Rectum cancer prognosis model construction method and system based on deep learning

Abstract

The invention relates to the technical field of biomedicine, and discloses a rectal cancer prognosis model construction method based on deep learning, and the method comprises the following steps: identifying disease characteristics of a rectal cancer patient, and defining prognosis requirements of the rectal cancer patient according to a baseline standard of the disease characteristics; determining focus types and iconography characteristics of the rectal cancer patients, and analyzing key prognosis indexes of the rectal cancer patients in combination with the focus types and the iconography characteristics; analyzing prognosis influence factors of the rectal cancer patient, and extracting characteristic data of the medical data; key prognosis features of the rectal cancer prognosis features are extracted, and derived features are determined; and setting a survival analysis result of the rectal cancer patient, and constructing a prognosis model of the rectal cancer patient in combination with the key prognosis characteristics, the derived characteristics and the survival analysis result. The use efficiency and generalization ability of the prognosis model can be improved.

Description

Technical Field

[0001] The present invention relates to the field of biomedical technology, and in particular to a method and system for constructing a rectal cancer prognosis model based on deep learning. Background Art

[0002] The construction of a prognostic model for rectal cancer refers to the use of statistics, machine learning and other methods, combined with clinical data and biological information, to develop a mathematical model that can predict the survival status and disease development of rectal cancer patients after treatment, so as to help doctors assess the patient's risk of disease progression and survival. Rectal cancer is one of the common malignant tumors, and its incidence and mortality rate account for a significant proportion worldwide. With the advancement of medical technology, most early-stage rectal cancer patients can be effectively treated through surgical resection combined with adjuvant radiotherapy and chemotherapy. However, due to the biological characteristics of rectal cancer and the diversity of clinical manifestations, the prognosis of different patients varies greatly. Therefore, it is particularly important to build a model that can accurately predict the prognosis of patients.

[0003] When building existing machine learning models, features need to be manually extracted based on experts' experience and domain knowledge, which leads to human errors in feature extraction and fails to fully capture and integrate the complex interactions between imaging, clinical and genomic data, limiting the effectiveness and generalization ability of prognostic models. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a method and system for constructing a rectal cancer prognosis model based on deep learning, which can improve the use efficiency and generalization ability of the prognosis model.

[0005] In a first aspect, the present invention provides a method for constructing a rectal cancer prognosis model based on deep learning, comprising:

[0006] Acquire a rectal cancer patient to be diagnosed, and collect medical data of the rectal cancer patient, identify disease characteristics of the rectal cancer patient based on the medical data, set baseline standards for the disease characteristics, and define prognostic requirements for the rectal cancer patient based on the baseline standards;

[0007] Determine the lesion type of the rectal cancer patient according to the disease characteristics, identify the imaging features corresponding to the lesion type, and analyze the key prognostic indicators of the rectal cancer patient in combination with the lesion type and the imaging features;

[0008] Based on the key prognostic indicators and the prognostic requirements, the prognostic influencing factors of the rectal cancer patients are analyzed, the key influencing factors of the prognostic influencing factors are identified, characteristic data of the medical data is extracted according to the key influencing factors, and the characteristic data is cleaned to obtain target data;

[0009] Extracting the rectal cancer prognostic features of the target data according to the key prognostic indicators, performing dimensionality reduction processing on the rectal cancer prognostic features to obtain key prognostic features, and determining derived features of the key prognostic features based on the disease characteristics;

[0010] Based on the key prognostic indicators, a survival analysis is performed on the rectal cancer patients to obtain survival analysis results, and a prognostic model for the rectal cancer patients is constructed by combining the key prognostic features, the imaging features and the survival analysis results.

[0011] In a second aspect, the present invention provides a rectal cancer prognosis model construction system based on deep learning, characterized in that the system comprises:

[0012] A demand analysis module, used for acquiring a rectal cancer patient to be diagnosed, collecting medical data of the rectal cancer patient, identifying disease characteristics of the rectal cancer patient based on the medical data, setting a baseline standard of the disease characteristics, and defining the prognosis needs of the rectal cancer patient based on the baseline standard;

[0013] A prognostic indicator identification module, used to determine the lesion type of the rectal cancer patient according to the disease characteristics, identify the imaging features corresponding to the lesion type, and analyze the key prognostic indicators of the rectal cancer patient in combination with the lesion type and the imaging features;

[0014] A data processing module, used for analyzing the prognostic influencing factors of the rectal cancer patient based on the key prognostic indicators and the prognostic requirements, identifying the key influencing factors of the prognostic influencing factors, extracting the characteristic data of the medical data according to the key influencing factors, and cleaning the characteristic data to obtain target data;

[0015] A feature extraction module, used to extract the rectal cancer prognostic features of the target data according to the key prognostic indicators, perform dimensionality reduction processing on the rectal cancer prognostic features to obtain key prognostic features, and determine derived features of the key prognostic features based on the disease characteristics;

[0016] The model building module is used to perform a survival analysis on the rectal cancer patients based on the key prognostic indicators to obtain survival analysis results, and to build a prognostic model for the rectal cancer patients by combining the key prognostic features, the imaging features and the survival analysis results.

