System and method for predicting tumor bone metastasis based on bone metastasis-specific bone markers

Abstract

The application provides a tumor bone metastasis prediction system and method based on bone metastasis specific bone markers, relates to the technical field of machine learning, and comprises the following steps: an index acquisition module is used for acquiring a plurality of bone marker indexes related to bone metastasis in patient serum obtained by serum detection of a patient; and a bone metastasis prediction module is connected to the index acquisition module, used for inputting each bone marker index into a tumor bone metastasis prediction model trained in advance, so as to predict the early bone metastasis incidence of the patient, thereby providing a reference for clinical diagnosis by doctors. The beneficial effect is that the bone marker content of calcium and phosphorus metabolism regulation indexes, bone formation and bone resorption markers, bone metabolism related hormones and cytokines, and electrolytes in serum is quantitatively detected, and the early bone metastasis incidence is predicted through machine learning, so that the method can be used for early auxiliary diagnosis of tumor bone metastasis, auxiliary evaluation of curative effect in a treatment process and monitoring of metastasis recurrence after treatment, and the detection result can be earlier than clinical symptoms.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning, in particular to a tumor bone metastasis prediction system and method based on bone metastasis specific bone markers. BACKGROUND

[0002] The incidence of bone metastases is 35-40 times that of primary bone malignancies. Cancer bone metastasis is one of the main causes of cancer pain. Its complications such as pathological fracture, spinal cord compression, hypercalcemia and bone marrow failure accelerate the development of the disease, seriously affect the quality of life of cancer patients, and shorten the survival time of patients by more than half. In the past ten years, many disciplines have made unremitting efforts in the mechanism of bone metastases, prevention and treatment methods, etc. However, so far, an effective cure has not been found.

[0003] In current clinical practice, the diagnosis and risk assessment of tumor bone metastasis mainly rely on imaging examination and bone marrow biopsy evaluation. However, these traditional methods have some limitations and shortcomings. X-ray, CT, PET-CT, magnetic resonance imaging and other imaging examinations are commonly used diagnostic methods for tumor bone metastasis. However, imaging examinations have radiation damage to the human body, are not suitable for repeated examination in a short period of time, are expensive, and are easily affected by subjective factors such as the experience, technique and meticulousness of the examiner. More importantly, these methods may not accurately detect early bone metastasis lesions, especially small bone metastasis lesions, leading to misdiagnosis or missed diagnosis. Bone marrow puncture is often difficult for patients to accept due to invasiveness and sampling bias. Although it has certain diagnostic value in some cases, bone marrow puncture lacks sensitivity for early or small bone metastasis and cannot accurately determine whether a tumor patient has bone metastasis and the early risk of bone metastasis. In summary, the existing clinical tools have certain limitations and shortcomings in the early diagnosis and risk assessment of tumor bone metastasis. Therefore, a new technical method needs to be developed to improve the accuracy of early diagnosis of tumor bone metastasis and clinical treatment support. SUMMARY

[0004] In view of the problems in the prior art, the present application provides a tumor bone metastasis prediction system based on bone metastasis specific bone markers, comprising:

[0005] An index acquisition module is configured to acquire a plurality of bone marker indexes related to bone metastasis in the serum of a patient obtained by serum detection of the patient;

[0006] A bone metastasis prediction module is connected to the index acquisition module and configured to input each of the bone marker indexes into a pre-trained tumor bone metastasis prediction model to predict the early bone metastasis incidence of the patient for reference by a doctor in clinical diagnosis.

[0007] Preferably, the bone marker indicators include calcium and phosphorus metabolism regulation indicators, bone formation and bone resorption marker indicators, bone metabolism related hormone and cytokine indicators, and electrolyte indicators.

[0008] Preferably, the tumor bone metastasis prediction model is at least one of an extreme gradient boosting model and a random forest model.

