The choice of the number of RCS nodes (k value) is closely related to the RCS results; the optimal k value is found using the principle of minimizing the AIC (Akaike information criterion). The AIC values ​​of all models from k=3 to k=8 are compared, and the k value with the smallest AIC value is selected as the optimal number of nodes for the restricted cubic spline.

[0214] RCS analysis results showed that, regardless of area ratio or count, after RCS conversion, T-region TLS still indicated a more positive DFS in triple-negative breast cancer, while P-region TLS remained associated with a poor DFS prognosis in triple-negative breast cancer (e.g., Figure 13 The mean AD (Total p < 0.05) was observed. This is consistent with the conclusions of the multivariate Cox regression analysis. Furthermore, the T-region TLS, including area ratio and count, showed a highly significant nonlinear association with DFS (e.g., ...). Figure 13 As shown in A: k = 3, as Figure 13As shown in C: k = 3, nonlinear p < 0.001). The threshold line is x = 1.5 (e.g. Figure 13 As shown in Figure A) and x = 3.0 (as shown in Figure A) Figure 13 As shown in Figure C), this indicates that when the TLS area ratio in region T is less than 1.5‰ or the count is less than 3, the increase in TLS in region T actually increases the negative prognostic effect (HR>1); only when the TLS area ratio in region T is greater than 1.5‰ or the count is greater than 3, is the increase in TLS in region T a prognostic protective factor (HR<1). In contrast, TLS in region P shows a linear correlation with survival (…). Figure 13 B: k = 5, Figure 13 In the middle D: k = 3, nonlinear p > 0.05).

[0215] An interesting conclusion was reached: in this study, fewer TLS in the proximal tumor (T) region became a significant prognostic risk factor. Only when there were more TLS in the proximal tumor (T) region could it actually provide protection. TLS in the distal tumor (P) region consistently remained a risk factor. Furthermore, an interesting correlation was found between the threshold line T count = 3.0 and the T level, and this threshold line perfectly matched the established definition of T level. According to the designed definition of T level, TLS in the proximal tumor (T) region of 1-3 is classified as level 1, i.e., T level = 1 is high risk, and T level > 1 is low risk. Among these, in… Figure 13 In this context, AD. Different models correspond to the confounding factor adjustment strategies in the multivariate Cox regression analysis above. The variables selected for different models are as follows: Based on clinical TNM staging, BMI, pCR, Ki67, and LVI, Model 2: T_per (‰) and P_per (‰); Model 3: T count and P count; Model 4: T level and P count. The y-axis represents the hazard ratio (HR) for DFS, and the x-axis represents the continuous characteristic of TLS. The solid yellow line represents the hazard ratio for different x-axis values, and the blue dots on the solid yellow line represent the number and values ​​of nodes (k). The light blue area represents the 95% confidence interval. The dashed black line is the reference line for HR=1. The solid red line is the threshold line for HR=1. Figure 13 The threshold line for A is x = 1.5. Figure 13 The threshold line for C is x = 3.0.

[0216] A disease-free survival (DFS) prediction model for triple-negative breast cancer was constructed based on the characteristics of the tertiary lymphoid structure, and its performance was compared.

[0217] Based on the multivariate Cox regression analysis in the previous section, four prediction models, Model 1 to Model 4, were constructed based on TLS features. According to the conclusion of the previous section, T count = 3.0 may be a cutoff point of significant importance for DFS prognosis. Therefore, the original four-category T level of TLS abundance rating system used in other literature was modified to obtain a two-category new T level: (1) no or low abundance (level 0) – 0 to 3 TLS observed in the T region; (2) medium to high abundance (level 1) – more than 3 TLS observed in the T region. Based on the new T level, the T level in Model 4 was replaced to construct a new model, Model 5. Figure 14 and Figure 15 The figure captions show the specific variables included in each model.

[0218] To screen for better models, the ROC curves, DCA curves, one-year disease-free survival calibration curves, three-year disease-free survival calibration curves, and C-index (Table 4) of Model 1-Model 5 were compared.

[0219] To assess the model's discriminative power, time-dependent ROC curve analysis quantified the model's ability to accurately distinguish different survival states over one and three years. A high AUC value indicates good discriminative power in predicting survival outcomes. Comparing the AUC values ​​of different models, after reducing the quantification of TLS abundance in the proximal tumor region by using new T-level binary variables, Model 5 (AUC1-Year = 0.796, AUC3-Year = 0.848) showed better discriminative power than Model 3 (AUC1-Year = 0.778, AUC3-Year = 0.816) using T-count variables and Model 1 (AUC1-Year = 0.791, AUC3-Year = 0.841) using TLS maturity variables. Furthermore, the difference in discriminative power compared to Model 4 (AUC1-Year = 0.798, AUC3-Year = 0.848) using T-level four-category variables was not significant.