[0017] Compared with the prior art, the technical principle and beneficial effects of this solution are:

[0018] The embodiment of the present invention can define the prognostic needs of the rectal cancer patient by identifying the disease characteristics of the rectal cancer patient and setting the baseline standard of the disease characteristics, so as to help the doctor to more comprehensively understand the disease status and treatment needs of the patient, thereby formulating a more accurate treatment plan and predicting possible treatment effects and disease progression; secondly, the embodiment of the present invention can determine the lesion type of the rectal cancer patient according to the disease characteristics, identify the imaging characteristics corresponding to the lesion type, and analyze the key prognostic indicators of the rectal cancer patient in combination with the lesion type and the imaging characteristics, which is helpful to capture the morphological diversity of rectal cancer and provide more accurate pathological information for the prognostic model, thereby improving the accuracy of the prognostic evaluation; thirdly, the embodiment of the present invention can identify the key influencing factors of the prognostic influencing factors by analyzing the prognostic influencing factors of the rectal cancer patient based on the key prognostic indicators and the prognostic needs, thereby identifying the patient's medical prognostic features in treatment data; further, the embodiment of the present invention can obtain the key prognostic features and derived features of rectal cancer patients by extracting the rectal cancer prognostic features of the target data according to the key prognostic indicators, helping the model to capture the deeper relationship and pattern between the key prognostic features, thereby improving the prediction accuracy of the model; finally, the embodiment of the present invention performs survival analysis on the rectal cancer patients based on the key prognostic indicators to obtain the survival analysis results, and combines the key prognostic features, the imaging features and the survival analysis results to construct the prognostic model of the rectal cancer patients. The fusion features can help the prognostic model capture the complex nonlinear relationship or interaction effect between the original imaging features of the patients, improve the prediction accuracy and generalization ability of the prognostic model, thereby accurately assessing the risk of recurrence, metastasis or survival of the patients, assisting doctors and patients to make more informed treatment choices, and improving the survival rate and quality of life of the patients. Therefore, the rectal cancer prognostic model construction method based on deep learning proposed by the present invention can improve the use efficiency and generalization ability of the prognostic model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0021] Figure 1 A schematic flow chart of a method for constructing a rectal cancer prognosis model based on deep learning provided in one embodiment of the present invention;

[0022] Figure 2 A schematic diagram of modules of a deep learning-based rectal cancer prognosis model construction system provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0023] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0024] The embodiment of the present invention provides a method for constructing a prognostic model for rectal cancer based on deep learning, and the execution subject of the method for constructing a prognostic model for rectal cancer based on deep learning includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present invention. In other words, the method for constructing a prognostic model for rectal cancer based on deep learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (Content De li very Network, CDN), and big data and artificial intelligence platforms.

[0025] See also Figure 1 FIG. 1 is a flow chart of a method for constructing a rectal cancer prognosis model based on deep learning according to an embodiment of the present invention. Figure 1 The deep learning-based rectal cancer prognostic model construction method described in the paper includes:

[0026] S1. Acquire a rectal cancer patient to be diagnosed, and collect medical data of the rectal cancer patient. According to the medical data, identify the disease characteristics of the rectal cancer patient, set a baseline standard for the disease characteristics, and define the prognosis requirements of the rectal cancer patient based on the baseline standard.

[0027] The embodiment of the present invention can provide data support for the subsequent construction of a prognosis model by acquiring rectal cancer patients to be prognosed and collecting medical data of the rectal cancer patients. The rectal cancer patients refer to patients who have been diagnosed with rectal cancer but have not yet received systemic treatment or who need further evaluation of treatment effects and prediction of future disease progression after treatment. The medical data refers to all medical information and records related to the rectal cancer patients, such as basic patient information, diagnostic reports, and laboratory test results.

[0028] Optionally, the medical data collection of the rectal cancer patient can be obtained through an electronic health record system.

[0029] Furthermore, the embodiment of the present invention can make the identification of disease characteristics more objective and accurate by identifying the disease characteristics of the rectal cancer patient based on the medical data, reduce the deviation of subjective judgment, and comprehensively reflect the patient's disease condition. The disease characteristics refer to the attribute characteristics that can describe and define the rectal cancer disease, such as clinical symptoms, imaging characteristics and tumor staging.

[0030] As an embodiment of the present invention, identifying the disease characteristics of the rectal cancer patient based on the medical data includes: identifying the preoperative evaluation data and postoperative pathology report of the rectal cancer patient based on the medical data; analyzing the preoperative clinical symptoms of the rectal cancer patient based on the preoperative evaluation data; scheduling the imaging data of the rectal cancer patient based on the preoperative clinical symptoms; analyzing the tumor characteristics of the rectal cancer patient based on the imaging data; identifying the postoperative physical condition of the rectal cancer patient based on the postoperative pathology report; and identifying the disease characteristics of the rectal cancer patient in combination with the tumor characteristics and the postoperative physical condition.

[0031] Among them, the preoperative evaluation data refers to the data obtained from a series of examinations and evaluations of the patient before the operation, including the patient's health status, surgical risks, physiological and psychological conditions, etc. The postoperative pathology report refers to the report issued after the pathological examination of the resected tissue samples after the operation. The preoperative clinical symptoms refer to various disease-related symptoms exhibited by the patient before the operation, such as pain, bleeding, obstruction, etc. The imaging data refers to the images and data about the patient's internal structures and lesions obtained through various imaging examinations (such as X-rays, CT, MRI, PET-CT, etc.). The tumor characteristics refer to a series of biological and pathological characteristics of the tumor, including the size, growth rate, invasiveness, etc. of the tumor. The postoperative physical condition refers to the patient's physical condition after the tissue is removed.

[0032] The embodiment of the present invention sets a baseline standard for the disease characteristics, and can determine the prognosis requirements of rectal cancer patients through the baseline standard, thereby providing a data learning range for the deep learning model. The baseline standard refers to the normalization level of the disease characteristics of rectal cancer patients.

[0033] Optionally, the baseline standard setting of the disease characteristics can be achieved by using personalized pathological data of rectal cancer patients, such as clinical diagnosis results, pathological examination results and general conditions of patients.

[0034] Furthermore, the embodiments of the present invention can help doctors to more comprehensively understand the disease status and treatment needs of patients with rectal cancer by defining the prognostic needs of the patients based on the baseline standards, thereby formulating more accurate treatment plans and predicting possible treatment effects and disease progression. The prognostic needs refer to the predictions and expectations of doctors, patients and their families regarding the future development of the disease, treatment effects, survival period and quality of life in medical practice.