[0009] Preferably, the tumor bone metastasis prediction system further comprises a prompt generation module connected to the bone metastasis prediction module, configured to generate corresponding prompt information when the early bone metastasis incidence output by any of the tumor bone metastasis prediction models is greater than a preset threshold, so as to prompt the doctor and / or the patient to perform further examination.

[0010] Preferably, the tumor bone metastasis prediction system further comprises a report generation module connected to the bone metastasis prediction module, configured to automatically generate a tumor bone metastasis prediction report of the patient, wherein the tumor bone metastasis prediction report comprises basic information of the patient, the bone marker indicators, and the corresponding early bone metastasis incidence.

[0011] The application further provides a tumor bone metastasis prediction method based on bone metastasis specific bone markers, applied to the tumor bone metastasis prediction system.

[0012] Step S1: obtaining a plurality of bone marker indicators related to bone metastasis in patient serum obtained through serum detection of the patient;

[0013] Step S2: inputting the bone marker indicators into a pre-trained tumor bone metastasis prediction model to predict the early bone metastasis incidence of the patient for reference by the doctor in clinical diagnosis.

[0014] Preferably, the bone marker indicators include calcium and phosphorus metabolism regulation indicators, bone formation and bone resorption marker indicators, bone metabolism related hormone and cytokine indicators, and electrolyte indicators.

[0015] Preferably, the tumor bone metastasis prediction model is at least one of an extreme gradient boosting model and a random forest model.

[0016] Preferably, the tumor bone metastasis prediction method further comprises the following steps after step S2:

[0017] When the early bone metastasis incidence output by any of the tumor bone metastasis prediction models is greater than a preset threshold, corresponding prompt information is generated to prompt the doctor and / or the patient to perform further examination.

[0018] Preferably, the tumor bone metastasis prediction method further comprises the following steps after step S2:

[0019] automatically generating a tumor bone metastasis prediction report of the patient, the tumor bone metastasis prediction report including basic information of the patient, each of the bone marker indexes and the corresponding early bone metastasis occurrence rate.

[0020] The above technical solution has the following advantages or beneficial effects:

[0021] 1) By quantitatively detecting the contents of bone markers such as calcium and phosphorus metabolism regulation indicators, bone formation and bone resorption markers, bone metabolism related hormones and cytokines, and electrolytes, and early predicting the bone metastasis occurrence rate through machine learning, the early auxiliary diagnosis of tumor bone metastasis, the auxiliary evaluation of treatment effect during treatment and the monitoring of metastasis recurrence after treatment can be realized, and the detection results can be earlier than clinical symptoms.

[0022] 2) It has the advantages of good specificity, high sensitivity, etc., and the detection technology involved is mature, and compared with traditional methods, it has short cycle and good specificity. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 In a preferred embodiment of the present application, a structure diagram of a tumor bone metastasis prediction system based on bone metastasis specific bone markers is provided.

[0024] Figure 2 In a preferred embodiment of the present application, a flowchart of a tumor bone metastasis prediction method based on bone metastasis specific bone markers is provided.

[0025] Figure 3 In a preferred embodiment of the present application, a diagram of the Shapley value of each bone marker index of the RF model is provided.

[0026] Figure 4 In a preferred embodiment of the present application, a diagram of the ROC curve corresponding to the RF model is provided.

[0027] Figure 5 In a preferred embodiment of the present application, a diagram of the Shapley value of each bone marker index of the XGB model is provided.

[0028] Figure 6 In a preferred embodiment of the present application, a diagram of the ROC curve corresponding to the XGB model is provided.

[0029] Figure 7 In a preferred embodiment of the present application, a diagram of a machine learning online calculation webpage is provided. DETAILED DESCRIPTION

[0030] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. The present application is not limited to this embodiment, and other embodiments can also fall within the scope of the present application as long as they comply with the spirit of the present application.

