Deep learning-based endometrial cancer detection method and system

Abstract

The invention provides a deep learning-based endometrial cancer detection method and system, and relates to the technical field of cancer detection, and the method comprises the steps: obtaining preliminary survey data; determining a preliminary survey result; determining whether histopathology detection is needed or not; if histopathology detection needs to be carried out, hysteroscope guided biopsy data of the patient to be detected are obtained, and a histopathology definite diagnosis result is determined; determining whether further detection is needed or not; if further detection is needed, acquiring patient physiological state data, medical image data and laboratory detection data of the to-be-detected patient; determining a physiological state influence coefficient of the patient; processing the medical image data, the laboratory detection data and the physiological state influence coefficient of the patient through a trained tumor deterioration time prediction model, and determining predicted tumor deterioration time; determining an endometrial cancer danger coefficient; and generating a detection report. According to the invention, the accuracy of the detection result of the endometrial cancer can be improved.

Description

Technical Field

[0001] The present invention relates to the field of cancer detection technology, and in particular to a deep learning-based endometrial cancer detection method and system. Background Art

[0002] In related technologies, endometrial cancer can be detected by performing in vivo pathological tests on the patient to be tested, which can improve the accuracy of the detection. However, related technologies make it difficult to predict the rate of tumor deterioration, that is, it is difficult to predict the time of tumor deterioration based on data such as ADC value and histological type.

[0003] The information disclosed in the background technology section of this application is only intended to deepen the understanding of the general background technology of this application, and should not be regarded as an admission or any form of suggestion that the information constitutes the prior art already known to those skilled in the art. Summary of the Invention

[0004] The present invention provides a deep learning-based detection method and system for endometrial cancer, which can solve the technical problem that related technologies are difficult to predict the time of tumor deterioration.

[0005] According to a first aspect of the present invention, a method for detecting endometrial cancer based on deep learning is provided, comprising: Obtain preliminary survey data from patients to be tested; Determine preliminary investigation results based on the preliminary investigation data; Determine whether histopathological examination is necessary based on the preliminary investigation results; If a histopathological examination is required, obtaining hysteroscopy-guided biopsy data of the patient to be tested, and determining a histopathological diagnosis based on the hysteroscopy-guided biopsy data; Determine whether further testing is necessary based on the tissue pathology confirmation results; If further testing is required, obtain the patient's physiological status data, medical imaging data, and laboratory test data; Determining a patient physiological state influence coefficient based on the patient physiological state data; Processing the medical imaging data, the laboratory test data, and the patient's physiological status influence coefficient using a trained tumor exacerbation time prediction model to determine a predicted tumor exacerbation time; determining the risk factor for endometrial cancer based on the predicted tumor progression time; A test report is generated based on the preliminary investigation results, the tissue pathology confirmation results and the endometrial cancer risk factor.

[0006] According to the present invention, determining preliminary investigation results based on the preliminary investigation data includes: determining abnormal bleeding identification results, abnormal discharge identification results, and abnormal pain identification results based on the preliminary investigation data; A preliminary investigation result is determined based on the abnormal bleeding identification result, the abnormal discharge identification result, and the abnormal pain identification result.

[0007] According to the present invention, determining the patient's physiological state influence coefficient based on the patient's physiological state data includes: Determining the patient's age, BMI, disease history, cytotoxic T cell density, helper regulatory T cell ratio, and macrophage ratio based on the patient's physiological status data; Obtaining diabetes identification results and cardiovascular disease identification results based on the patient's medical history; Determining a patient age risk identification result according to the patient age; The patient's physiological state influence coefficient is determined based on the patient's age risk identification result, the patient's BMI, the toxic T cell density, the helper regulatory T cell ratio, the macrophage ratio, the diabetes identification result, and the cardiovascular disease identification result.

[0008] According to the present invention, determining the patient's physiological state influence coefficient based on the patient's age risk identification result, the patient's BMI, the cytotoxic T cell density, the helper regulatory T cell ratio, the macrophage ratio, the diabetes identification result, and the cardiovascular disease identification result includes: According to the formula

[0009] Determine the patient physiological status influence coefficient of the i-th patient to be tested , where if is a conditional function and and is the logical operator of "and". is the diabetes identification result of the i-th patient to be tested, , is the BMI of the i-th patient to be tested, is the patient age risk identification result of the i-th patient to be tested, , is the cardiovascular disease identification result of the i-th patient to be tested, , is the cytotoxic T cell density of the i-th patient to be tested, To preset the cytotoxic T cell density threshold, is the ratio of helper regulatory T cells of the i-th patient to be tested, To preset the threshold of the ratio of helper regulatory T cells, is the macrophage ratio of the i-th patient to be tested, The preset macrophage ratio threshold.

[0010] According to the present invention, the training step of the tumor progression time prediction model includes: Obtain historical patient physiological status data, historical medical imaging data, and historical laboratory test data for multiple historical patients; Determining an influence coefficient of the historical patient physiological state based on the historical patient physiological state data; Based on the historical medical imaging data, determine the historical degree of myometrial invasion, historical ADC value, and historical depth of cervical stromal invasion; Based on the historical laboratory test data, determine the historical histological type risk level and historical lymphovascular space infiltration identification results; The historical patient physiological state influence coefficient, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space infiltration identification result are processed by a tumor deterioration time prediction model to obtain a historical training deterioration time; Obtain the historical actual deterioration time of multiple historical patients; Determining a training loss function for a tumor deterioration time prediction model based on the historical patient physiological state influence coefficient, the historical training deterioration time, the historical actual deterioration time, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space invasion identification result; The tumor exacerbation time prediction model is trained according to the training loss function to obtain a trained tumor exacerbation time prediction model.

