CN112037919A - Risk assessment model for papillary carcinoma of thyroid nodule patient - Google Patents

Abstract

The invention discloses a risk assessment model for papillary carcinoma of patients with thyroid nodules, which is characterized by comprising: using Logistic regression to analyze the risk factors of PTC, selecting a specific index as an independent variable of the model; constructing an input layer including an input layer BP neural network model with a three-layer feedforward neural network structure, hidden layer and output layer, use the data including the above-mentioned specific indicators to train and predict it; compare the predicted value output by the model with the actual value, and draw ROC curve, the area under the curve AUC is obtained, and the AUC value is used to evaluate the diagnostic performance of the model. The model whose diagnostic performance meets the standard is the target risk assessment model. The risk assessment model of the invention can be used as a reference for doctors to diagnose papillary thyroid cancer, and the accuracy of diagnosis of papillary thyroid cancer can be improved.

Description

A risk assessment model for papillary carcinoma in patients with thyroid nodules

technical field

The invention relates to a risk assessment model constructed based on a neural network model, which is specifically used for the risk assessment of papillary carcinoma in patients with thyroid nodules.

Background technique

In recent years, people in the field generally believe that the popularization of thyroid disease screening and the advancement of detection technology are the main reasons for the significant increase in the incidence data of thyroid cancer that we have obtained. Among the known incidence data of thyroid cancer, most of them are papillary thyroid cancer (PTC) with better prognosis. The increase in a large number of PTC cases not only increases the medical burden, but also increases the possibility of overtreatment. In terms of preoperative diagnosis of thyroid cancer, both domestic and international guidelines recommend neck ultrasound as the first choice for all patients with known or suspected thyroid nodules, which can clearly demonstrate the nodule's echogenic texture, edge morphology, calcification, internal and surrounding thyroid nodules. Blood flow, growth trend and other characteristics, but its diagnostic accuracy is not enough to be used as an independent diagnostic method, patients suspected of thyroid cancer need follow-up cytopathological examination of thyroid nodule fine needle aspiration biopsy, as an invasive examination, its There are certain operational risks and the probability of puncture failure, and there are some cases that cannot be clearly diagnosed by cytology. Previous studies have shown that in addition to imaging examinations, the general condition of patients and some serological indicators are also correlated with the benign and malignant nodules, but because of their low independent diagnostic performance, they are often only used as clinical reference indicators.

SUMMARY OF THE INVENTION

The technical purpose of the present invention is to provide a risk assessment model for papillary carcinoma in patients with thyroid nodules, and the output index can be used as a basis for assisting doctors in making a diagnosis, thereby improving the diagnostic accuracy of papillary thyroid carcinoma in patients.

The technical scheme of the present invention is:

A risk assessment model for papillary carcinoma in patients with thyroid nodules, characterized by comprising the following steps:

S1. Collect the clinical data of the patients, analyze the risk factors of PTC by Logistic regression, select age, gender, systolic blood pressure, thyroid stimulating hormone, thyroid autoantibodies, thyroglobulin, thyroid ultrasound TI-RADS classification, apolipoprotein A-I, The 9 indexes of albumin are used as independent variables of the model, and the classification indexes among the 9 indexes are assigned, and the classification indexes include gender, thyroid autoantibodies and thyroid ultrasound TI-RADS classification;

S2, construct a BP neural network model, the BP neural network model includes a three-layer feedforward neural network structure, which are an input layer, a hidden layer and an output layer, and the input indicators of the input layer are the 9 indicators selected in step S1 , the output index of the output layer is the prediction result of papillary carcinoma in patients with thyroid nodules;

S3. Collect the clinical data of the patient, create a training set and a prediction set, and input the data of the training set into the BP neural network model constructed in step S2 for training; after the training is completed, input the data of the prediction set into the model, and output the prediction of the model The value is compared with the real value, and the ROC curve is drawn to obtain the area under the curve AUC. The AUC value is used to evaluate the diagnostic performance of the model. The model whose diagnostic performance meets the standard is the target risk assessment model.