[0035] As an embodiment of the present invention, the prognosis needs of the rectal cancer patients are defined based on the baseline standard, including: identifying the disease severity of the rectal cancer patients based on the baseline standard; setting the disease treatment method for the rectal cancer patients according to the disease severity; identifying the patient disease characteristics of the baseline standard; analyzing the survival period of the rectal cancer patients and the tumor circumferential margin status of the rectal cancer patients based on the patient disease characteristics and the disease severity; analyzing the disease recurrence risk of the rectal cancer patients based on the tumor circumferential margin status; and defining the prognosis needs of the rectal cancer patients in combination with the disease severity, the survival period and the disease recurrence risk.

[0036] Among them, the disease severity refers to the degree of progression of rectal cancer, including tumor size (T), lymph node involvement (N) and whether there is distant metastasis (M); the disease treatment method refers to the treatment plan determined according to the severity of the disease; the patient's disease characteristics refer to other disease-related characteristics other than TNM staging, such as tumor grade, biological marker level, patient age, gender, underlying diseases, etc.; the survival period refers to the length of time a patient is expected to survive from the start of a specific diagnosis or treatment; the tumor circumferential margin status refers to the distance between the deepest point of tumor infiltration and the radial margin of the soft tissue around the rectum; the disease recurrence risk refers to the possibility of the disease recurring after treatment.

[0037] Optionally, the identification of the disease severity of the rectal cancer patient based on the baseline standard can be achieved using the rectal cancer DI STANCE assessment program.

[0038] In an optional embodiment of the present invention, based on the patient's disease characteristics and the severity of the disease, the following formula is used to analyze the survival period of the rectal cancer patient:

[0039]

[0040] Among them, U(t) represents the survival period of patients with rectal cancer, t j represents the time point when disease event j occurs in patients with rectal cancer, b j represents the number of events of disease event j in the group corresponding to rectal cancer patients, wj represents the weight of the patient's disease characteristics and disease severity on disease event j, m j Indicates that at time t j The number of groups corresponding to previous rectal cancer patients.

[0041] In another optional embodiment of the present invention, based on the patient's disease characteristics and the severity of the disease, the following evaluation formula is used to analyze the tumor circumferential margin status of the rectal cancer patient:

[0042]

[0043] Among them, T represents the circumferential resection margin status of the tumor in patients with rectal cancer. T=0 means that the circumferential resection margin status of the tumor is negative when the distance between the tumor and the circumferential resection margin is greater than 1 mm, and T=1 means that the circumferential resection margin status of the tumor is positive when the distance between the tumor and the circumferential resection margin is less than or equal to 1 mm.

[0044] It should be noted that T = 0 means that the resection margin is more than 1 mm away from the tumor edge, and the risk of disease recurrence in rectal cancer patients is low in this state. T = 1 means that the tumor is very close to or in contact with the resection margin, and the risk of disease recurrence in rectal cancer patients is high in this state.

[0045] S2. Determine the lesion type of the rectal cancer patient based on the disease characteristics, identify the imaging features corresponding to the lesion type, and analyze the key prognostic indicators of the rectal cancer patient in combination with the lesion type and the imaging features.

[0046] In an embodiment of the present invention, by determining the lesion type of the rectal cancer patient according to the disease characteristics, the prognostic factors of the rectal cancer patient can be determined through the biological behavior, staging, pathological response and molecular characteristics of the tumor. The lesion type refers to the classification of the morphology, properties and characteristics of the lesion site in medical imaging, pathology and clinical diagnosis.

[0047] Optionally, according to the disease characteristics, the lesion type of the rectal cancer patient can be determined by histopathological examination.

[0048] Furthermore, the embodiments of the present invention can help doctors assess patients' risk of recurrence by identifying the imaging features corresponding to the lesion types, thereby providing patients with more accurate prognosis assessments and treatment plans. The imaging features refer to specific manifestations of rectal cancer observed through imaging techniques (such as X-rays, CT, MRI, ultrasound, etc.), such as tumor signal characteristics, surrounding tissue invasion characteristics, lymph node metastasis characteristics, etc.

[0049] Optionally, the imaging feature recognition corresponding to the lesion type can be implemented using a convolutional neural network, such as converting low-level features into high-level features through multi-layer feature extraction of the convolutional neural network.

[0050] The embodiment of the present invention combines the lesion type and the imaging features to analyze the key prognostic indicators of the rectal cancer patients, which helps to capture the morphological diversity of rectal cancer and provide more accurate pathological information for the prognostic model, thereby improving the accuracy of the prognostic evaluation. The key prognostic indicators refer to important parameters that can predict or evaluate the patient's disease progression, treatment effect and final outcome.

[0051] As an embodiment of the present invention, the combination of the lesion type and the imaging features to analyze the key prognostic indicators of the rectal cancer patient includes collecting genomic expression factors of the rectal cancer patient based on the lesion type; analyzing the correlation between the imaging features and the genomic expression factors; identifying the imaging feature combination of the genomic expression factors based on the correlation; identifying the pathological stage of the rectal cancer patient based on the imaging feature combination; analyzing the pathological response of the rectal cancer patient based on the pathological stage; identifying the survival rate and recurrence risk ratio of the rectal cancer patient based on the pathological stage and the pathological response; and analyzing the key prognostic indicators of the rectal cancer patient based on the survival rate and the recurrence risk ratio.