[0220] The DCA curve assesses the clinical utility of a model by calculating the net benefit of using the corresponding model compared to not using any model (None) or treating all patients (All) across a range of potential threshold probabilities. The DCA curves for Models 1-3 intersect with the blue line at low predictive risk (approximately <0.1), indicating that the net benefit of Models 1-3 is low at low risk thresholds, and their use should be avoided. Models 4-5, however, exhibit high net benefit across almost the entire threshold probability range, demonstrating significant clinical benefit from their use.

[0221] The calibration curve assesses the consistency between the model's predicted survival probability and the actual observed survival probability in patients. The dashed reference line represents the ideal model, where the predicted and actual rates are exactly the same. The closer the fitted curve is to the diagonal, the closer the predicted and actual rates are, and the more consistent the model is. The bootstrap resampling method simulates the process of drawing multiple samples from the population, thus providing a more robust estimate of the model's calibration performance. Comparing one-year survival rates with three-year predictive rates reveals that Model 5 outperforms Models 1-4 in predictive consistency.

[0222] The C-index primarily measures the consistency between the risk score predicted by the model and the actual survival outcome. It considers information about individual pairings, time, and events, making it more effective than traditional ROC analysis as a metric for evaluating the discriminative power of survival prediction models. Models 1-5 all exhibited similar C-indexes (0.77-0.80), indicating good model discriminative power (Table 4). Among them, in... Figure 14 In the figure, the ROC curves for Model 1 through Model 5 are shown. Three-year survival prediction is marked with a solid red line, and one-year survival prediction is marked with a solid blue line. The horizontal axis represents the false positive probability (FPR, 1-Specificity), and the vertical axis represents the true positive probability (TPR, Sensitivity). The area under the curve (AUC) is marked on the graph; it is the area enclosed by the curve and the coordinate axes. The closer the AUC is to 1, the higher the model's predictive discrimination and accuracy. An AUC less than 0.5 indicates that the model's predictive ability is worse than random prediction.

[0223] B. Multivariate Cox regression decision curves for Model 1-Model 5. The horizontal axis represents the threshold probability, and the vertical axis represents the net benefit. The green line (None) represents no treatment using any model, with a net benefit of 0. The blue line (All) represents treatment for all patients. Figure 15This is a calibration plot of the one-year survival rate (blue line) and three-year survival rate (red line) for a multivariate Cox regression model. The calibration curve shows the relationship between the actual and predicted incidence rates. The horizontal axis represents the predicted survival rate, and the vertical axis represents the actual survival rate. The dashed diagonal line serves as a reference line, representing that the predicted probability equals the actual probability. The closer the fitted curve is to the diagonal line, the stronger the correspondence between the predicted and actual incidence rates, indicating a higher model quality.

[0224] Model 1: cTNM+BMI+Ki67+pCR+LVI+TLS Mature.

[0225] Model 2: cTNM+BMI+Ki67+pCR+LVI+T_per(‰)+P_per(‰).

[0226] Model 3: cTNM+BMI+Ki67+pCR+LVI+T count+P count.

[0227] Model 4: cTNM+BMI+Ki67+pCR+LVI+T level+P count.

[0228] Model 5: cTNM+BMI+Ki67+pCR+LVI+new T level+P count.

[0229] Table 4 shows the C-index and standard error of each model.

[0230] Model C-index SE Model 1 0.78 0.03 Model 2 0.79 0.02 Model 3 0.77 0.03 Model 4 0.80 0.02 Model 5 0.79 0.02

[0231] Final Model Construction and Clinical Application: Considering both the accuracy comparison results and the ease of application, Model 5 was ultimately selected as the final prediction model. The specific modeling coefficients of Model 5 are shown in Table 5 below. Based on the seven parameters of Model 5, a Nomogram prediction model for disease-free survival (DFS) prognosis of triple-negative breast cancer was constructed with disease progression and death as target events. Figure 16 In the Nomogram prediction model, a score is obtained for each predictor of a case, and the predicted probability corresponding to the total score obtained by adding these scores is the probability of one-year and three-year disease-free survival for that case. Figure 16A. A nomogram built based on the Model 5 prediction model. Each predictor has a corresponding score, and the total score corresponds to the one-year and three-year disease-free survival rates. B. An interactive interface for an online dynamic nomogram built based on the prediction model. The image shows the input content for two triple-negative breast cancer cases. C. The display interface for the online dynamic nomogram built based on the prediction model, including numerical summaries, survival rate prediction visualization, and a graph showing survival rate changes with follow-up time. (Items 1-2 are...) Figure 16 The one-year and three-year disease-free survival rates for Case 1 (black and blue lines, black survival rate curve); numbers 3-4 are... Figure 16 The one-year and three-year disease-free survival results for Case 2 (red and green lines, blue survival curve).