[0031] In a preferred embodiment of the present invention, based on the above-mentioned problems existing in the prior art, a tumor bone metastasis prediction system based on bone metastasis-specific bone markers is provided, such as... Figure 1 As shown, it includes:

[0032] The indicator acquisition module 1 is used to acquire multiple bone marker indicators related to bone metastasis in the patient's serum obtained from serum testing.

[0033] Bone metastasis prediction module 2, connected to indicator acquisition module 1, is used to input various bone marker indicators into the pre-trained tumor bone metastasis prediction model to predict the early bone metastasis incidence rate of patients for doctors to use as a reference for clinical diagnosis.

[0034] Specifically, in this embodiment, bone biomarkers include calcium and phosphorus metabolism regulation indicators, bone formation and bone resorption biomarkers, bone metabolism-related hormones and cytokines, and electrolyte indicators.

[0035] Furthermore, indicators of calcium and phosphorus metabolism regulation include calcium (Ca), phosphorus (P), parathyroid hormone (PTH), and calcitonin (CT). Calcium (Ca) and phosphorus (P) are the main components of bone tissue, working together to maintain bone stability. Tumor bone metastases disrupt the calcium-phosphorus balance in bone; bone destruction releases calcium and phosphorus ions into the bloodstream, leading to elevated serum calcium and phosphorus levels. Parathyroid hormone (PTH) is a hormone that regulates calcium and phosphorus metabolism, affecting serum calcium and phosphorus levels by modulating renal reabsorption of calcium and phosphorus. In tumor bone metastases, destroyed bone releases more calcium ions, stimulating the parathyroid glands to secrete PTH, resulting in elevated serum PTH levels. Calcitonin (CT) is a hormone secreted by the parathyroid glands that regulates serum calcium and phosphorus levels by inhibiting the release of calcium ions from bone. In tumor bone metastases, the calcium ions released from destroyed bone reduce calcitonin secretion, leading to decreased serum calcitonin levels.

[0036] Bone formation and bone resorption marker indicators include BAP, tP1NP, OPG, β-CTx. Among them, bone formation alkaline phosphatase (BAP) is a marker for evaluating bone formation. In bone metastasis, the destruction of bone can inhibit the bone formation process, resulting in a decrease in serum BAP level. Total 1 type collagen amino terminal extension peptide (tPINP) is another marker of bone formation, which reflects the degree of collagen synthesis. Bone metastasis can interfere with the normal process of bone formation, resulting in a decrease in serum tP1NP level. Osteoprotegerin (OPG) is a protein that inhibits bone resorption process, which inhibits the activity of bone resorption cells by binding to RANKL. In bone metastasis, tumor cells may produce excess OPG, resulting in decreased bone resorption and elevated serum OPG levels, which can be used as a diagnostic marker for lung cancer, prostate cancer, breast cancer and other bone metastases. β-collagen degradation product (β-CTX): in the process of bone resorption, CTX is released into the blood with the degradation of type I collagen molecules, so the concentration of CTX in the blood can specifically reflect the level of bone resorption, and β-CTX is confirmed as a sensitive and specific bone resorption marker. In bone metastasis, the destruction of bone can lead to increased release of β-CTx, resulting in elevated serum β-CTx levels.

[0037] Bone metabolism related hormone and cytokine indicators include FT3, FT4, TSH, IL-6, PTHrP. Among them, free triiodothyronine (FT3) and free thyroxine (FT4) are two forms of thyroid hormone, which participate in the regulation of bone metabolism. In bone metastasis, bone destruction can interfere with the normal function of thyroid hormone, leading to changes in serum FT3 and FT4 levels. Thyroid stimulating hormone (TSH) is a hormone that regulates the secretion of thyroid hormone, which stimulates the thyroid gland to synthesize and release thyroid hormone. In bone metastasis, tumor cells may produce certain substances that interfere with the normal regulation of TSH, leading to changes in serum TSH levels. Interleukin-6 (IL-6) is a cytokine that participates in the regulation of inflammation and bone metabolism. In bone metastasis, IL-6 produced by tumor cells can promote bone destruction and tumor growth, resulting in elevated serum IL-6 levels. Parathyroid hormone-related protein (PTHrP) is a protein produced by tumor cells, which has similar effects to PTH and can increase the concentration of calcium ions in the blood. In bone metastasis, PTHrP produced by tumor cells can interfere with the balance of calcium and phosphorus metabolism, leading to elevated serum calcium levels.