[0011] According to the present invention, a training loss function of a tumor deterioration time prediction model is determined based on the historical patient physiological state influence coefficient, the historical training deterioration time, the historical actual deterioration time, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space invasion identification result, including: According to the formula

[0012] Determine the training loss function of the tumor progression time prediction model ,in, is the historical training deterioration time of the kth historical patient, is the actual historical deterioration time of the kth historical patient, is the historical ADC value of the kth historical patient, To preset the ADC value threshold, is the historical muscle invasion degree of the kth historical patient, To preset the threshold of muscle invasion degree, is the historical cervical stromal invasion depth of the kth historical patient, To preset the historical cervical stromal invasion depth threshold, is the historical histological type risk level of the kth historical patient, To preset the histological type risk level threshold, is the historical lymphovascular space infiltration identification result of the kth historical patient, , is the influence coefficient of the physiological status of the kth historical patient, is the preset threshold of the patient's physiological status influence coefficient, K is the number of historical patients, k≤K, and both k and K are positive integers.

[0013] According to the present invention, the medical imaging data, the laboratory test data, and the patient's physiological status influence coefficient are processed by a trained tumor exacerbation time prediction model to determine the predicted tumor exacerbation time, including: Determine the degree of myometrial invasion, ADC value, and depth of cervical stromal invasion based on the medical imaging data; Determine the histological risk level and lymphovascular space invasion identification results based on the laboratory test data; The patient's physiological state influence coefficient, the degree of myometrial invasion, the ADC value, the cervical stromal invasion depth, the histological type risk level and the lymphovascular space infiltration identification result are processed through the trained tumor deterioration time prediction model to obtain the predicted tumor deterioration time.

[0014] According to the present invention, determining the risk factor for endometrial cancer based on the predicted tumor progression time includes: determining a first difference according to the predicted tumor exacerbation time and a set tumor exacerbation time threshold; determining a first ratio according to the first difference and a set tumor progression time threshold; The risk factor for endometrial cancer is determined based on the first ratio.

[0015] According to a second aspect of the present invention, a deep learning-based endometrial cancer detection system is provided, comprising: Preliminary data module, to obtain preliminary survey data of patients to be tested; A preliminary result module, which determines preliminary investigation results based on the preliminary investigation data; A pathology detection module determines whether a tissue pathology test is needed based on the preliminary investigation results; Pathology result module, if a tissue pathology test is required, obtains the hysteroscopy-guided biopsy data of the patient to be tested, and determines the tissue pathology diagnosis result based on the hysteroscopy-guided biopsy data; A deep detection module determines whether further testing is needed based on the tissue pathology diagnosis results; Detailed data module, if further testing is required, to obtain the patient's physiological status data, medical imaging data and laboratory test data; An influence coefficient module, for determining an influence coefficient of the patient's physiological state according to the patient's physiological state data; A prediction time module processes the medical imaging data, the laboratory test data, and the patient's physiological status influence coefficient using a trained tumor exacerbation time prediction model to determine a predicted tumor exacerbation time; a risk coefficient module, which determines the risk coefficient of endometrial cancer based on the predicted tumor progression time; The test report module generates a test report based on the preliminary investigation results, the tissue pathology confirmation results and the endometrial cancer risk factor.

[0016] Technical effect: According to the present invention, a preliminary investigation can be conducted on the patient to be tested to determine whether a histopathological examination is needed, and the accuracy of the test results can be improved through histopathological examination. Furthermore, in the case of confirmed endometrial cancer, the time of tumor deterioration can be predicted based on the patient's physiological status data, medical imaging data and laboratory test data of the patient to be tested, the risk level of the tumor can be evaluated based on the predicted tumor deterioration time, the endometrial cancer risk factor can be determined, and a test report can be generated, thereby improving the comprehensiveness and accuracy of the endometrial cancer test results. When determining the patient's physiological status influence coefficient, the patient's physiological status influence coefficient can be determined based on the patient's age risk identification results, patient BMI, toxic T cell density, auxiliary regulatory T cell ratio, macrophage ratio, diabetes identification results and cardiovascular disease identification results. During the calculation process, the health status of the patient to be tested can be evaluated based on four aspects: metabolic status, age status, cardiovascular disease status and immune status, thereby improving the comprehensiveness and accuracy of the patient's physiological status influence coefficient. When determining the training loss function, the training loss function of the tumor deterioration time prediction model can be determined based on the historical patient physiological status influence coefficient, historical training deterioration time, historical actual deterioration time, historical myometrial invasion degree, historical ADC value, historical cervical stromal invasion depth, historical histological type risk level, and lymphovascular space invasion identification results. During the calculation process, the influence of the historical patient physiological status influence coefficient, historical training deterioration time, historical actual deterioration time, historical myometrial invasion degree, historical ADC value, historical cervical stromal invasion depth, historical histological type risk level, and lymphovascular space invasion identification results on the tumor deterioration rate can be used to determine the impact of the above data on the error of the historical training deterioration time. Based on this impact and the relative error of the historical training deterioration time, the training loss function is set to reduce the training loss function of the tumor deterioration time prediction model during the training process, thereby more specifically improving the accuracy of the tumor deterioration time prediction model.

[0017] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and not limiting of the present invention. Other features and aspects of the present invention will become more apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can derive other embodiments based on these drawings without inventive efforts. Figure 1A schematic diagram exemplarily illustrates a flow chart of a method for detecting endometrial cancer based on deep learning according to an embodiment of the present invention; Figure 2 A schematic diagram exemplarily illustrates a method for determining a preliminary investigation result according to an embodiment of the present invention; Figure 3 A schematic diagram illustrating, by way of example, determining a patient's physiological state influence coefficient according to an embodiment of the present invention; Figure 4 A schematic diagram exemplarily shows a method for determining and predicting tumor deterioration time according to an embodiment of the present invention; Figure 5 A schematic diagram exemplarily illustrates a method for determining a risk factor for endometrial cancer according to an embodiment of the present invention; Figure 6 A block diagram of a deep learning-based endometrial cancer detection system according to an embodiment of the present invention is exemplarily shown. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0020] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0021] Figure 1 A flowchart of a method for detecting endometrial cancer based on deep learning according to an embodiment of the present invention is exemplarily shown. The method includes: Step S1, obtaining preliminary survey data of the patient to be tested; Step S2, determining preliminary investigation results based on the preliminary investigation data; Step S3, determining whether a tissue pathology test is needed based on the preliminary investigation results; Step S4: if a histopathological examination is required, obtaining hysteroscopy-guided biopsy data of the patient to be tested, and determining a histopathological diagnosis result based on the hysteroscopy-guided biopsy data; Step S5, determining whether further testing is required based on the tissue pathology diagnosis result; Step S6, if further testing is required, obtaining the patient's physiological status data, medical imaging data, and laboratory test data of the patient to be tested; Step S7, determining the patient's physiological state influence coefficient based on the patient's physiological state data; Step S8, processing the medical imaging data, the laboratory test data, and the patient's physiological status influence coefficient using the trained tumor exacerbation time prediction model to determine a predicted tumor exacerbation time; Step S9, determining the risk factor of endometrial cancer based on the predicted tumor progression time; Step S10: generating a test report based on the preliminary investigation results, the histopathological diagnosis results, and the endometrial cancer risk factor.