On the basis of the above scheme, further improved or preferred schemes also include:

Further, in step S1, the assignment of the classification index is:

Gender: 1 for males, 0 for females;

Thyroid autoantibodies: if any one of TgAb and TPOAb is positive, it is positive (+) and assigned as 1; if both are negative, it is negative (-) and assigned as 2;

Thyroid ultrasound TI-RADS classification: 0 for 2; 1 for 3; 2 for 4a; 3 for 4b; 4 for 4c; 5 for 5;

In step S2, the output index of the output layer is assigned as follows: benign nodule is 1, and papillary thyroid carcinoma is 2.

Preferably, in step S2, the number of hidden layers of the constructed BP neural network model is 1, wherein the number of neuron nodes is 15, the training function is the Levenberg-Marquardt function, and the weight update method of error back propagation is the gradient descent method, The number of model iterations is 1000 and the learning rate is 0.01.

Beneficial effects:

Although the current preoperative non-invasive diagnostic methods for papillary thyroid cancer are abundant, they still have their own shortcomings in terms of independent diagnostic performance. In the past, physicians often used ultrasound results as the main reference, combined with the patient's general information and serological indicators to give the next step. medical opinion. The present invention constructs a kind of objective preoperative non-invasive diagnostic model (data processing model) with some serological indexes and general data of the patient together with the ultrasound results, makes full use of the clinical data of the patient, can be used as a reference to assist the doctor in making a diagnosis, and increases the Accuracy of preoperative diagnosis. Moreover, the application of this model does not require additional examinations, nor does it significantly increase the medical cost of patients, so it is suitable for promotion. The BP neural network model belongs to the category of machine learning. The richer the patient data provided for training, the more accurate the model prediction obtained.

Description of drawings

Fig. 1 is the structural representation of BP neural network model of the present invention;

FIG. 2 is a ROC curve of a BP neural network prediction model in an embodiment of the present invention.

Detailed ways

In order to clarify the technical solution and working principle of the present invention, the present invention will be further introduced below with reference to the accompanying drawings and specific embodiments.

A risk assessment model for papillary carcinoma in patients with thyroid nodules, comprising the following steps:

S1. Collect clinical data of patients, including ① age, gender and other general information; ② thyroid-related hormone and antibody test results; ③ biochemical test results; ④ thyroid nodule ultrasound TI-RADS grading results, etc. The present invention uses Logistic regression to analyze the risk factors of PTC, and selects age (years), gender (male/female), systolic blood pressure (mmHg), thyroid stimulating hormone (mIU/L), thyroid autoantibodies (positive/negative), Nine indicators including thyroglobulin (nmol/L), thyroid ultrasound TI-RADS grade, apolipoprotein A-I (g/L), and albumin (g/L) were used as independent variables of the model, and the nine indicators were The classification index in , the classification index includes gender, thyroid autoantibody and thyroid ultrasound TI-RADS grade, and the specific assignment is:

1) Gender: Male is assigned a value of 1, and female is assigned a value of 0;

2) Thyroid autoantibodies: if any one of TgAb and TPOAb is positive, it is positive (+) and assigned as 1; if both of them are negative, it is negative (-) and assigned as 2;

3) Thyroid ultrasound TI-RADS grading: grade 2 is 0; grade 3 is 1; grade 4a is 2; grade 4b is 3; grade 4c is 4; grade 5 is 5;

S2. Build a BP neural network model with the neural network toolbox that comes with the MATLAB software. The BP neural network model includes a three-layer feedforward neural network structure, which are an input layer, a hidden layer and an output layer. The structure is shown in Figure 1. As shown, the input indicators of the input layer are the 9 indicators selected in step S1, the output indicators of the output layer are the prediction results of papillary carcinoma in patients with thyroid nodules, and the output indicators of the output layer are assigned as follows: benign nodules are 1, Papillary thyroid carcinoma is 2. The number of hidden layers in the BP neural network model is 1, the number of neuron nodes is 15, the training function is the Levenberg-Marquardt function, the weight update method of error back propagation is the gradient descent method, the number of model iterations is 1000, and the learning The rate is 0.01.