[0038] The electrolyte indicators include K, Na, Mg. Among them, potassium (K), sodium (Na) and magnesium (Mg) are electrolytes in the body, which play an important role in maintaining bone health and normal bone metabolism. In bone metastasis, the balance of electrolytes may be affected. For example, the calcium ions released by bone destruction can affect the normal function of the sodium-potassium pump, leading to changes in serum potassium and sodium levels. In addition, magnesium is an important component of bone, and bone destruction can affect serum magnesium levels.

[0039] In this embodiment, the above-mentioned 16 bone marker indicators are all used as inputs of the tumor bone metastasis prediction model, which overcomes the defect that the sensitivity and specificity of a single indicator cannot meet the requirements of clinical diagnosis.

[0040] Further specifically, the content of the 16 bone marker indicators in the serum of the patient can be detected by a test paper or a test box. The specific detection method of the 16 bone marker indicators is as follows:

[0041] 1. The concentration of PTHrP and OPG is detected by double antibody sandwich enzyme-linked immunosorbent assay (ELISA):

[0042] First, the PTHrP or OPG standard and the test sample are added to the pre-coated PTHrP or OPG antibody microplate, and then another enzyme-labeled PTHrP or OPG antibody is added. After incubation and thorough washing, the unbound components are removed. Then, a sandwich complex is formed on the solid phase surface of the microplate, including the structure of solid phase antibody-antigen-HRP labeled antibody. By adding a chromogenic substrate, the substrate is catalyzed by HRP to produce a blue product, which is finally converted to yellow under the action of a termination solution. The concentration of PTHrP or OPG in the sample can be calculated by measuring the absorbance (OD value) at 450 nm wavelength using an enzyme label instrument. By fitting the standard curve, the concentration of PTHrP or OPG in the sample can be calculated.

[0043] 2. BAP is determined by one-step enzyme immunoassay:

[0044] First, the BAP-specific mouse monoclonal antibody is added to the reaction tube containing goat anti-mouse polyclonal antibody-coated paramagnetic particles, and then the standard, quality control and sample containing BAP are added to the coated particles and combined with the anti-BAP monoclonal antibody. With the formation of the solid phase / capture antibody / BAP complex, the material combined with the solid phase will be adsorbed in the magnetic field, and the unbound material will be washed away. Next, the chemiluminescent substrate Lumi Phos 530 is added to the reaction tube, and the amount of light generated is measured using a luminometer. Finally, the amount of light generated is proportional to the concentration of BAP in the sample.

[0045] 3. Electrochemiluminescence method is used to detect β-CTx, TSH, tP1NP, NMID, PTH, CT, IL-6, FT4 and FT3 in serum:

[0046] The principle is that the sample to be measured, paramagnetic particles coated with antibodies and antibodies labeled with luminescent agents are added to the reaction cup, and are incubated to form a complex. The complex material is sucked into the flow chamber, and the magnetic microspheres are adsorbed on the electrode. The unbound substances are washed away by the cleaning solution, and the electrode generates chemiluminescence after a voltage is applied.

[0047] 4. Ion selective electrode method is used to detect K, Na, Ca and Mg in serum:

[0048] The principle is that the sensing film selectively responds to a certain specific ion in the mixed solution, similar to the working principle of the pH electrode, and the potential between the ion-selective membrane and the sample solution changes with the change of the activity of the ion to be measured.