[0232] Table 5 shows the regression coefficients of the variables in Model 5. 1

[0233] Model 5 coef exp(coef) SE pvalue cTNM 0.420 1.522 0.100 <0.001 BMI 0.079 0.079 0.030 0.009 Ki67 0.122 1.130 0.292 0.676 pCR -1.027 -1.027 0.612 0.094 LVI 1.144 3.138 0.248 <0.001 Pcount 0.082 1.085 0.033 0.013 newTlevel -1.328 0.265 0.286 <0.001

[0234] 1 coef, regression coefficient. exp(coef), hazard ratio, i.e., HR value. SE, standard error.

[0235] Nomograms are a traditional, visual statistical predictive model that integrates multiple predictor variables to estimate patient disease prognosis. However, traditional nomograms are typically static in nature, making them ill-suited to adapting to new data or changes. Furthermore, these models cannot be updated in real time, limiting their potential practical applications. It is worth noting that traditional nomograms are often paper-based or require manual reading, which can lead to user misinterpretation or calculation errors.

[0236] Dynamic nomograms, or interactive nomograms, are contemporary enhancements and iterations of traditional models, typically built on web or software applications. Users can input data through simple input and clicks in an interactive interface, and the model calculates and updates the results in real time, reducing computational errors. Therefore, this model can calculate and update results instantly, significantly reducing the possibility of computational errors. Furthermore, due to their web- or application-based nature, these dynamic nomograms can be easily accessed from any device with an internet connection, making them readily applicable in clinical settings.

[0237] The predictive model was deployed online using the DynNom package in R and a shinyapps.io cloud account linked to RStudio. The model is open source and can be accessed directly over the network.

[0238] The access address is as follows: https: / / 4ravewaving.shinyapps.io / TLS_Nomogram / .

[0239] Two cases of triple-negative breast cancer were included for comparison. In the conventional nomogram: Case 1 (cTNM: IIIA, 90 points; BMI: 24, 10 points; Ki67: Low, 0 points; pCR: No, 15 points; LVI: Yes, 15 points; new T level: No to Low, 20 points; P count: 2, 5 points), with a total score of 155 points, had a one-year disease-free survival rate of approximately 0.65% and a three-year disease-free survival rate of approximately 0.25%. Case 2 (cTNM: IIIA, 90 points; BMI: 24, 10 points; Ki67: Low, 0 points; pCR: No, 15 points; LVI: No, 0 points; new T level: Medium to High, 0 points; P count: 2, 5 points), with a total score of 120 points, had a one-year disease-free survival rate of approximately 0.97% and a three-year disease-free survival rate of approximately 0.8%. In the dynamic nomogram, the same baseline characteristics of the two cases were entered in the interactive interface. Serial numbers 1-2 indicate that the one-year disease-free survival rate for case 1 is 0.680 and the three-year disease-free survival rate is 0.226; serial numbers 3-4 indicate that the one-year survival rate for case 2 is 0.970 and the three-year disease-free survival rate is 0.880. In addition to being more convenient and faster, the dynamic nomogram also provides more accurate survival rate predictions and can provide visualization of 95% confidence intervals.

[0240] Prognostic stratification based on immune risk score and spatial typing of tertiary lymphoid structure: Based on the regression coefficients of Model 5, the formula for calculating the immune risk score based on TLS was derived.

[0241] TLS Risk Score = 0.082 * P count - 1.328 * new Tlevel (1)

[0242] Wherein, new T level: none or low abundance = 0, medium to high abundance = 1.

[0243] In the SYSU cohort, TLS immunity risk scores were calculated and divided into low-risk and high-risk groups using the median as the cut-off value. KM survival analysis showed significant risk stratification in the primary outcome DFS (log-rank p < 0.0001). Figure 17 (A). However, in the secondary outcome overall survival (OS), the survival difference was not significant (log-rank p = 0.091, e.g., Figure 17 (C) but the low-risk group showed a significantly better prognostic trend.

[0244] In the TCGA cohort that did not receive neoadjuvant therapy, the TLS immune risk score was applied, and the median was used as the cutoff value to divide the patients into low-risk and high-risk groups. Compared to the TLS immune risk score, which directly grouped patients based on the presence or absence of TLS in Section 3.3, this method achieved DFS (log-rank p = 0.0009, as shown in Section 3.3). Figure 17 (B) and OS (log-rank p = 0.0010, such as B ... Figure 17 The significant distinction between D) risk and TLS immunity risk score demonstrates robustness in other heterogeneous cohorts.

[0245] To further verify the conclusions, TLS spatial distribution was used to classify them. Type I: T region has no or few TLS, P region has no TLS; Type II: T region contains many TLS, P region has no TLS; Type III: T region contains many TLS, P region contains TLS; Type IV: T region has no or few TLS, P region contains TLS. A schematic diagram of the four TLS spatial classifications is shown below. Figure 17 As shown.