S3. Collect enough clinical data of patients to create a training set and a prediction set, and input the data of the training set into the BP neural network model constructed in step S2 for training; after the training is completed, input the data of the prediction set into the model, and use the The predicted value output by the model is compared with the actual value of the disease, and the ROC curve is drawn to obtain the area under the curve AUC. The AUC value is used to evaluate the diagnostic performance of the model. The model whose diagnostic performance meets the standard is the target risk assessment model. For models whose diagnostic performance is not up to standard, the accuracy of model prediction can be improved by increasing the patient data in the training set, that is, the more abundant the patient data provided for training, the more accurate the model prediction obtained.

In this example, a training set is created based on the clinical data of 1622 patients for learning, and the clinical data of 406 patients is used as a prediction set, and the result is predicted to obtain the ROC curve as shown in Figure 2, the area under the curve AUC=0.943, 95%CI0 .912 to 0.973, standard error = 0.016, P < 0.001. In the coordinate points of the ROC curve output by SPSS, the approximate value of the predicted value with the highest sensitivity + specificity of 1.7835 is also selected as the diagnostic cut point, and the sensitivity of the BP neural network model is 90.1%, specificity is 90.2%, NPV (negative Predictive value) was 75.4% and PPV (positive predictive value) was 96.5%. The prediction results of this model are compared with those of the traditional ultrasound TI-RADS grading independent diagnosis and multivariate Logistic regression combined diagnosis model for a total of 2028 patients to predict the properties of nodules. Table 1 shows that the prediction is accurate. The degree of diagnosis is higher than the diagnostic results of the traditional ultrasound and multivariate Logistic regression combined diagnostic model, and has high reliability.

Note: AUC represents the area under the receiver operating characteristic curve (ROC), with 95% confidence intervals in parentheses. Youden index = sensitivity + specificity -1.

The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above-mentioned embodiments, and the descriptions in the above-mentioned embodiments and the description are only to illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have Various changes and improvements, the claimed scope of the present invention is defined by the appended claims, description and their equivalents.

Claims (3)

1. a risk assessment model for thyroid nodule patient papillary carcinoma, is characterized in that, comprises the following steps: S1. Collect the clinical data of the patients, analyze the risk factors of PTC by Logistic regression, select age, gender, systolic blood pressure, thyroid stimulating hormone, thyroid autoantibodies, thyroglobulin, thyroid ultrasound TI-RADS classification, apolipoprotein A-I, The 9 indexes of albumin are used as independent variables of the model, and the classification indexes among the 9 indexes are assigned, and the classification indexes include gender, thyroid autoantibodies and thyroid ultrasound TI-RADS classification; S2, construct a BP neural network model, the BP neural network model includes a three-layer feedforward neural network structure, which are an input layer, a hidden layer and an output layer, and the input indicators of the input layer are the 9 indicators selected in step S1 , the output index of the output layer is the prediction result of papillary carcinoma in patients with thyroid nodules; S3. Collect the clinical data of the patient, create a training set and a prediction set, and input the data of the training set into the BP neural network model constructed in step S2 for training; after the training is completed, input the data of the prediction set into the model, and output the prediction of the model The value is compared with the real value, and the ROC curve is drawn to obtain the area under the curve AUC. The AUC value is used to evaluate the diagnostic performance of the model. The model whose diagnostic performance meets the standard is the target risk assessment model. 2. a kind of risk assessment model for thyroid nodule patient papillary carcinoma according to claim 1, is characterized in that: In step S1, the assignment of the classification index is: Gender: 1 for males, 0 for females; Thyroid autoantibodies: if any one of TgAb and TPOAb is positive, it is positive (+) and assigned as 1; if both are negative, it is negative (-) and assigned as 2; Thyroid ultrasound TI-RADS classification: 0 for 2; 1 for 3; 2 for 4a; 3 for 4b; 4 for 4c; 5 for 5; In step S2, the output index of the output layer is assigned as follows: benign nodule is 1, and papillary thyroid carcinoma is 2. 3. a kind of risk assessment model for thyroid nodule patient papillary carcinoma according to claim 1, is characterized in that: In step S2, the number of hidden layers of the constructed BP neural network model is 1, the number of neuron nodes is 15, the training function is the Levenberg-Marquardt function, the weight update method of error back propagation is the gradient descent method, and the number of model iterations is is 1000, and the learning rate is 0.01.

Images