[0049] 5. Spectrophotometry is used to detect the content of P in serum:

[0050] The principle is that after removing organic phosphorus in serum, inorganic phosphorus salt and ammonium molybdate reagent generate phosphomolybdate, which is reduced to blue by ferrous sulfate and has light absorption at 620 nm. The phosphorus content in blood is calculated by measuring the absorbance at 620 nm.

[0051] In the implementation process of the present application, specifically includes:

[0052] Step (A): 4ml of peripheral blood of the person to be measured is drawn into a BD coagulation tube, and serum is obtained by centrifugation for direct detection or freezing at-80℃ for detection;

[0053] Step (B): The above serum is detected by using the above corresponding detection technology to obtain the concentration content of 16 bone markers in serum.

[0054] Among them, the processing of peripheral blood of the person to be measured in step (A): when collecting the serum sample, the blood of the person to be measured should be left to stand for at least 30 minutes after being drawn, and then centrifuged at 1000xg for 10 minutes to collect the serum. The number of freeze-thaw cycles of the collected plasma should be less than 3 times.

[0055] The detection technology of bone markers used in step (B). According to the specification and standardized process of the corresponding technology, the whole experimental process must be strictly carried out according to the process.

[0056] As a preferred embodiment, the tumor bone metastasis prediction system of the present application adopts a machine learning online calculation webpage in the form of a webpage for doctors to use, wherein the doctors can input the detected 16 bone marker indicators into the machine learning online calculation webpage to predict the corresponding bone metastasis occurrence rate. More preferably, the machine learning online calculation webpage can also be directly connected to the hospital information management system to automatically call the 16 bone marker indicators of the corresponding patient from the information management system, and then predict the corresponding bone metastasis occurrence rate, which is not limited here.

[0057] In a preferred embodiment of the present application, the tumor bone metastasis prediction model is at least one of an extreme gradient boosting model and a random forest model.

[0058] In a preferred embodiment of the present application, a prompt generation module 3 is further included, connected to the bone metastasis prediction module 2, for generating corresponding prompt information when the early bone metastasis occurrence rate output by any tumor bone metastasis prediction model is greater than a preset threshold, to prompt the doctor and / or patient to perform further examination.

[0059] Specifically, in the present embodiment, the tumor bone metastasis prediction model is one, and as long as the early bone metastasis occurrence rate output by the tumor bone metastasis prediction model is greater than the preset threshold, it indicates that the current patient has a high risk of bone metastasis, and corresponding prompt information is generated, otherwise it indicates that the current patient has a low risk of bone metastasis, and a low-risk prompt can also be generated accordingly. When the tumor bone metastasis prediction model is two, i.e., when it includes an extreme gradient boosting model and a random forest model, the two models complement each other, and as long as the early bone metastasis occurrence rate output by one of the models is greater than the preset threshold, it indicates that the current patient has a high risk of bone metastasis, and corresponding prompt information is generated, otherwise it indicates that the current patient has a low risk of bone metastasis, and a low-risk prompt can also be generated accordingly.

[0060] In a preferred embodiment of the present application, a report generation module 4 is further included, connected to the bone metastasis prediction module 2, for automatically generating a tumor bone metastasis prediction report of the patient, wherein the tumor bone metastasis prediction report includes the basic information of the patient, the bone marker indicators, and the corresponding early bone metastasis occurrence rate.

[0061] Specifically, in the present embodiment, through the tumor bone metastasis prediction report, the doctor or patient can directly view the specific values of the bone marker indicators and the corresponding early bone metastasis occurrence rate, and a risk prompt can also be provided to assist the doctor in giving further clinical diagnosis.

[0062] The present application also provides a tumor bone metastasis prediction method based on bone metastasis-specific bone markers, which is applied to the tumor bone metastasis prediction system described above, as shown in Figure 2 The tumor bone metastasis prediction method includes:

[0063] Step S1, obtaining a plurality of bone marker indicators related to bone metastasis in the serum of the patient obtained by serum detection of the patient;

[0064] Step S2, inputting each bone marker indicator into a pre-trained tumor bone metastasis prediction model to predict the early bone metastasis incidence of the patient for the doctor to make a clinical diagnosis reference.