[0052] Among them, the genomic expression factor refers to the measured value of the gene expression level in the tumor tissue of rectal cancer patients, the correlation degree refers to the statistical correlation strength between the imaging features and the genomic expression factors, the imaging feature combination refers to the set of imaging features that are significantly correlated with the genomic expression factors selected based on the correlation degree analysis results, the pathological stage refers to the clinical stage of rectal cancer determined according to the size, depth, lymph node involvement and whether there is distant metastasis of the tumor, the pathological response refers to the effect of pathological stage on patient treatment and prognosis, the survival rate refers to the proportion of patients who are still alive within 5 years after diagnosis, and the recurrence risk ratio refers to the possibility of recurrence of patients after treatment.

[0053] Optionally, based on the lesion type, the collection of genomic expression factors of the rectal cancer patient can be achieved using high-throughput sequencing technology, based on the pathological staging, the pathological response analysis of the rectal cancer patient can be obtained through tumor regression grading, based on the pathological staging and the pathological response, the survival rate identification of the rectal cancer patient can be determined using the Kaplan-Meier survival curve, and the recurrence risk ratio identification of the rectal cancer patient can be obtained through a Log-rank test.

[0054] S3. Based on the key prognostic indicators and the prognostic requirements, the prognostic factors of the rectal cancer patients are analyzed, the key prognostic factors are identified, and according to the key prognostic factors, the characteristic data of the medical data is extracted, and the characteristic data is cleaned to obtain the target data.

[0055] The embodiment of the present invention analyzes the prognostic influencing factors of the rectal cancer patients based on the key prognostic indicators and the prognostic requirements, so as to understand the impact mechanism of the prognostic influencing factors on the patient's survival rate and treatment effect, and help doctors formulate personalized treatment plans to improve the patient's survival rate and quality of life. The prognostic influencing factors refer to various factors that can affect the development of the disease and the patient's final outcome, such as tumor location, tumor stage, patient's psychological state, etc.

[0056] As an embodiment of the present invention, the method of analyzing the factors affecting the prognosis of the rectal cancer patient based on the key prognostic indicators and the prognostic requirements includes: analyzing the pathological remission status of the rectal cancer patient based on the key prognostic indicators and the prognostic requirements; identifying the complete remission status in the pathological remission status; identifying the degree of tumor regression of the rectal cancer patient based on the complete remission status; analyzing the adverse factors affecting the complete remission status; analyzing the tumor resection effect of the rectal cancer patient based on the adverse factors; and analyzing the factors affecting the prognosis of the rectal cancer patient in combination with the pathological remission status, the tumor regression degree and the tumor resection effect.

[0057] Among them, the pathological remission state refers to the pathological response state of the tumor in patients with rectal cancer after receiving neoadjuvant therapy (such as chemoradiotherapy), the complete remission state refers to the state in which no tumor cells are found to remain in the radically resected tumor specimen under a microscope, the tumor regression degree refers to the response degree of the tumor to neoadjuvant therapy, including the degree of no response, partial response and complete response, the adverse influencing factors refer to the threatening factors that affect the complete resection of the tumor in patients with rectal cancer, including the maximum diameter of the postoperative scar / lesion, the lymph node metastasis status and the distance between the lower edge of the tumor and the anal verge, and the tumor resection effect refers to the thoroughness of the surgical resection of the tumor, including the status of the circumferential resection margin (CRM) and the distal resection margin.

[0058] Optionally, based on the key prognostic indicators and the prognostic requirements, the analysis of the pathological remission status of the rectal cancer patient can be obtained by the degree of tumor regression after neoadjuvant chemoradiotherapy (NCRT), according to the complete remission status, the identification of the tumor regression degree of the rectal cancer patient can be achieved using the ROC curve, and based on the adverse influencing factors, the analysis of the tumor resection effect of the rectal cancer patient can be determined by the status of the circumferential resection margin (CRM) and the distal resection margin.

[0059] Furthermore, the embodiments of the present invention can help predict the patient's response to a specific treatment by identifying the key influencing factors of the prognostic influencing factors, so as to more accurately evaluate the patient's prognosis. The key influencing factors refer to factors that play a decisive role in the treatment and prognostic evaluation of rectal cancer patients, such as the degree of tumor regression.

[0060] Optionally, the identification of key influencing factors of the prognostic factors can be obtained through a Cox proportional hazard model.

[0061] The embodiment of the present invention can help identify the prognostic characteristics of rectal cancer patients by extracting characteristic data of the medical data based on the key influencing factors. The characteristic data refers to specific data that can quantify or describe information such as the patient's condition, disease characteristics, and treatment response.

[0062] Optionally, the feature data extraction of the medical data based on the key influencing factors can be achieved by using a multivariate regression analysis method.

[0063] Furthermore, the embodiment of the present invention obtains target data by cleaning the feature data, and can identify missing values, erroneous values ​​and duplicates in the feature data, ensure that the relationship between the data is reasonable, and improve the efficiency and accuracy of data processing. The target data refers to a data set that has been cleaned, optimized and prepared for further analysis after a series of data preprocessing steps.

[0064] Optionally, the cleaning process of the feature data can be implemented using the Melissa Clean Suite data processing tool.

[0065] S4. According to the key prognostic indicators, the rectal cancer prognostic features of the target data are extracted, and the rectal cancer prognostic features are subjected to dimensionality reduction processing to obtain key prognostic features, and based on the disease characteristics, derived features of the key prognostic features are determined.

[0066] The embodiment of the present invention can help doctors divide patients into different risk groups by extracting the rectal cancer prognostic features of the target data according to the key prognostic indicators, so as to improve the quality of life of patients and optimize medical resources. The rectal cancer prognostic features refer to clinical pathological characteristics and biological markers that can predict the disease development and outcome (such as survival period, recurrence risk, etc.) of rectal cancer patients.