[0246] KM survival analysis was performed on four TLS spatial classifications based on DFS and OS in the SYSU and TCGA queues. It was found that types II and III had better prognoses than types I and IV. Type IV with TLS in the P region had a worse prognosis than type I without TLS at all. Figure 19 This further confirms the conclusion.

[0247] Among them, Figure 17 In this study, AD was stratified based on the TLS immune risk scores calculated using Model 5 regression coefficients. The median risk score was used as the cutoff point for dividing participants into low-risk and high-risk groups. KM survival analysis was performed on DFS and OS in the SYSU and TCGA cohorts, respectively. Differences between groups were tested using log rank.

[0248] exist Figure 18 In the context of TLS, spatial classification is as follows: Type I: T region has no or few TLS, P region has no TLS; Type II: T region has many TLS, P region has no TLS; Type III: T region has many TLS, P region has TLS; Type IV: T region has no or few TLS, P region has TLS.

[0249] According to research, TLS in the T region are mostly PFL-TLS containing Ki67+B cells, while TLS in the P region are mostly E-TLS containing Ki67-B cells.

[0250] exist Figure 19In the AD group, stratification was performed based on the TLS spatial classification. KM survival analysis was conducted on DFS and OS in the SYSU and TCGA queues, respectively. Differences between groups were tested using log rank.

[0251] Sensitivity analysis: In the TCGA cohort, the abundance of TLS in region T was subjected to the same nonlinear restricted cubic spline analysis based on DFS (e.g., Figure 20 (A) It was found that after incomplete adjustments to the variables based on Model 5, the T-region TLS count exhibited an inflection point in HR at x = 3.0, meaning that when x < 3.0, the T-region TLS count was a risk factor for disease-free survival (HR > 1). However, when x > 3.0, the T-region TLS count became a protective factor against disease-free survival, although this protective effect became insignificant as the number of T-region TLS increased. This finding suggests robustness to the non-linear association between T-region TLS and disease-free survival, and that the T-region TLS count at 3 may be a statistically significant cutoff point. Furthermore, it indicates that TLS is only associated with a more positive DFS prognosis when there are sufficiently high numbers of T-region TLS.

[0252] Multivariate Cox regression analysis was also performed on variables in Model 5 (excluding BMI and pCR) based on DFS in the TCGA cohort (e.g. Figure 20 (B) In the TCGA cohort, TLS abundance rating in the T region was significantly associated with a favorable disease-free survival (DFS) outcome (HR = 0.335, p = 0.022), while clinical TNM stage (HR = 1.835, p < 0.001) and vascular invasion (HR = 9.415, p < 0.001) were significantly associated with a poor outcome. However, no significant association was found between TLS count in the P region and prognosis.

[0253] exist Figure 20 In this study, A. nonlinear analysis was performed on the adjusted T count in the TCGA queue. Due to missing BMI and pCR, the adjustment variables were cTNM, Ki67, and LVI. The y-axis represents the hazard ratio (HR) for DFS, and the x-axis represents the TLS count in the proximal (T) region. The solid yellow line represents the hazard ratio for different x-axis values, and the blue dots on the solid yellow line represent the number and values ​​of nodes (k). The light blue area represents the 95% confidence interval. The dashed black line is the reference line for HR=1. The solid red line is the threshold line for HR=1, with a threshold of x=3.0.

[0254] B. Multivariate Cox regression forest plot for disease-free survival (DFS) in the TCGA cohort. Due to missing BMI and pCR, the included variables were cTNM, Ki67, LVI, new T level, and P count from Model 5.

[0255] Discussion and Conclusion: This study revealed the complex role of TLS in triple-negative breast cancer patients who received neoadjuvant chemotherapy, based on both quantity and spatial location. The spatial distribution of TLS was closely related to the degree of maturity, with TLS in nearby tumors mainly exhibiting the characteristics of PFL-TLS, while TLS in distant tumors mainly exhibited the characteristics of E-TLS. This is not consistent with the findings of Ding et al. in intrahepatic cholangiocarcinoma

[32] , where TLS in distant tumors also exhibited the characteristics of E-TLS, while TLS in nearby tumors mainly exhibited the characteristics of SFL-TLS. In Ding et al.'s study

[32] , a total of 87 SFL-TLS were found, accounting for 15.2% of the total TLS (571). In the triple-negative breast cancer cohort, only 37 SFL-TLS were found, accounting for 2.4% of the total TLS (1570). Cross-cancer TLS studies

[57] have found that the quantity and quality of TLS differ in different cancer types. Therefore, it can be speculated that the proportion of fully mature TLS in triple-negative breast cancer is lower than that in intrahepatic cholangiocarcinoma, which may be related to the differences in factors that promote TLS maturation in different tumor immune microenvironments. This may suggest that promoting TLS maturation in triple-negative breast cancer could be a novel immunotherapy strategy.