[0022] According to the deep learning-based endometrial cancer detection method of an embodiment of the present invention, a preliminary investigation can be conducted on the patient to be tested to determine whether a tissue pathology test is needed, and the accuracy of the test results can be improved through tissue pathology testing. Furthermore, in the case of a confirmed endometrial cancer, the tumor deterioration time is predicted based on the patient's physiological status data, medical imaging data and laboratory test data of the patient to be tested, the risk level of the tumor is evaluated based on the predicted tumor deterioration time, the endometrial cancer risk factor is determined, and a test report is generated, thereby improving the comprehensiveness and accuracy of the endometrial cancer test results.

[0023] According to one embodiment of the present invention, in step S1, preliminary investigation data of the patient to be tested is obtained. For example, a preliminary investigation of the patient can be conducted through oral questioning or questionnaire survey to investigate whether the patient has abnormal symptoms and obtain preliminary investigation data.

[0024] According to an embodiment of the present invention, in step S2, a preliminary investigation result is determined based on the preliminary investigation data.

[0025] Figure 2 The following is a schematic diagram illustrating, by way of example, determining a preliminary investigation result according to an embodiment of the present invention.

[0026] According to one embodiment of the present invention, step S2 includes: Step S21, determining abnormal bleeding recognition results, abnormal discharge recognition results, and abnormal pain recognition results based on the preliminary investigation data; Step S22: determining preliminary investigation results based on the abnormal bleeding identification result, the abnormal discharge identification result, and the abnormal pain identification result.

[0027] For example, oral inquiries or questionnaires are used to determine whether the patient to be tested has abnormal uterine bleeding (such as postmenopausal bleeding, abnormal premenopausal bleeding), abnormal vaginal discharge (such as serous or bloody discharge) and abnormal lower abdominal pain. If abnormal uterine bleeding exists, the abnormal bleeding identification result is 1, otherwise, the abnormal bleeding identification result is 0. If abnormal vaginal discharge exists, the abnormal discharge identification result is 1, otherwise, the abnormal discharge identification result is 0. If abnormal lower abdominal pain exists, the abnormal pain identification result is 1, otherwise, the abnormal pain identification result is 0. The preliminary investigation results are determined based on the sum of the abnormal bleeding identification result, the abnormal discharge identification result and the abnormal pain identification result.

[0028] According to one embodiment of the present invention, in step S3, it is determined whether a tissue pathology test is required based on the preliminary investigation results.

[0029] For example, if the preliminary investigation result is equal to 0, it means that no warning symptoms have occurred and no tissue pathology test is required. If the preliminary investigation result is greater than 0, it means that abnormal uterine bleeding, abnormal vaginal discharge, or abnormal lower abdominal pain exists and a tissue pathology test is required.

[0030] According to one embodiment of the present invention, in step S4, if a histopathological test is required, hysteroscopy-guided biopsy data of the patient to be tested is obtained, and a histopathological diagnosis result is determined based on the hysteroscopy-guided biopsy data.

[0031] For example, if a histopathological test is required, a hysteroscopically guided biopsy is performed on the patient to be tested to obtain hysteroscopically guided biopsy data, and a histopathological examination is performed on the hysteroscopically guided biopsy data obtained to determine the histopathological diagnosis result. If it is confirmed that the patient has endometrial cancer, the histopathological diagnosis result is 1, otherwise, the histopathological diagnosis result is 0.

[0032] According to one embodiment of the present invention, in step S5, it is determined whether further testing is required based on the tissue pathology diagnosis result.

[0033] For example, if the histopathological diagnosis result is 1, it indicates that the patient has endometrial cancer and further testing is required. If the histopathological diagnosis result is 0, it indicates that the patient has not had endometrial cancer and no further testing is required.

[0034] According to one embodiment of the present invention, in step S6, if further testing is required, the patient's physiological status data, medical imaging data, and laboratory test data of the patient to be tested are obtained.

[0035] For example, based on the medical history and normal physiological condition of the patient to be tested, the patient's physiological status data is obtained, the patient to be tested is subjected to transvaginal ultrasound, pelvic MRI and other tests to obtain medical imaging data, and the samples obtained by hysteroscopy-guided biopsy are subjected to laboratory testing to obtain laboratory test data.

[0036] According to one embodiment of the present invention, in step S7, the patient's physiological state influence coefficient is determined based on the patient's physiological state data.

[0037] Figure 3 A schematic diagram of determining the influence coefficient of a patient's physiological state according to an embodiment of the present invention is exemplarily shown.

[0038] According to one embodiment of the present invention, step S7 includes: Step S71, determining the patient's age, BMI, medical history, cytotoxic T cell density, helper regulatory T cell ratio, and macrophage ratio based on the patient's physiological status data; Step S72, obtaining diabetes recognition results and cardiovascular disease recognition results based on the patient's medical history; Step S73, determining the patient age risk identification result according to the patient age; Step S74, determining the patient's physiological state influence coefficient based on the patient's age risk identification result, the patient's BMI, the cytotoxic T cell density, the helper regulatory T cell ratio, the macrophage ratio, the diabetes identification result, and the cardiovascular disease identification result.