[0067] As an embodiment of the present invention, the method of extracting the rectal cancer prognostic features of the target data according to the key prognostic indicators includes: identifying the data item attributes of the target data according to the key prognostic indicators; extracting the patient group characteristics of the key prognostic indicators based on the data item attributes; classifying the patient group characteristics to obtain classification features; calculating the attribute correlation degree between the classification features and the data item attributes; analyzing the importance of the classification features to the key prognostic indicators; and extracting the rectal cancer prognostic features of the target data based on the attribute correlation degree and the importance.

[0068] Among them, the data item attributes refer to the characteristics of each variable or field in the data set, such as age, gender, tumor size, pathological stage, etc. The patient group characteristics refer to the statistical characteristics or patterns extracted from the data item attributes that can describe a specific patient group, such as the average age, gender ratio, tumor stage distribution, etc. of rectal cancer patients. The classification feature refers to the conversion of continuous data item attributes into different categories. The attribute correlation degree refers to the correlation or dependency between the feature items of different classification features and the data item attributes. The importance refers to the influence or contribution of the classification feature to the prediction of key prognostic indicators.

[0069] Optionally, the extraction of patient population features of the key prognostic indicators based on the data item attributes can be achieved using a receiver operating characteristic (ROC) curve evaluation model, and the importance analysis of the classification features to the key prognostic indicators can be determined by a feature importance score of a machine learning model such as a random forest.

[0070] In an optional embodiment of the present invention, the attribute correlation degree between the classification feature and the data item attribute is calculated using the following formula:

[0071]

[0072] Among them, R represents the attribute correlation between the classification feature and the data item attribute, and D p represents the attribute value of the p-th classification feature, C p represents the attribute value of the data item attribute corresponding to the p-th classification feature, max(D p ,C p ) represents the maximum value of the attribute value of the p-th classification feature and its corresponding data item attribute, p represents the serial number of the classification feature, and n represents the number of identical attributes of the classification feature and its corresponding data item attribute.

[0073] The embodiment of the present invention performs dimensionality reduction processing on the rectal cancer prognostic features to obtain key prognostic features, and can screen out the key features that have the greatest impact on the prognosis of rectal cancer, so that the prognostic model can more accurately predict the patient's prognosis. The key prognostic features refer to factors that have a significant impact on the treatment effect and survival of rectal cancer patients, such as molecular marker characteristics.

[0074] As an embodiment of the present invention, the dimensionality reduction processing of the rectal cancer prognostic features to obtain key prognostic features includes: collecting prognostic feature data of the rectal cancer prognostic features; analyzing the distribution of the rectal cancer prognostic features and feature correlations based on the prognostic feature data; identifying data characteristics of the prognostic feature data based on the distribution; determining a dimensionality reduction method for the rectal cancer prognostic features based on the feature correlations; and performing dimensionality reduction processing on the rectal cancer prognostic features in combination with the data characteristics and the dimensionality reduction method to obtain key prognostic features.

[0075] Among them, the prognostic feature data refers to various data related to disease prognosis collected from rectal cancer patients, including clinical parameters (such as age, gender, tumor size, stage), genomic data, biomarker levels, imaging features, etc. The distribution situation refers to the statistical distribution characteristics of the prognostic feature data, such as whether it presents a normal distribution, a skewed distribution, whether there are outliers or outliers, etc. The feature relationship refers to the correlation or dependency between different features in the prognostic feature data, such as linear relationships, nonlinear relationships, interactions, etc. between features. The data characteristics refer to the intrinsic properties of the prognostic feature data, such as the scale (dimension) of the data, the type of data (continuous, categorical), the sparsity or density of the data, etc. The dimensionality reduction method refers to the technology or method used to reduce the dimension of the prognostic feature data, such as principal component analysis (PCA), linear discriminant analysis (LDA), t-SNE, autoencoder, etc.

[0076] Optionally, the distribution analysis of the rectal cancer prognostic features based on the prognostic feature data can be obtained through a histogram, the feature relationship analysis of the rectal cancer prognostic features based on the prognostic feature data can be achieved using the Pearson correlation coefficient, and the data characteristic identification of the prognostic feature data based on the distribution can be determined by cluster analysis.

[0077] Furthermore, the embodiments of the present invention can help the model capture deeper relationships and patterns between key prognostic features based on the disease characteristics, thereby improving the prediction accuracy of the model. The derived features refer to new features obtained by performing mathematical operations, statistical analysis or other forms of processing on one or more original features, such as interaction terms (the combined effect of different gene activities in patients with rectal cancer), time series features (changes in physiological indicators before and after treatment), survival time, etc.

[0078] As an embodiment of the present invention, determining the derived features of the key prognostic features based on the disease features includes: scheduling patient tumor data corresponding to the key prognostic features based on the disease features; extracting the tumor volume, tumor markers and gene activity of the patient tumor data; analyzing the concentration level of the tumor marker, and identifying the treatment response speed and treatment effect of the tumor cells based on the concentration level; analyzing the growth pattern of the tumor cells based on the tumor volume; calculating the risk score of the tumor cells based on the gene activity; and determining the derived features of the key prognostic features in combination with the treatment response speed, the treatment effect, the growth pattern and the risk score.

[0079] Among them, the patient tumor data refers to information data related to the tumor diagnosed in the patient, such as tumor size, the tumor volume refers to the size of the space occupied by the tumor in the body, the tumor marker refers to a substance related to tumor growth and existence that can be detected in the blood, such as carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9), the gene activity refers to the expression level of a specific gene in tumor cells, the concentration level refers to the concentration of the tumor marker in the blood, the treatment response speed refers to the response speed of the tumor to treatment (such as chemotherapy, radiotherapy), the treatment effect refers to the inhibitory effect of treatment on tumor growth, the growth pattern refers to the rate and method of tumor growth, and the risk score refers to a score calculated based on the activity of multiple genes.