[0256] Mature TLS and the abundance of TLS in the tumor-proximal region are effective predictors of good prognosis in DFS of triple-negative breast cancer, while the presence of TLS in the tumor-distal region is significantly associated with poor prognosis in DFS. In previous studies, the dual function of spatially different TLS has been recognized in intrahepatic cholangiocarcinoma

[32] , liver metastases from colon cancer

[31] , and clear cell renal cell carcinoma

[30] , but the underlying mechanism remains uncertain. Zhang et al.

[31] and Ding et al.

[32] found that TLS near the tumor contained more CD4+Bcl6+Tfh cells than TLS far from the tumor. Tfh cells are an important component in activating B cell proliferation, building germinal centers, and forming mature TLS, which helps to shape the anti-tumor immune microenvironment. The assessment of TLS maturity by Ki67 labeling proliferating B cells also proved that TLS near the tumor region is mostly PFL-TLS containing more Ki67+ proliferating B cells and is associated with better prognosis. Wolf et al.

[66] reported that B cells and their function are highly dependent on the presence and maturation of TLS, and highly proliferating and fully mature B cells grow into plasma cells and produce high-affinity antitumor IgG and IgA antibodies, thereby forming an anti-cancer immune microenvironment. This may explain one of the mechanisms by which the number and spatial location of TLS near the tumor region play a complex role in prognosis in studies.

[0257] Regarding the relationship between peritumoral TLS and adverse outcomes, this is consistent with previous findings in other subtypes of breast cancer

[47] . Previous studies in a mouse liver cancer model

[67] reported that peritumoral TLS reflects an inflammatory environment that supports tumor growth and may serve as a site for tumor progenitor cell migration. Sofopoulos et al.'s study on peritumoral TLS in breast cancer

[47] also found that many peritumoral TLSs far from the tumor were infiltrated by clusters of cancer cells. It can be hypothesized that peritumoral TLS may reflect the process by which tumor cells gradually migrate from the tumor area to the tumor periphery associated with invasion.

[0258] Although the results in the SYSU cohort can indirectly verify this hypothesis, the relationship between TLS near the tumor region and TLS far from the tumor region may be far more complex than imagined. A study in intrahepatic cholangiocarcinoma

[32] reported that as the abundance of peritumoral TLS increased, the frequency of Treg cells in intratumoral TLS increased significantly, which means that there is a potential link between peritumoral TLS and intratumoral TLS. In addition to its own immature and dysfunctional state, peritumoral TLS may also have an anti-tumor immune effect by destroying intratumoral TLS. However, the study did not further analyze the specific cell phenotypes of TLS in different regions of triple-negative breast cancer through multiplex immunofluorescence analysis, so the relevant conclusions could not be verified.

[0259] Compared to previous studies of similar types, a nonlinear analysis of TLS abundance in both proximal and distal tumor regions was performed, leading to a new conclusion: in the SYSU cohort, triple-negative breast cancer patients who received neoadjuvant chemotherapy and whose proximal tumor region TLS abundance count threshold exceeded 3.0 indicated a good prognosis. Furthermore, according to commonly used TLS abundance rating criteria in the literature [30-32], low-abundance proximal tumor region TLS may have a worse prognosis than TLS without proximal tumor region TLS, which indirectly supports the study results. Sensitivity analysis was also performed in the TCGA triple-negative cohort, confirming this conclusion, with the count threshold also being 3.0 in the TCGA cohort. The underlying mechanism may be related to the interaction between TLS, suggesting that there may be a positive feedback mechanism that promotes the maturation of proximal tumor region TLS, while low-abundance TLS is insufficient to reach the positive feedback threshold for TLS promotion.

[0260] Based on the findings of this study, the commonly used rating methods in the literature were modified in the construction of the prediction model. A new T-level was proposed, merging near-tumor region TLS without abundance and near-tumor region TLS with low abundance into a high-risk item, while near-tumor region TLS with medium to high abundance were classified as low-risk items. Multiple prediction models were constructed, and the final model was selected through performance comparison (AUC1-Year = 0.796, AUC3-Year = 0.848, C-index = 0.79). The model's risk score also achieved risk stratification in the TCGA cohort.