[0039] For example, according to the medical records of the patient to be tested, the patient's age, height, weight and disease history are obtained, and the patient's BMI is determined according to the patient's height and weight. The CD8+ cytotoxic T cell density, CD4+ helper T cell / regulatory T cell ratio and CD68+ / CD163+ macrophage ratio, that is, cytotoxic T cell density, helper regulatory T cell ratio and macrophage ratio, are detected by immunohistochemistry staining, multicolor immunofluorescence and IHC double staining respectively; according to the patient's disease history, the diabetes identification result and cardiovascular disease identification result are determined. If the patient to be tested has a history of diabetes, then diabetes The identification result is 1, otherwise the diabetes identification result is 0. If the patient to be tested has a history of cardiovascular disease, the cardiovascular disease identification result is 0.5, otherwise the cardiovascular disease identification result is 1; if the patient to be tested is greater than or equal to 70 years old, the patient's age risk identification result is 0.5, otherwise the patient's age risk identification result is 1; based on the patient's age risk identification result, patient BMI, cytotoxic T cell density, auxiliary regulatory T cell ratio, macrophage ratio, diabetes identification result and cardiovascular disease identification result, the patient's immunity and health status are evaluated to determine the patient's physiological status influence coefficient.

[0040] According to one embodiment of the present invention, step S74 includes: determining the patient physiological state influence factor of the i-th patient to be tested according to formula (1): , (1) Among them, if is a conditional function, and is the logical operator of "and". is the diabetes identification result of the i-th patient to be tested, , is the BMI of the i-th patient to be tested, is the patient age risk identification result of the i-th patient to be tested, , is the cardiovascular disease identification result of the i-th patient to be tested, , is the cytotoxic T cell density of the i-th patient to be tested, To preset the cytotoxic T cell density threshold, is the ratio of helper regulatory T cells of the i-th patient to be tested, To preset the threshold of the ratio of helper regulatory T cells, is the macrophage ratio of the i-th patient to be tested, The preset macrophage ratio threshold.

[0041] According to one embodiment of the present invention, in formula (1), the conditional function The value of includes the following two cases, when it satisfies When the condition is met, it means that the i-th patient has a history of diabetes and is overweight. When the patient has diabetes and BMI is greater than or equal to 35, the patient may suffer from metabolic syndrome. Metabolic abnormalities may lead to chronic inflammation, and chronic inflammation promotes tumor growth. The patient's physical condition is poor in terms of metabolism. The value of the condition function is 0.5. When it is not met When the condition is met, it means that the i-th patient to be tested has no history of diabetes and is overweight, and the possibility of the patient to be tested suffering from metabolic syndrome is small. The patient's physical condition in terms of metabolism is normal, and the value of the conditional function is 1.

[0042] According to one embodiment of the present invention, is the patient age risk identification result of the i-th patient to be tested. When the age of the i-th patient to be tested is greater than or equal to 70 years old, the tolerance of patients over 70 years old is poor. is 0.5, indicating that the physical condition of the patient to be tested is poor according to age assessment. When the age of the i-th patient to be tested is less than 70 years old, A value of 1 indicates that the patient's physical condition is normal based on age. is the cardiovascular disease identification result of the i-th patient to be tested. When the i-th patient to be tested has cardiovascular disease, cardiovascular disease will limit the implementation of treatment. is 0.5, indicating that the physical condition of the patient to be tested is poor according to the cardiovascular disease status assessment. When the i-th patient to be tested does not have cardiovascular disease, A value of 1 indicates that the physical condition of the patient under test is normal according to the cardiovascular disease status assessment.

[0043] According to one embodiment of the present invention, cytotoxic T cells are the main force in directly killing tumors. The higher the density of cytotoxic T cells, the stronger the immune activity. is the ratio of the cytotoxic T cell density of the i-th patient to be tested to the preset cytotoxic T cell density threshold. The larger the ratio, the higher the cytotoxic T cell density of the i-th patient to be tested, and the stronger the immune activity of the i-th patient to be tested. Can be set to 20 / HPF, Indicates the physical condition of the patient to be tested determined by the density of cytotoxic T cells. The bigger the body, the better the physical condition. is the ratio of CD4+ helper T cells to regulatory T cells of the i-th patient to be tested. Helper T cells are anti-tumor type, and regulatory T cells are tumor-promoting type. The larger the value, the more favorable it is for tumor treatment. is the ratio of the helper regulatory T cell ratio of the i-th patient to be tested to the preset helper regulatory T cell ratio threshold. The larger the ratio, the larger the helper regulatory T cell ratio of the i-th patient to be tested, and the more suitable the immune status of the i-th patient to be tested is for tumor treatment. Can be set to 2.5, Indicates the physical condition of the patient to be tested determined based on the ratio of helper regulatory T cells, The bigger the body, the better the physical condition. is the ratio of CD68+ macrophages to CD163+ macrophages of the i-th patient to be tested. CD68+ macrophages are anti-tumor and CD163+ macrophages are pro-metastasis. The larger the value, the more favorable it is for tumor treatment. is the ratio of the macrophage ratio of the i-th patient to be tested to the preset macrophage ratio threshold. The larger the ratio, the larger the macrophage ratio of the i-th patient to be tested, and the more suitable the immune status of the i-th patient to be tested is for tumor treatment. Can be set to 1.5, Indicates the physical condition of the patient to be tested determined based on the macrophage ratio, The bigger it is, the better the physical condition is.

[0044] According to one embodiment of the present invention, It means that the health status of the patient to be tested is evaluated based on the patient's metabolic status, age status, cardiovascular disease status and immune status, and the influence coefficient of the patient's physiological status is determined.

[0045] In this way, the patient's physiological status influence coefficient can be determined based on the patient's age risk identification results, patient BMI, toxic T cell density, auxiliary regulatory T cell ratio, macrophage ratio, diabetes identification results and cardiovascular disease identification results. During the calculation process, the health status of the patient to be tested can be evaluated through four aspects: metabolic status, age status, cardiovascular disease status and immune status, thereby improving the comprehensiveness and accuracy of the patient's physiological status influence coefficient.