[0080] Optionally, the concentration level analysis of the tumor marker can be obtained through laboratory testing, and the identification of the treatment response speed of the tumor cells based on the concentration level can be achieved by evaluating the rate of decrease of the tumor marker concentration. The growth pattern analysis of the tumor cells based on the tumor volume can be determined by continuously measuring the changes in tumor volume.

[0081] In an optional embodiment of the present invention, the risk score of the tumor cell is calculated according to the gene activity using the following formula:

[0082]

[0083] Among them, K represents the risk score of tumor cells, W f represents the contribution of the f-th gene to the risk score, X f represents the gene activity of the f-th gene, g represents the number of genes, and f represents the sequence number of the gene.

[0084] S5. Based on the key prognostic indicators, a survival analysis is performed on the rectal cancer patients to obtain survival analysis results, and a prognostic model for the rectal cancer patients is constructed by combining the key prognostic features, the imaging features and the survival analysis results.

[0085] The embodiment of the present invention performs survival analysis on the rectal cancer patients based on the key prognostic indicators to obtain survival analysis results, which can predict the patient's long-term survival rate and the possibility of disease recurrence, and help doctors choose the most appropriate treatment plan. The survival analysis results refer to specific indicators or events for evaluating the patient's treatment effect and predicting future health status, including disease-free survival, overall survival, disease recurrence rate, metastasis rate, etc.

[0086] As an embodiment of the present invention, the survival analysis of the rectal cancer patients is performed based on the key prognostic indicators to obtain survival analysis results, including: based on the key prognostic indicators, the rectal cancer patients are risk stratified to obtain patient risk grading; according to the patient risk grading, the genomic data and imaging feature data of the rectal cancer patients are collected; based on the genomic data and the imaging feature data, the rectal cancer patients are scored for recurrence risk to obtain a recurrence risk score; according to the recurrence risk score, the survival recurrence events and follow-up time of the rectal cancer patients are analyzed; according to the survival recurrence events and the follow-up time, the survival influencing indicators of the rectal cancer patients are identified; in combination with the survival recurrence events, the recurrence risk score and the survival influencing indicators, the rectal cancer patients are subjected to survival analysis to obtain survival analysis results.

[0087] Among them, the patient risk grading refers to dividing patients into different risk groups based on their key prognostic indicators (such as tumor size, stage, patient age, gene expression pattern, etc.); the genomic data refers to data on the activity of specific genes collected from tumor samples of rectal cancer patients; the imaging feature data refers to a series of quantitative and qualitative information about rectal cancer obtained through imaging examinations; the survival recurrence event refers to the reappearance or recurrence of the disease after a period of disease-free survival or disease stability; the follow-up time refers to the time period from the completion of the initial treatment (such as surgery, radiotherapy, chemotherapy, etc.) of rectal cancer patients to regular examinations and evaluations; the survival influencing indicators refer to various factors that can affect the patient's survival, including disease recurrence and metastasis, patient functional status, changes in the levels of biomarkers, etc.

[0088] Optionally, the analysis of the survival recurrence events of the colorectal cancer patients based on the genomic data can be determined by multivariate analysis, and the identification of indicators affecting the survival of the colorectal cancer patients based on the survival recurrence events can be achieved by utilizing the changing patterns of the patients' biomarkers and functional status.

[0089] In an optional embodiment of the present invention, combining the survival recurrence event, the recurrence risk score and the survival impact index, the following formula is used to perform survival analysis on the rectal cancer patient to obtain a survival analysis result:

[0090]

[0091] Among them, Loss represents the survival analysis results of rectal cancer patients, N represents the total number of rectal cancer patients, and v r represents the weight coefficient of the recurrence risk score, r i represents the risk score of the i-th rectal cancer patient, v e represents the weight coefficient of the recurrence event during survival, e i The indicator variable indicating whether the i-th rectal cancer patient has a recurrence event during survival, w a represents the weight coefficient of follow-up time, a i The follow-up time of the i-th rectal cancer patient, k represents the total number of recurrence risk score, survival recurrence events and follow-up time, h represents the prediction category index of the prediction model corresponding to the survival analysis result, s ih It represents the prediction score of the prediction model corresponding to the survival analysis result for the hth category of the i-th rectal cancer patient.

[0092] Furthermore, the embodiment of the present invention constructs a prognostic model for the rectal cancer patient by combining the key prognostic features, the imaging features and the survival analysis results. The fusion features can help the prognostic model capture the complex nonlinear relationship or interaction effect between the original imaging features of the patient, thereby improving the prediction accuracy and generalization ability of the prognostic model, thereby accurately assessing the risk of recurrence, metastasis or survival of the patient, assisting doctors and patients to make more informed treatment choices, and improving the survival rate and quality of life of the patient. The prognostic model refers to a statistical or machine learning model, such as a deep learning model, used to predict the future health status of rectal cancer patients.