[0261] Conclusions: (1) In the triple-negative breast cancer cohort receiving neoadjuvant chemotherapy, the spatial distribution of TLS was closely related to maturity. Near-tumor TLS mainly exhibited the characteristics of PFL-TLS, while distant-tumor TLS mainly exhibited the characteristics of E-TLS. (2) Analysis revealed that the abundance of mature TLS and near-tumor TLS was associated with more positive disease-free survival, while the abundance of distant-tumor TLS was associated with poor disease-free survival prognosis. (3) The abundance of near-tumor TLS showed a significant non-linear correlation with disease-free survival prognosis. When there were fewer TLS in the near-tumor region (area ratio less than 1.5‰ or count less than 3), near-tumor TLS was associated with poor prognosis, and only when there were more TLS in the near-tumor region was it associated with more positive disease-free survival. (4) Four models were constructed based on commonly used standards, and a fifth model was constructed based on the conclusions. After comprehensively comparing the performance of the five models, the best model, Model 5, was obtained (AUC1-Year = 0.796, AUC3-Year = 0.848, C-index = 0.79). (5) A simple and convenient online dynamic Nomogram analysis tool was built based on the best model and made open source on the Internet for use by a wide range of users.

[0262] To achieve the above objectives, the present invention also provides a prognostic prediction system for neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure. This system is applied to the aforementioned prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure, such as... Figure 21 As shown, the system includes:

[0263] The data acquisition and processing unit is used to acquire paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer, and to slice and process the corresponding paraffin block of breast cancer tissue according to the paraffin block data of breast cancer tissue.

[0264] The first data generation unit is used to generate first-level data corresponding to TLS density and second-level data corresponding to TLS maturity based on the three-level lymphoid structure and the paraffin block data of the breast cancer tissue, and to generate estimated data corresponding to the level data; wherein, the first-level data is a semi-quantitative abundance level; and the second-level data is a maturity level.

[0265] The second data generation unit is used to construct a prediction model corresponding to neoadjuvant triple-negative breast cancer, and combine the estimated data to generate prediction data corresponding to the prognosis of neoadjuvant triple-negative breast cancer in real time.

[0266] The data acquisition and processing unit further includes:

[0267] The first processing module is used to perform H&E staining and Ki67 immunohistochemical staining on paraffin blocks of breast cancer tissue, and generate corresponding KFB format files.

[0268] The second processing module is used to convert the KFB format file into corresponding pathological image data in SVS format.

[0269] The third processing module is used to perform region delineation processing, area measurement processing, and ROI region cell number calculation processing on the SVS format WSI file respectively.

[0270] The second data generation unit further includes:

[0271] The first judgment module is used to acquire sample data corresponding to neoadjuvant triple-negative breast cancer, and to determine and estimate whether there are significant prognostic differences between different groups of samples.

[0272] The first data generation module is used to perform Cox proportional hazards regression analysis on each covariate individually and generate covariate data that are significantly associated with survival time.

[0273] The second data generation module is used to construct a multi-factor Cox model, and at the same time convert continuous variables into restricted cubic splines and add them into the multi-factor Cox model to generate corresponding node number data.

[0274] The second data generation unit further includes:

[0275] The third data generation module is used to generate data on the relationship between the three-level lymphoid structure and disease-free survival and overall survival based on the prediction model corresponding to neoadjuvant triple-negative breast cancer.

[0276] The fourth data generation module is used to acquire clinicopathological feature data corresponding to neoadjuvant triple-negative breast cancer, and generate correlation data between the clinicopathological feature data and the tertiary lymphoid structure feature data.

[0277] The fifth data generation module is used to create linear correlation data between the three-level lymphoid structure feature data and prognostic data, and to generate corresponding correction data based on the linear correlation data; wherein, the linear correlation data includes linear correlation data and non-linear correlation data;

[0278] The second data generation unit further includes:

[0279] The sixth data generation module is used to construct a DFS prediction model for triple-negative breast cancer based on the three-level lymphoid structure feature data, compare the performance of each prediction model, and generate model performance evaluation index data corresponding to the performance.

[0280] In the system embodiment of the present invention, the specific details of the method steps involved in the prognostic prediction of neoadjuvant-adjuvant triple-negative breast cancer based on the three-level lymphoid structure have been described above. That is to say, the functional modules in the system are used to implement the steps or sub-steps in the above method embodiment, which will not be repeated here.

[0281] To achieve the above objectives, the present invention also provides a prognostic prediction platform for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure, such as... Figure 22 As shown, the system includes a processor, a memory, and a control program for a neoadjuvant-adjuvant-adjuvant-triple-negative breast cancer prognostic prediction platform based on a tertiary lymphoid structure. The processor executes the control program, which is stored in the memory. This control program implements the steps of the neoadjuvant-adjuvant-adjuvant-triple-negative breast cancer prognostic prediction method based on a tertiary lymphoid structure. For example:

[0282] S01. Obtain paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer, and process the corresponding paraffin block of breast cancer tissue into sections based on the paraffin block data of breast cancer tissue.

[0283] S02. Based on the three-level lymphoid structure and combined with the paraffin block data of the breast cancer tissue, first-level data corresponding to TLS density and second-level data corresponding to TLS maturity are generated, and estimated data corresponding to the level data are generated; wherein, the first-level data is a semi-quantitative abundance level; the second-level data is a maturity level.