[0046] According to one embodiment of the present invention, in step S8, the medical imaging data, the laboratory test data and the patient's physiological status influence coefficient are processed by a trained tumor deterioration time prediction model to determine a predicted tumor deterioration time.

[0047] Figure 4 The following is a schematic diagram illustrating, by way of example, determining the predicted tumor deterioration time according to an embodiment of the present invention.

[0048] According to one embodiment of the present invention, step S8 includes: Step S81, determining the degree of myometrial invasion, ADC value, and depth of cervical stromal invasion based on the medical imaging data; Step S82, determining the histological type risk level and lymphovascular space infiltration identification result based on the laboratory test data; Step S83, using the trained tumor deterioration time prediction model to process the patient's physiological state influence coefficient, the myometrial infiltration degree, the ADC value, the cervical stromal invasion depth, the histological type risk level, and the lymphovascular space infiltration identification result to obtain a predicted tumor deterioration time.

[0049] For example, T2WI imaging features are determined based on pelvic MRI, and the degree of infiltration is determined based on T2WI imaging features. For example, when the high signal of the tumor does not penetrate the midline of the low signal junction zone, it indicates that the degree of infiltration is less than 50%. When the high signal of the tumor penetrates the low signal zone and reaches the outer 1 / 2 of the muscle layer, it indicates that the degree of infiltration is greater than or equal to 50%. Diffusion-weighted imaging (DWI) combined with apparent diffusion coefficient (ADC) quantitative analysis is used to determine the DWI / ADC value, that is, the ADC value. The depth of cervical stromal invasion is determined by combining MRI sagittal and axial observations. The histological type is determined by histopathological HE staining combined with immunohistochemistry (IHC) assistance. The risk level of the histological type is determined based on the histological type. For example, regular glandular structure and nuclear atypia less than 5% are low risk, and histological type The histological risk level is 1, with glandular fusion or cribriform structure accounting for 6% to 50% as intermediate risk. The histological risk level is 2, with solid areas greater than 50% or significant nuclear atypia as high risk. The histological risk level is 3. Lymphovascular space invasion detection is performed based on the double verification of HE staining and IHC observation. When the test shows LVSI positive, the lymphovascular space invasion identification result is 1.4. Otherwise, the lymphovascular space invasion identification result is 1. The patient's physiological status influence coefficient, myometrial invasion degree, ADC value, cervical stromal invasion depth, histological type risk level and lymphovascular space invasion identification result are processed through the trained tumor deterioration time prediction model to obtain the time when the tumor may relapse locally, metastasize distantly or upgrade the pathological grade, that is, predict the time of tumor deterioration.

[0050] According to one embodiment of the present invention, the step of training the tumor progression time prediction model includes: Obtain historical patient physiological status data, historical medical imaging data, and historical laboratory test data for multiple historical patients; Determining an influence coefficient of the historical patient physiological state based on the historical patient physiological state data; Based on the historical medical imaging data, determine the historical degree of myometrial invasion, historical ADC value, and historical depth of cervical stromal invasion; Based on the historical laboratory test data, determine the historical histological type risk level and historical lymphovascular space infiltration identification results; The historical patient physiological state influence coefficient, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space infiltration identification result are processed by a tumor deterioration time prediction model to obtain a historical training deterioration time; Obtain the historical actual deterioration time of multiple historical patients; Determining a training loss function for a tumor deterioration time prediction model based on the historical patient physiological state influence coefficient, the historical training deterioration time, the historical actual deterioration time, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space invasion identification result; The tumor exacerbation time prediction model is trained according to the training loss function to obtain a trained tumor exacerbation time prediction model.

[0051] For example, obtain the historical patient physiological status data, historical medical imaging data and historical laboratory test data of historical patients who have completed clinical treatment; determine the historical patient physiological status influence coefficient based on the historical patient physiological status data, and the calculation method of the historical patient physiological status influence coefficient is similar to the calculation method of the patient physiological status influence coefficient in formula (1), which will not be repeated here; determine the historical myometrial invasion depth, historical ADC value and historical cervical stromal invasion depth of historical patients based on the historical medical imaging data, and the determination method of the historical myometrial invasion depth, historical ADC value and historical cervical stromal invasion depth of historical patients is the same as the determination method of the myometrial invasion depth, ADC value and historical cervical stromal invasion depth, which will not be repeated here; determine the historical histological type risk level and historical lymphovascular space invasion identification results of historical patients based on the historical laboratory test data, and the historical histological type risk level and historical lymphovascular space invasion identification results are confirmed by the histological type risk level and lymphovascular space invasion identification results. The determination method is the same and will not be repeated here; the historical patient physiological state influence coefficient, historical myometrial invasion depth, historical ADC value, historical cervical stromal invasion depth, historical histological type risk level and historical lymphovascular space infiltration identification results are processed by the tumor deterioration time prediction model to obtain the time when the historical patient's tumor may undergo local recurrence, distant metastasis or pathological grade upgrade, that is, the historical training deterioration time; the historical actual deterioration time of multiple historical patients in actual clinical treatment is obtained; according to the historical patient physiological state influence coefficient, historical training deterioration time, historical actual deterioration time, historical myometrial invasion depth, historical ADC value, historical cervical stromal invasion depth, historical histological type risk level and historical lymphovascular space infiltration identification results, the training loss function of the tumor deterioration time prediction model is determined; the tumor deterioration time prediction model is trained according to the training loss function to improve the accuracy of the tumor deterioration time prediction model for tumor deterioration time prediction, and a trained tumor deterioration time prediction model is obtained.