[0065] In a preferred embodiment of the present application, the bone marker indicators include calcium and phosphorus metabolism regulation indicators, bone formation and bone resorption marker indicators, bone metabolism related hormone and cytokine indicators, and electrolyte indicators.

[0066] In a preferred embodiment of the present application, the tumor bone metastasis prediction model is at least one of an extreme gradient boosting model and a random forest model.

[0067] In a preferred embodiment of the present application, after step S2 is performed, the method further comprises:

[0068] When the early bone metastasis incidence output by any tumor bone metastasis prediction model is greater than a preset threshold, corresponding prompt information is generated to prompt the doctor and / or the patient to perform further examination.

[0069] In a preferred embodiment of the present application, after step S2 is performed, the method further comprises:

[0070] Automatically generating a tumor bone metastasis prediction report of the patient, wherein the tumor bone metastasis prediction report includes the basic information of the patient, each bone marker indicator, and the corresponding early bone metastasis incidence.

[0071] As a preferred embodiment of the present application, 123 healthy controls, 339 lung cancer patients without bone metastasis, and 331 lung cancer patients with bone metastasis were collected from July 2015 to October 2020, a total of 508 serum samples (from Shanghai No. 6 People's Hospital) were detected for 16 bone marker indicators, including:

[0072] Peripheral blood of the to-be-tested person is extracted, and serum is obtained by centrifugation, and direct detection or freezing at-80℃ for detection is performed;

[0073] The above serum is detected by using the corresponding detection technology, the concentration of PTHrP and OPG is detected by using double antibody sandwich enzyme-linked immunosorbent assay (ELISA), BAP is detected by using one-step enzyme immunoassay, and the concentration of β-CTx, TSH, tP1NP, PTH, CT, IL-6, FT4 and FT3 in the serum is detected by electrochemiluminescence method. The ion selective electrode method is used to detect the concentration of K, Na, Ca and Mg in the serum, and the spectrophotometric method is used to detect the content of P in the serum.

[0074] The serum samples of 123 healthy controls were set as group 0, the serum samples of 339 lung cancer patients without bone metastasis were set as group 1, and the serum samples of 331 lung cancer patients with bone metastasis were set as group 2. The statistical analysis of 18 indexes including age and gender was as follows:

[0075] (a) In the first round of screening, each index was compared between groups, T-test was used, the median and quartile of each marker were listed, and the p value of comparison between groups was calculated. The comparison results of each index are shown in Table 1 as follows:

[0076] Table 1 Comparison results of each index

[0077]

[0078] (b) According to the analysis of 16 bone marker indexes in the last round, the ROC curve of each single index was drawn. The diagnostic value of each bone marker index for lung cancer bone metastasis was analyzed as shown in Table 2, wherein the area AUC of the ROC curve of the related index, the 95% CI of AUC, the sensitivity, the specificity, the negative predictive value, the positive predictive value, the accuracy, the Youden index and the Cut-off value represent the diagnostic value of each bone marker index for lung cancer bone metastasis.

[0079] The indexes include 16 bone markers such as calcium and phosphorus metabolism regulation indexes (Ca, P, PTH, CT), bone formation and bone resorption markers (BAP, tP1NP, OPG, β-CTx), bone metabolism related hormones and cytokines (FT3, FT4, TSH, IL-6, PTHrP) and electrolytes (K, Na, Mg).

[0080] Table 2 Diagnostic value analysis of lung cancer bone metastasis

[0081]

[0082]

[0083] In the model training, a total of 670 patient detection data were included, including 339 lung cancer patients without bone metastasis and 331 lung cancer patients with bone metastasis. From the two groups, 70% were extracted as the training set (469 cases), and the remaining 30% were extracted as the test set (201 cases).