[0284] S03. Construct a prediction model corresponding to neoadjuvant triple-negative breast cancer, and combine the estimated data to generate prediction data corresponding to the prognosis of neoadjuvant triple-negative breast cancer in real time.

[0285] The specific details of the steps have been explained above and will not be repeated here.

[0286] In this embodiment of the invention, the built-in processor of the neoadjuvant-adjuvant-adjuvant-adjuvant-adjuvant-triple-negative breast cancer prognostic prediction platform based on the three-level lymphatic structure can be composed of integrated circuits. For example, it can be composed of a single packaged integrated circuit, or it can be composed of multiple integrated circuits packaged with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor connects to various components using various interfaces and lines, and executes programs or units stored in memory, and calls data stored in memory, to perform various functions and process data for neoadjuvant- ...

[0287] The memory is used to store program code and various data. It is installed in the neoadjuvant-adjuvant triple-negative breast cancer prognosis prediction platform based on the three-level lymphatic structure and enables high-speed and automatic access to programs or data during operation.

[0288] The memory includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0289] The present invention obtains paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer, and processes the corresponding paraffin block of breast cancer tissue into sections based on the paraffin block data.

[0290] Based on the three-level lymphoid structure and combined with the paraffin block data of the breast cancer tissue, first-level data corresponding to TLS density and second-level data corresponding to TLS maturity are generated, and estimated data corresponding to the level data are generated; wherein, the first-level data is a semi-quantitative abundance level; and the second-level data is a maturity level.

[0291] A predictive model corresponding to neoadjuvant triple-negative breast cancer is constructed, and predictive data corresponding to the prognosis of neoadjuvant triple-negative breast cancer is generated in real time by combining the estimated data; and a system and platform corresponding to the method are also provided, which can accurately evaluate the maturity and spatial positioning characteristics of TLS in the prognostic assessment of triple-negative breast cancer, and at the same time construct a more accurate prognostic predictive model, so as to provide a strong basis for the formulation of personalized treatment plans, thereby promoting the improvement of survival rate and quality of life of triple-negative breast cancer patients.

[0292] In other words, this solution is a prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure. By constructing a prediction model corresponding to neoadjuvant-adjuvant triple-negative breast cancer and combining it with the estimated data, it generates high-precision prediction data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time.

[0293] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A prognostic prediction method for neoadjuvant-adjuvant triple-negative breast cancer based on tertiary lymphoid structure, characterized in that, The method includes the following steps: Obtain paraffin-embedded breast cancer tissue data corresponding to neoadjuvant-adjuvant triple-negative breast cancer. Based on the tertiary lymphoid structure, and combined with the paraffin-embedded breast cancer tissue data, generate first-level data corresponding to the density of TLS in the tertiary lymphoid structure and second-level data corresponding to the maturity of TLS in the tertiary lymphoid structure, and generate estimated data corresponding to the level data; wherein, the first-level data is a semi-quantitative abundance level; the second-level data is a maturity level; construct a prediction model corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and combine the estimated data to generate prediction data corresponding to the prognosis of neoadjuvant-adjuvant triple-negative breast cancer in real time, including the following steps: To obtain sample data corresponding to neoadjuvant-adjuvant triple-negative breast cancer, and to determine and estimate whether there are significant prognostic differences between different sample groups. Each covariate was individually subjected to Cox proportional hazards regression analysis, and data on covariates significantly associated with survival time were generated. A multi-factor Cox model is constructed, and continuous variables are converted into restricted cubic splines and added to the multi-factor Cox model to generate corresponding node count data. The distribution area of ​​tertiary lymphoid structures (TLS) was subdivided into two sub-regions: the near-tumor T region and the far-tumor P region. The near-tumor T region includes invasive carcinoma areas, areas with residual degenerative cancer cells, areas where cancer cells have died or completely disappeared and been replaced by granulation tissue or fibrosis, and areas within 0.5 mm of their edges. The far-tumor P region is the area on the slice that does not meet the above criteria for the near-tumor T region. The abundance and maturity of tertiary lymphoid structures (TLS) in the near-tumor T region and the far-tumor P region were evaluated respectively.

2. The prediction method according to claim 1, characterized in that, The step of obtaining paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer includes: H&E staining and Ki67 immunohistochemical staining were performed on paraffin blocks of breast cancer tissue, and whole-section WSI scanning was performed to generate corresponding KFB format files. Based on the KFB format file, corresponding pathological image data in SVS format is generated; the SVS format whole-slice scan WSI file is processed by region delineation, area measurement, and ROI region cell count calculation.