[0052] According to one embodiment of the present invention, the training loss function of the tumor deterioration time prediction model is determined based on the historical patient physiological state influence coefficient, the historical training deterioration time, the historical actual deterioration time, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level and the historical lymphovascular space infiltration identification result, including: determining the training loss function of the tumor deterioration time prediction model according to formula (2): , (2) in, is the historical training deterioration time of the kth historical patient, is the actual historical deterioration time of the kth historical patient, is the historical ADC value of the kth historical patient, To preset the ADC value threshold, is the historical muscle invasion degree of the kth historical patient, To preset the threshold of muscle invasion degree, is the historical cervical stromal invasion depth of the kth historical patient, To preset the historical cervical stromal invasion depth threshold, is the historical histological type risk level of the kth historical patient, To preset the histological type risk level threshold, is the historical lymphovascular space infiltration identification result of the kth historical patient, , is the influence coefficient of the physiological status of the kth historical patient, is the preset threshold of the patient's physiological status influence coefficient, K is the number of historical patients, k≤K, and both k and K are positive integers.

[0053] According to one embodiment of the present invention, is the ratio of the historical myometrial invasion degree of the kth historical patient to the preset myometrial invasion degree threshold. The larger the ratio is, the greater the historical myometrial invasion degree of the kth historical patient is. The preset myometrial invasion degree threshold can be set to 50%. is the ratio of the historical cervical stromal invasion depth of the kth historical patient to the preset historical cervical stromal invasion depth threshold. The larger the ratio, the deeper the historical cervical stromal invasion depth of the kth historical patient. Can be set to 3mm, It is the ratio of the historical histological type risk level of the kth historical patient to the preset histological type risk level threshold. The larger the ratio is, the higher the historical histological type risk level of the kth historical patient is. The preset histological type risk level threshold can be set to level 1. is the historical lymphovascular space infiltration identification result of the kth historical patient. When lymphovascular space infiltration occurs, =1.4, indicating that when lymphovascular invasion occurs, the risk of tumor metastasis increases by 40%, and when lymphovascular invasion does not occur, is equal to 1, It is the ratio of the historical patient physiological state influence coefficient of the kth historical patient to the preset patient physiological state influence coefficient threshold. The larger the ratio is, the greater the historical patient physiological state influence coefficient of the kth historical patient is. The preset patient physiological state influence coefficient threshold can be set to 1. is the ratio of the historical ADC value of the kth historical patient to the preset ADC value threshold. The larger the ratio is, the larger the historical ADC of the kth historical patient is. Can be set to , It indicates that the historical patient physiological status influence coefficient, historical ADC value and historical training deterioration time are positively correlated. For example, the greater the historical patient physiological status influence coefficient, the more favorable the physical health condition of the patient to be tested is for tumor treatment, the larger the value of historical training deterioration time is, the larger the historical ADC value is, usually indicating a better treatment effect, indicating that the growth of the tumor is inhibited or delayed, and the clinical deterioration rate is slowed down. The larger the value of historical training deterioration time is, the negative correlation is between the historical myometrial invasion degree, the historical cervical stromal invasion depth, the historical histological type risk level, the historical lymphovascular space infiltration identification result and the historical training deterioration time. For example, the historical The greater the degree of myometrial invasion, the easier it is to break through the uterine serosa and invade adjacent organs (such as the bladder and rectum), causing a sudden increase in the risk of lymph node metastasis. The smaller the value of the historical training deterioration time is. The deeper the historical cervical stromal invasion depth is, the easier it is to break through the physical barrier of the cervical matrix and quickly spread to the paracervical tissue. The smaller the value of the historical training deterioration time is. The higher the historical histological type risk level is, the faster the tumor grows, and the lower the sensitivity to radiotherapy is. The smaller the value of the historical training deterioration time is. When lymphovascular space infiltration occurs, a "high-speed channel" is provided for tumor cells, and the risk of lymph node metastasis increases by 40%. The smaller the value of the historical training deterioration time is. Therefore, the items related to the historical patient physiological status influence coefficient and the historical ADC value are placed in the numerator, and the items related to the historical myometrial invasion degree, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space infiltration identification results are placed in the denominator, indicating that and The larger the value of , the larger the value of historical training deterioration time. 、 、 and The larger the value of , the smaller the value of historical training deterioration time.

[0054] According to one embodiment of the present invention, is the relative error between the historical training deterioration time and the historical actual deterioration time of the kth historical patient, using The training loss function is obtained by taking a weighted average of the relative errors between the historical training exacerbation time and the historical actual exacerbation time for each patient. During the training process, this training loss function is reduced, thereby reducing the error between the historical training exacerbation time and the historical actual exacerbation time, thereby improving the prediction accuracy of the tumor exacerbation time prediction model.

[0055] In this way, the training loss function of the tumor deterioration time prediction model can be determined based on the historical patient physiological status influence coefficient, historical training deterioration time, historical actual deterioration time, historical myometrial invasion degree, historical ADC value, historical cervical stromal invasion depth, historical histological type risk level, and lymphovascular space invasion identification results. During the calculation process, the influence of the historical patient physiological status influence coefficient, historical training deterioration time, historical actual deterioration time, historical myometrial invasion degree, historical ADC value, historical cervical stromal invasion depth, historical histological type risk level, and lymphovascular space invasion identification results on the tumor deterioration rate can be used to determine the impact of the above data on the error of the historical training deterioration time. Based on this impact and the relative error of the historical training deterioration time, the training loss function is set to reduce the training loss function of the tumor deterioration time prediction model during the training process, thereby more specifically improving the accuracy of the tumor deterioration time prediction model.

[0056] According to one embodiment of the present invention, in step S9, the risk factor of endometrial cancer is determined based on the predicted tumor progression time.

[0057] Figure 5 The following is a schematic diagram showing, by way of example, how to determine the risk factor for endometrial cancer according to an embodiment of the present invention.

[0058] According to one embodiment of the present invention, step S9 includes: Step S91, determining a first difference value based on the predicted tumor worsening time and a set tumor worsening time threshold; Step S92, determining a first ratio according to the first difference and a set tumor progression time threshold; Step S93: determining the risk factor of endometrial cancer based on the first ratio.