[0084] The random forest RF model and XGB model of machine learning were used to determine whether the lung cancer patients had bone metastasis, and the sensitivity, specificity and accuracy were calculated as the evaluation indexes of the classification model. Finally, the report generation model will automatically generate a diagnosis report, and the accuracy and timeliness of the report will be used as the evaluation index of the report generation model.

[0085] Wherein, the Shapley value of each bone marker index when adopting the random forest model is as shown in the following table 1: Figure 3 As shown in the following table 2, in the 201 test sets, the sensitivity, specificity and accuracy of the random forest (RF) model are 86%, 91% and 89% respectively, and the corresponding model ROC curve is as shown in the following table 3: Figure 4

[0086] The Shapley value of each bone marker index when adopting the extreme gradient boosting model is as shown in the following table 4: Figure 5 As shown in the following table 5, in the 201 test sets, the sensitivity, specificity and accuracy of the extreme gradient boosting (XGB) model are 88%, 84% and 86% respectively, and the corresponding model ROC curve is as shown in the following table 6: Figure 6

[0087] Under the normal network environment, the machine learning model in the application takes about several seconds from the original data input to the final report generation, which has obvious timeliness compared with manual interpretation (many indicators cannot be given a conclusion by manual interpretation).

[0088] In order to analyze the value of each model in the early diagnosis of lung cancer bone metastasis, we analyzed 316 lung cancer patients without bone metastasis in the follow-up study, of which 123 patients developed bone metastasis during the follow-up period, and the bone metastasis rate was 38.92%. We used the RF model and the XGB model to predict the serum bone marker detection results of the patients when they were diagnosed with lung cancer without bone metastasis for the first time.

[0089] According to the best critical value (cut off value) of the model prediction (the average threshold value of the two algorithms = 40%), the risk stratification is carried out. The patients with risk probability lower than the best critical value are classified into the low-risk group, and the patients with risk probability higher than the critical value are classified into the high-risk group.

[0090] The prediction results are shown in the following table 3:

[0091] Table 3: Risk prediction of RF model and XGB model in follow-up study

[0092]

[0093] It can be seen that, compared with the high-risk group patients, the incidence of bone metastasis in the high-risk group patients is 3-5 times or more than that in the low-risk group patients (P<0.001). The study also calculates the matched actual and expected probabilities.

[0094] The consistency analysis of predicting bone metastasis and actually occurring bone metastasis shows that the prediction consistency of the RF model and the XGB model is the highest, and the Kappa values are 0.87 and 0.83 respectively.

[0095] ​​Further analysis found that, according to the initial diagnosis data of patients to predict lung cancer bone metastasis, to the patient diagnosed with bone metastasis, the RF model and the XGB model can be earlier than the average of 10.27±3.58 months and 10.35±3.64 months of imaging.

[0096] This shows that the RF model and the XGB model have important value in the early diagnosis of lung cancer bone metastasis. Patients in the high-risk group are recommended to perform bone scanning or further examination. Patients in the low-risk group are recommended to perform less imaging examination.

[0097] Further, the machine learning online calculation webpage of the present application is shown as Figure 7 As shown, doctors and patients can perform online calculation through the webpage.

[0098] In summary, the present application can detect the content of 16 bone markers such as calcium and phosphorus metabolism regulation indicators (Ca, P, PTH, CT), bone formation and bone resorption markers (BAP, tP1NP, OPG, β-CTx), bone metabolism related hormones and cytokines (FT3, FT4, TSH, IL-6, PTHrP) and electrolytes (K, Na, Mg) in the serum of patients, and assist doctors in early diagnosis of bone metastasis of solid tumors such as lung cancer through the online calculation webpage based on the RF model and the XGB model. The detection method is mature, simple, short cycle and high sensitivity. In the test set, the serum content of 16 bone markers is detected, and the diagnostic accuracy of the RF model and the XGB model reaches 86% and 89%; according to the initial diagnosis data of patients to predict lung cancer bone metastasis, to the patient diagnosed with bone metastasis, the RF model and the XGB model can be earlier than the average of 10.27±3.58 months and 10.35±3.64 months of imaging, so as to effectively supplement the existing indicators.