[0093] It can be seen that the embodiment of the present invention can define the prognostic needs of the rectal cancer patient by identifying the disease characteristics of the rectal cancer patient and setting the baseline standards of the disease characteristics, thereby helping doctors to more comprehensively understand the patient's disease status and treatment needs, thereby formulating a more accurate treatment plan and predicting possible treatment effects and disease progression; secondly, the embodiment of the present invention determines the lesion type of the rectal cancer patient according to the disease characteristics, identifies the imaging features corresponding to the lesion type, and can analyze the key prognostic indicators of the rectal cancer patient in combination with the lesion type and the imaging features, which helps to capture the morphological diversity of rectal cancer and provide more accurate pathological information for the prognostic model, thereby improving the accuracy of the prognostic evaluation; thirdly, the embodiment of the present invention analyzes the prognostic influencing factors of the rectal cancer patient based on the key prognostic indicators and the prognostic needs, thereby identifying the key influencing factors of the prognostic influencing factors, thereby identifying the key prognostic influencing factors. Prognostic features in the patient's medical data; further, the embodiment of the present invention can obtain the key prognostic features and derived features of the rectal cancer patient by extracting the rectal cancer prognostic features of the target data according to the key prognostic indicators, so as to help the model capture the deeper relationship and pattern between the key prognostic features, thereby improving the prediction accuracy of the model; finally, the embodiment of the present invention performs survival analysis on the rectal cancer patient based on the key prognostic indicators to obtain the survival analysis results, and combines the key prognostic features, the imaging features and the survival analysis results to construct the prognostic model of the rectal cancer patient. The fusion features can help the prognostic model capture the complex nonlinear relationship or interaction effect between the original imaging features of the patient, improve the prediction accuracy and generalization ability of the prognostic model, thereby accurately assessing the risk of recurrence, metastasis or survival of the patient, assisting doctors and patients to make more informed treatment choices, and improving the survival rate and quality of life of the patient. Therefore, the rectal cancer prognostic model construction method based on deep learning proposed by the present invention can improve the use efficiency and generalization ability of the prognostic model.

[0094] like Figure 2 Shown is a system function module diagram of the rectal cancer prognosis model constructed based on deep learning in the present invention.

[0095] The rectal cancer prognosis model construction system 200 based on deep learning of the present invention can be installed in an electronic device. According to the functions implemented, the rectal cancer prognosis model construction system based on deep learning can include a demand analysis module 201, a prognosis indicator identification module 202, a data processing module 203, a feature extraction module 204, and a model construction module 205. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0096] In the embodiment of the present invention, the functions of each module / unit are as follows:

[0097] The demand analysis module 201 is used to obtain a rectal cancer patient to be diagnosed, collect medical data of the rectal cancer patient, identify disease characteristics of the rectal cancer patient according to the medical data, set a baseline standard of the disease characteristics, and define the prognosis needs of the rectal cancer patient based on the baseline standard;

[0098] The prognostic indicator identification module 202 is used to determine the lesion type of the rectal cancer patient according to the disease characteristics, identify the imaging features corresponding to the lesion type, and analyze the key prognostic indicators of the rectal cancer patient in combination with the lesion type and the imaging features;

[0099] The data processing module 203 is used to analyze the prognostic influencing factors of the rectal cancer patient based on the key prognostic indicators and the prognostic requirements, identify the key influencing factors of the prognostic influencing factors, extract the characteristic data of the medical data according to the key influencing factors, and clean the characteristic data to obtain target data;

[0100] A feature extraction module 204 is used to extract the colorectal cancer prognostic features of the target data according to the key prognostic indicators, perform dimensionality reduction processing on the colorectal cancer prognostic features to obtain key prognostic features, and determine derived features of the key prognostic features based on the disease characteristics;

[0101] The model building module 205 is used to perform a survival analysis on the rectal cancer patients based on the key prognostic indicators to obtain survival analysis results, and to build a prognostic model for the rectal cancer patients in combination with the key prognostic features, the imaging features and the survival analysis results.

[0102] In detail, each module in the rectal cancer prognosis model construction system 200 based on deep learning in the embodiment of the present invention is used in the same manner as described above. Figure 1The same technical means are used as the method for constructing a rectal cancer prognosis model based on deep learning described in , and can produce the same technical effects, so I will not go into details here.

[0103] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0104] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0105] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0106] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0107] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for constructing a rectal cancer prognosis model based on deep learning, characterized in that: The method comprises: Acquire a rectal cancer patient to be diagnosed, and collect medical data of the rectal cancer patient, identify disease characteristics of the rectal cancer patient based on the medical data, set baseline standards for the disease characteristics, and define prognostic requirements for the rectal cancer patient based on the baseline standards; Determine the lesion type of the rectal cancer patient according to the disease characteristics, identify the imaging features corresponding to the lesion type, and analyze the key prognostic indicators of the rectal cancer patient in combination with the lesion type and the imaging features; Based on the key prognostic indicators and the prognostic requirements, the prognostic influencing factors of the rectal cancer patients are analyzed, the key influencing factors of the prognostic influencing factors are identified, characteristic data of the medical data is extracted according to the key influencing factors, and the characteristic data is cleaned to obtain target data; Extracting the rectal cancer prognostic features of the target data according to the key prognostic indicators, performing dimensionality reduction processing on the rectal cancer prognostic features to obtain key prognostic features, and determining derived features of the key prognostic features based on the disease characteristics; Based on the key prognostic indicators, a survival analysis is performed on the rectal cancer patients to obtain survival analysis results, and a prognostic model for the rectal cancer patients is constructed by combining the key prognostic features, the imaging features and the survival analysis results.

2. The method according to claim 1, characterized in that The step of identifying the disease characteristics of the rectal cancer patient according to the medical data includes: Based on the medical data, identifying the preoperative evaluation data and postoperative pathology report of the rectal cancer patient; Based on the preoperative evaluation data, analyzing the preoperative clinical symptoms of the rectal cancer patient; According to the preoperative clinical symptoms, scheduling imaging data of the rectal cancer patient; Analyzing tumor characteristics of the rectal cancer patient based on the imaging data; identifying the postoperative physical condition of the rectal cancer patient according to the postoperative pathology report; The disease characteristics of the rectal cancer patient are identified by combining the tumor characteristics and the postoperative physical condition.

3. The method according to claim 1, characterized in that: Defining the prognostic requirements of the rectal cancer patient based on the baseline criteria includes: identifying the disease severity of the rectal cancer patient based on the baseline criteria; According to the severity of the disease, setting a disease treatment method for the rectal cancer patient; Identify patient disease characteristics for the baseline criteria; Based on the patient's disease characteristics and the severity of the disease, analyzing the survival period of the rectal cancer patient; Analyzing the risk of disease recurrence of the rectal cancer patient according to the circumferential resection margin status of the tumor; The prognostic requirements of the rectal cancer patient are defined in combination with the severity of the disease, the survival period and the risk of disease recurrence.