[0059] For example, a first difference is determined based on the set tumor deterioration time threshold minus the predicted tumor deterioration time, and the predicted tumor deterioration time can be set to 6 months; a first ratio is determined based on the ratio of the first difference and the set tumor deterioration time threshold; when the first ratio is less than 0, the endometrial cancer risk factor is 1, indicating a certain risk; when the first ratio is greater than or equal to 0 and less than or equal to 0.5, the endometrial cancer risk factor is 2, indicating a high degree of risk; when the first ratio is greater than 0.5, the endometrial cancer risk factor is 3, indicating a serious degree of risk.

[0060] According to one embodiment of the present invention, in step S10, a test report is generated based on the preliminary investigation results, the tissue pathology diagnosis results, and the endometrial cancer risk factor.

[0061] For example, based on preliminary investigation results, histopathological diagnosis results and endometrial cancer risk coefficient, it is determined whether the patient to be tested is diagnosed with endometrial cancer and the risk level of the diagnosed endometrial cancer.

[0062] According to the deep learning-based endometrial cancer detection method of an embodiment of the present invention, a preliminary investigation can be conducted on the patient to be tested to determine whether a histopathological examination is needed, and the accuracy of the test results can be improved through histopathological examination. Furthermore, in the case of a confirmed diagnosis of endometrial cancer, the tumor deterioration time can be predicted based on the patient's physiological status data, medical imaging data, and laboratory test data of the patient to be tested. The risk level of the tumor is evaluated based on the predicted tumor deterioration time, the endometrial cancer risk factor is determined, and a test report is generated, thereby improving the comprehensiveness and accuracy of the endometrial cancer test results. When determining the patient's physiological status influence coefficient, the patient's physiological status influence coefficient can be determined based on the patient's age risk identification results, patient BMI, cytotoxic T cell density, auxiliary regulatory T cell ratio, macrophage ratio, diabetes identification results, and cardiovascular disease identification results. During the calculation process, the health status of the patient to be tested can be evaluated based on four aspects: metabolic status, age status, cardiovascular disease status, and immune status, thereby improving the comprehensiveness and accuracy of the patient's physiological status influence coefficient. When determining the training loss function, the training loss function of the tumor deterioration time prediction model can be determined based on the historical patient physiological status influence coefficient, historical training deterioration time, historical actual deterioration time, historical myometrial invasion degree, historical ADC value, historical cervical stromal invasion depth, historical histological type risk level, and lymphovascular space invasion identification results. During the calculation process, the influence of the historical patient physiological status influence coefficient, historical training deterioration time, historical actual deterioration time, historical myometrial invasion degree, historical ADC value, historical cervical stromal invasion depth, historical histological type risk level, and lymphovascular space invasion identification results on the tumor deterioration rate can be used to determine the impact of the above data on the error of the historical training deterioration time. Based on this impact and the relative error of the historical training deterioration time, the training loss function is set to reduce the training loss function of the tumor deterioration time prediction model during the training process, thereby more specifically improving the accuracy of the tumor deterioration time prediction model.

[0063] Figure 6 A block diagram of a deep learning-based endometrial cancer detection system according to an embodiment of the present invention is exemplarily shown. The system includes: Preliminary data module, to obtain preliminary survey data of patients to be tested; A preliminary result module, which determines preliminary investigation results based on the preliminary investigation data; A pathology detection module determines whether a tissue pathology test is needed based on the preliminary investigation results; Pathology result module, if a tissue pathology test is required, obtains the hysteroscopy-guided biopsy data of the patient to be tested, and determines the tissue pathology diagnosis result based on the hysteroscopy-guided biopsy data; A deep detection module determines whether further testing is needed based on the tissue pathology diagnosis results; Detailed data module, if further testing is required, to obtain the patient's physiological status data, medical imaging data and laboratory test data; An influence coefficient module, for determining an influence coefficient of the patient's physiological state according to the patient's physiological state data; A prediction time module processes the medical imaging data, the laboratory test data, and the patient's physiological status influence coefficient using a trained tumor exacerbation time prediction model to determine a predicted tumor exacerbation time; a risk coefficient module, which determines the risk coefficient of endometrial cancer based on the predicted tumor progression time; The test report module generates a test report based on the preliminary investigation results, the tissue pathology confirmation results and the endometrial cancer risk factor.

[0064] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0065] Those skilled in the art will appreciate that the embodiments of the present invention described above and shown in the accompanying drawings are intended to be illustrative only and are not intended to limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and illustrated in the embodiments. Any variations or modifications may be made to the embodiments of the present invention without departing from the principles described.

Claims

1. A method for detecting endometrial cancer based on deep learning, characterized in that: include: Obtain preliminary survey data from patients to be tested; Determine preliminary investigation results based on the preliminary investigation data; Determine whether histopathological examination is necessary based on the preliminary investigation results; If a histopathological examination is required, obtaining hysteroscopy-guided biopsy data of the patient to be tested, and determining a histopathological diagnosis based on the hysteroscopy-guided biopsy data; Determine whether further testing is necessary based on the tissue pathology confirmation results; If further testing is required, obtain the patient's physiological status data, medical imaging data, and laboratory test data; Determining a patient physiological state influence coefficient based on the patient physiological state data; Processing the medical imaging data, the laboratory test data, and the patient's physiological status influence coefficient using a trained tumor exacerbation time prediction model to determine a predicted tumor exacerbation time; determining the risk factor for endometrial cancer based on the predicted tumor progression time; A test report is generated based on the preliminary investigation results, the tissue pathology confirmation results and the endometrial cancer risk factor.

2. The deep learning-based endometrial cancer detection method according to claim 1, characterized in that: Based on the preliminary investigation data, determine the preliminary investigation results, including: determining abnormal bleeding identification results, abnormal discharge identification results, and abnormal pain identification results based on the preliminary investigation data; A preliminary investigation result is determined based on the abnormal bleeding identification result, the abnormal discharge identification result, and the abnormal pain identification result.