[0099] The above only describes the preferred embodiments of the present application, and does not limit the implementation and protection scope of the present application. Those skilled in the art should be able to realize that any equivalent replacement and obvious changes made according to the present application and drawings should be included in the protection scope of the present application.

Claims

1. A bone metastasis prediction system based on bone metastasis-specific bone markers, characterized by, The system comprises: an index acquisition module, configured to acquire a plurality of bone marker indexes related to bone metastasis in serum of a patient obtained by serum detection on the patient; a bone metastasis prediction module, connected to the index acquisition module, configured to input each of the bone marker indexes into a pre-trained tumor bone metastasis prediction model to predict an early bone metastasis incidence of the patient for reference by a doctor in clinical diagnosis; the bone marker indexes comprise calcium and phosphorus metabolism regulation indexes, bone formation and bone resorption marker indexes, bone metabolism related hormone and cytokine indexes, and electrolyte indexes; the calcium and phosphorus metabolism regulation indexes comprise calcium (Ca), phosphorus (P), parathyroid hormone (PTH), and calcitonin (CT), the bone formation and bone resorption marker indexes comprise bone-specific alkaline phosphatase (BAP), total type I procollagen amino-terminal propeptide (tP1NP), osteoprotegerin (OPG), and β-type I collagen carboxy-terminal peptide (β-CTx), the bone metabolism related hormone and cytokine indexes comprise free triiodothyronine (FT3), free thyroxine (FT4), thyroid-stimulating hormone (TSH), parathyroid hormone-related protein (PTHrP), and interleukin-6 (IL-6), and the electrolyte indexes comprise potassium (K), sodium (Na), and magnesium (Mg); wherein the system is applied to prediction of lung cancer bone metastasis; the tumor bone metastasis prediction model is at least one of an extreme gradient boosting model and a random forest model.

2. The tumor bone metastasis prediction system of claim 1, wherein, The system further comprises a prompt generation module, connected to the bone metastasis prediction module, configured to generate corresponding prompt information to prompt the doctor and / or the patient to perform further examination when the early bone metastasis incidence output by any of the tumor bone metastasis prediction models is greater than a preset threshold.

3. The tumor bone metastasis prediction system of claim 1, wherein, The system further comprises a report generation module, connected to the bone metastasis prediction module, configured to automatically generate a tumor bone metastasis prediction report of the patient, wherein the tumor bone metastasis prediction report comprises basic information of the patient, each of the bone marker indexes, and the corresponding early bone metastasis incidence.

4. A method for predicting tumor bone metastasis based on bone metastasis-specific bone markers, characterized by, The tumor bone metastasis prediction method comprises: step S1, acquiring a plurality of bone marker indexes related to bone metastasis in serum of a patient obtained by serum detection on the patient; step S2, inputting each of the bone marker indexes into a pre-trained tumor bone metastasis prediction model to predict an early bone metastasis incidence of the patient for reference by a doctor in clinical diagnosis.

5. The method of claim 4, wherein the tumor bone metastasis prediction method is characterized by, After the step S2 is performed, the method further comprises: generating corresponding prompt information to prompt the doctor and / or the patient to perform further examination when the early bone metastasis incidence output by any of the tumor bone metastasis prediction models is greater than a preset threshold.

6. The method of claim 4, wherein the tumor bone metastasis prediction method is characterized by, After the step S2 is performed, the method further comprises: automatically generating a tumor bone metastasis prediction report of the patient, wherein the tumor bone metastasis prediction report comprises basic information of the patient, each of the bone marker indexes, and the corresponding early bone metastasis incidence.

Images