3. The prediction method according to claim 1, characterized in that, The step of constructing a predictive model corresponding to neoadjuvant triple-negative breast cancer and, in conjunction with the estimated data, generating predictive data corresponding to the prognosis of neoadjuvant triple-negative breast cancer in real time, further includes the following steps: Based on the prediction model corresponding to neoadjuvant triple-negative breast cancer, data on the relationship between tertiary lymphoid structure and disease-free survival and overall survival were generated. Acquire clinicopathological feature data corresponding to neoadjuvant triple-negative breast cancer, and generate correlation data between the clinicopathological feature data and the tertiary lymphoid structure feature data; A correlation is established between the three-level lymphoid structure feature data and prognostic data, and corresponding correction data is generated based on the correlation data; wherein, the correlation data includes linear correlation data and nonlinear correlation data.

4. The prediction method according to claim 3, characterized in that, The step of constructing a predictive model corresponding to neoadjuvant triple-negative breast cancer and, in conjunction with the estimated data, generating predictive data corresponding to the prognosis of neoadjuvant triple-negative breast cancer in real time, further includes the following steps: Based on the three-level lymphoid structure feature data, a disease-free survival (DFS) prediction model for triple-negative breast cancer was constructed, and the performance of each prediction model was compared to generate model performance evaluation index data corresponding to the performance.

5. A prognostic prediction system for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure, characterized in that, The system is applied to the prediction method as described in any one of claims 1-4, and the system comprises: The data acquisition and processing unit is used to acquire paraffin block data of breast cancer tissue corresponding to neoadjuvant triple-negative breast cancer. The first data generation unit is used to generate, based on the tertiary lymphoid structure and combined with the paraffin block data of the breast cancer tissue, first-level data corresponding to the density of TLS in the tertiary lymphoid structure and second-level data corresponding to the maturity of TLS in the tertiary lymphoid structure, and generate estimated data corresponding to the level data; wherein, the first-level data is a semi-quantitative abundance level; and the second-level data is a maturity level. The second data generation unit is used to construct a prediction model corresponding to neoadjuvant triple-negative breast cancer, and combine the estimated data to generate prediction data corresponding to the prognosis of neoadjuvant triple-negative breast cancer in real time. The second data generation unit further includes: The first judgment module is used to acquire sample data corresponding to neoadjuvant triple-negative breast cancer, and to determine and estimate whether there are significant prognostic differences between different groups of samples. The first data generation module is used to perform Cox proportional hazards regression analysis on each covariate individually and generate covariate data that are significantly associated with survival time. The second data generation module is used to construct a multi-factor Cox model, and at the same time convert continuous variables into restricted cubic splines and add them into the multi-factor Cox model to generate corresponding node number data. The distribution area of ​​tertiary lymphoid structures (TLS) was subdivided into two sub-regions: the near-tumor T region and the far-tumor P region. The near-tumor T region includes invasive carcinoma areas, areas with residual degenerative cancer cells, areas where cancer cells have died or completely disappeared and been replaced by granulation tissue or fibrosis, and areas within 0.5 mm of their edges. The far-tumor P region is the area on the slice that does not meet the above criteria for the near-tumor T region. The abundance and maturity of tertiary lymphoid structures (TLS) in the near-tumor T region and the far-tumor P region were evaluated respectively.

6. The prediction system according to claim 5, characterized in that, The data acquisition and processing unit further includes: The first processing module performs H&E staining and Ki67 immunohistochemical staining on paraffin blocks of breast cancer tissue, performs whole-slice scanning (WSI), and generates corresponding KFB format files. The second processing module is used to convert the KFB format file into corresponding pathological image data in SVS format. The third processing module is used to perform region delineation, area measurement, and ROI region cell number calculation on the SVS format whole-slice scan WSI file, respectively.

7. The prediction system according to claim 5, characterized in that, The second data generation unit further includes: The third data generation module is used to generate data on the relationship between the three-level lymphoid structure and disease-free survival and overall survival based on the prediction model corresponding to neoadjuvant triple-negative breast cancer. The fourth data generation module is used to acquire clinicopathological feature data corresponding to neoadjuvant triple-negative breast cancer, and generate correlation data between the clinicopathological feature data and the tertiary lymphoid structure feature data. The fifth data generation module is used to create correlation data between the three-level lymphoid structure feature data and prognostic data, and to generate corresponding correction data based on the correlation data; wherein, the correlation data includes linear correlation data and nonlinear correlation data; The sixth data generation module is used to construct a disease-free survival (DFS) prediction model for triple-negative breast cancer based on the three-level lymphoid structure feature data, compare the performance of each prediction model, and generate model performance evaluation index data corresponding to the performance.

8. A prognostic prediction platform for neoadjuvant-adjuvant triple-negative breast cancer based on a three-tiered lymphatic structure, characterized in that, The prediction platform includes a processor, a memory, and a prediction platform control program; wherein the prediction platform control program is stored in the memory, and when the processor executes the prediction platform control program, it implements the prediction method as described in any one of claims 1 to 4.

Images