3. The method for detecting endometrial cancer based on deep learning according to claim 1, characterized in that: Determining the patient's physiological state influence coefficient according to the patient's physiological state data includes: Determining the patient's age, BMI, disease history, cytotoxic T cell density, helper regulatory T cell ratio, and macrophage ratio based on the patient's physiological status data; Obtaining diabetes identification results and cardiovascular disease identification results based on the patient's medical history; Determining a patient age risk identification result according to the patient age; The patient's physiological state influence coefficient is determined based on the patient's age risk identification result, the patient's BMI, the toxic T cell density, the helper regulatory T cell ratio, the macrophage ratio, the diabetes identification result, and the cardiovascular disease identification result.

4. The method for detecting endometrial cancer based on deep learning according to claim 3, characterized in that: Determining the patient's physiological state influence coefficient according to the patient's age risk identification result, the patient's BMI, the cytotoxic T cell density, the helper regulatory T cell ratio, the macrophage ratio, the diabetes identification result, and the cardiovascular disease identification result, including: According to the formula Determine the patient physiological status influence coefficient of the i-th patient to be tested , where if is a conditional function and and is the logical operator of "and". is the diabetes identification result of the i-th patient to be tested, , is the BMI of the i-th patient to be tested, is the patient age risk identification result of the i-th patient to be tested, , is the cardiovascular disease identification result of the i-th patient to be tested, , is the cytotoxic T cell density of the i-th patient to be tested, To preset the cytotoxic T cell density threshold, is the ratio of helper regulatory T cells of the i-th patient to be tested, To preset the threshold of the ratio of helper regulatory T cells, is the macrophage ratio of the i-th patient to be tested, The preset macrophage ratio threshold.

5. The method for detecting endometrial cancer based on deep learning according to claim 1, characterized in that: The training steps of the tumor progression time prediction model include: Obtain historical patient physiological status data, historical medical imaging data, and historical laboratory test data for multiple historical patients; Determining an influence coefficient of the historical patient physiological state based on the historical patient physiological state data; Based on the historical medical imaging data, determine the historical degree of myometrial invasion, historical ADC value, and historical depth of cervical stromal invasion; Based on the historical laboratory test data, determine the historical histological type risk level and historical lymphovascular space infiltration identification results; The historical patient physiological state influence coefficient, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space infiltration identification result are processed by a tumor deterioration time prediction model to obtain a historical training deterioration time; Obtain the historical actual deterioration time of multiple historical patients; Determining a training loss function for a tumor deterioration time prediction model based on the historical patient physiological state influence coefficient, the historical training deterioration time, the historical actual deterioration time, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space invasion identification result; The tumor exacerbation time prediction model is trained according to the training loss function to obtain a trained tumor exacerbation time prediction model.

6. The method for detecting endometrial cancer based on deep learning according to claim 5, characterized in that: Determining a training loss function for a tumor deterioration time prediction model based on the historical patient physiological state influence coefficient, the historical training deterioration time, the historical actual deterioration time, the historical myometrial invasion degree, the historical ADC value, the historical cervical stromal invasion depth, the historical histological type risk level, and the historical lymphovascular space invasion identification result includes: According to the formula Determine the training loss function of the tumor progression time prediction model ,in, is the historical training deterioration time of the kth historical patient, is the actual historical deterioration time of the kth historical patient, is the historical ADC value of the kth historical patient, To preset the ADC value threshold, is the historical muscle invasion degree of the kth historical patient, To preset the threshold of muscle invasion degree, is the historical cervical stromal invasion depth of the kth historical patient, To preset the historical cervical stromal invasion depth threshold, is the historical histological type risk level of the kth historical patient, To preset the histological type risk level threshold, is the historical lymphovascular space infiltration identification result of the kth historical patient, , is the influence coefficient of the physiological status of the kth historical patient, is the preset threshold of the patient's physiological status influence coefficient, K is the number of historical patients, k≤K, and both k and K are positive integers.

7. The method for detecting endometrial cancer based on deep learning according to claim 1, characterized in that: The medical imaging data, the laboratory test data, and the patient's physiological status influence coefficient are processed by a trained tumor exacerbation time prediction model to determine a predicted tumor exacerbation time, including: Determine the degree of myometrial invasion, ADC value, and depth of cervical stromal invasion based on the medical imaging data; Determine the histological risk level and lymphovascular space invasion identification results based on the laboratory test data; The patient's physiological state influence coefficient, the degree of myometrial invasion, the ADC value, the cervical stromal invasion depth, the histological type risk level and the lymphovascular space infiltration identification result are processed through the trained tumor deterioration time prediction model to obtain the predicted tumor deterioration time.

8. The method for detecting endometrial cancer based on deep learning according to claim 1, characterized in that: Based on the predicted tumor progression time, the risk factor for endometrial cancer is determined, including: determining a first difference according to the predicted tumor exacerbation time and a set tumor exacerbation time threshold; determining a first ratio according to the first difference and a set tumor progression time threshold; The risk factor for endometrial cancer is determined based on the first ratio.

9. A deep learning-based endometrial cancer detection system, characterized in that: include: Preliminary data module, to obtain preliminary survey data of patients to be tested; A preliminary result module, which determines preliminary investigation results based on the preliminary investigation data; A pathology detection module determines whether a tissue pathology test is needed based on the preliminary investigation results; Pathology result module, if a tissue pathology test is required, obtains the hysteroscopy-guided biopsy data of the patient to be tested, and determines the tissue pathology diagnosis result based on the hysteroscopy-guided biopsy data; A deep detection module determines whether further testing is needed based on the tissue pathology diagnosis results; Detailed data module, if further testing is required, to obtain the patient's physiological status data, medical imaging data and laboratory test data; An influence coefficient module, for determining an influence coefficient of the patient's physiological state according to the patient's physiological state data; A prediction time module processes the medical imaging data, the laboratory test data, and the patient's physiological status influence coefficient using a trained tumor exacerbation time prediction model to determine a predicted tumor exacerbation time; a risk coefficient module, which determines the risk coefficient of endometrial cancer based on the predicted tumor progression time; The test report module generates a test report based on the preliminary investigation results, the tissue pathology confirmation results and the endometrial cancer risk factor.

Images