4. The method according to claim 1, characterized in that: The key prognostic indicators of the rectal cancer patient are analyzed by combining the lesion type and the imaging features, including: Based on the lesion type, collecting genomic expression factors of the rectal cancer patient; Analyzing the correlation between the imaging features and the genomic expression factors; According to the degree of correlation, identifying the image feature combination of the genomic expression factor; Based on the combination of imaging features, identifying the pathological stage of the rectal cancer patient; Analyzing the pathological response of the rectal cancer patient according to the pathological staging; identifying the survival rate and recurrence risk ratio of the rectal cancer patient based on the pathological stage and the pathological response; Based on the survival rate and the recurrence risk ratio, key prognostic indicators of the rectal cancer patients were analyzed.

5. The method according to claim 1, characterized in that The analyzing the prognostic influencing factors of the rectal cancer patient based on the key prognostic indicators and the prognostic requirements includes: Based on the key prognostic indicators and the prognostic requirements, analyzing the pathological remission status of the rectal cancer patient; identifying a complete remission state within the pathological remission state; identifying the extent of tumor regression in the rectal cancer patient according to the complete remission status; Analyze the adverse factors affecting the complete remission state; Based on the adverse influencing factors, analyzing the tumor resection effect of the rectal cancer patient; The factors affecting the prognosis of the rectal cancer patients were analyzed based on the pathological remission status, the tumor shrinkage degree and the tumor resection effect.

6. The method according to claim 1, characterized in that The step of extracting the rectal cancer prognostic features of the target data according to the key prognostic indicators includes: According to the key prognostic indicator, identifying a data item attribute of the target data; Extracting patient population characteristics of the key prognostic indicators based on the data item attributes; Classifying the patient group characteristics to obtain classification characteristics; Calculating the attribute correlation between the classification feature and the attribute of the data item; Analyzing the importance of the classification features to the key prognostic indicators; The rectal cancer prognosis features of the target data are extracted by combining the attribute correlation degree and the importance.

7. The method according to claim 1, characterized in that The rectal cancer prognostic features are subjected to dimensionality reduction processing to obtain key prognostic features, including: Collecting prognostic feature data of the rectal cancer prognostic feature; Analyzing the distribution of the rectal cancer prognostic features and the relationship between the features according to the prognostic feature data; Based on the distribution, identifying data characteristics of the prognostic feature data; Determining a dimensionality reduction method for the rectal cancer prognostic features according to the interrelationships between the features; Combining the data characteristics and the dimension reduction method, the rectal cancer prognostic features are subjected to dimension reduction processing to obtain key prognostic features.

8. The method according to claim 1, characterized in that Determining the derived features of the key prognostic features based on the disease features includes: Based on the disease characteristics, scheduling patient tumor data corresponding to the key prognostic features; extracting tumor volume, tumor markers and gene activity of the patient's tumor data; Analyzing the concentration level of the tumor marker, and identifying the treatment response speed and treatment effect of the tumor cells according to the concentration level; analyzing the growth pattern of the tumor cells based on the tumor volume; Calculating a risk score of the tumor cell according to the gene activity; Derivative features of the key prognostic features are determined by combining the treatment response rate, the treatment effect, the growth pattern and the risk score.

9. The method according to claim 1, characterized in that: Based on the key prognostic indicators, a survival analysis is performed on the rectal cancer patient to obtain a survival analysis result, including: Based on the key prognostic indicators, risk stratifying the rectal cancer patients to obtain patient risk classification; According to the risk classification of the patient, collecting genome data and imaging feature data of the rectal cancer patient; Based on the genomic data and the imaging feature data, performing a recurrence risk score on the rectal cancer patient to obtain a recurrence risk score; According to the recurrence risk score, analyzing the survival recurrence events and follow-up time of the rectal cancer patients; According to the survival recurrence events and the follow-up time, identifying indicators affecting the survival of the rectal cancer patient; Combining the survival recurrence events, the recurrence risk score and the survival influencing index, a survival analysis is performed on the rectal cancer patients to obtain a survival analysis result.

10. A rectal cancer prognosis model construction system based on deep learning, the system implementing the method according to claim 1, characterized in that: The system comprises: A demand analysis module, used for acquiring a rectal cancer patient to be diagnosed, collecting medical data of the rectal cancer patient, identifying disease characteristics of the rectal cancer patient based on the medical data, setting a baseline standard of the disease characteristics, and defining the prognosis needs of the rectal cancer patient based on the baseline standard; A prognostic indicator identification module, used to determine the lesion type of the rectal cancer patient according to the disease characteristics, identify the imaging features corresponding to the lesion type, and analyze the key prognostic indicators of the rectal cancer patient in combination with the lesion type and the imaging features; A data processing module, used for analyzing the prognostic influencing factors of the rectal cancer patient based on the key prognostic indicators and the prognostic requirements, identifying the key influencing factors of the prognostic influencing factors, extracting the characteristic data of the medical data according to the key influencing factors, and cleaning the characteristic data to obtain target data; A feature extraction module, used to extract the rectal cancer prognostic features of the target data according to the key prognostic indicators, perform dimensionality reduction processing on the rectal cancer prognostic features to obtain key prognostic features, and determine derived features of the key prognostic features based on the disease characteristics; The model building module is used to perform a survival analysis on the rectal cancer patients based on the key prognostic indicators to obtain survival analysis results, and to build a prognostic model for the rectal cancer patients by combining the key prognostic features, the imaging features and the survival analysis results.

Images