Method for predicting characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network

Abstract

The present invention proposes a method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network. The feedforward neural network technology is introduced, combined with a bioheat transfer mathematical model and the establishment of a model based on the different heat generations of normal tissues and abnormal tissues. The data generated by modeling is used to train the feedforward neural network to reach the best fitting state, so as to improve the prediction accuracy of new data. On this basis, the data is flattened, so that a matrix composed of multiple data generated by one model becomes a vector. Multiple models are established to obtain a data set composed of data of multiple models. At the same time, the angle parameter of the abnormal tissue in the normal tissue is increased to improve the prediction of the state of the abnormal tissue. The present invention can make the prediction accuracy of the size and position of the abnormal tissue of a given proportion ellipsoid greater than 98%, and it is found that the prediction accuracy of the posture angle of the abnormal tissue is significantly correlated with the irregularity of the abnormal tissue, playing a role in assisting scientific research.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning and the modeling of the surface temperature distribution of biological tissues, and particularly relates to a method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network. Background Art

[0002] During the process of cell division and differentiation of biological tissues, affected by their own gene characteristics and environmental factors, cells may undergo lesions, forming abnormal tissues in a certain area of biological tissues, which affects the normal physiological activities of organisms. Therefore, timely detection of abnormal tissues and prediction of their characteristics are helpful for biological research. The prediction results can be further processed to obtain effective abnormal tissue images in the organism, and the artificial neural network is a method for processing the prediction results. Summary of the Invention

[0003] The theoretical basis of the present invention lies in that there are certain differences in the temperature of the abnormal tissue parts of organisms compared with healthy tissues. Therefore, the temperature abnormality in the abnormal tissue area can be used as an effective index for diagnosing the characteristics of abnormal tissues contained in biological tissues, and the finite element method is used to solve the one-dimensional unsteady bioheat equation. The thermal imaging map caused by the infrared radiation on the surface of biological tissues is obtained through infrared thermal imaging technology, so as to obtain the final temperature distribution, and further information related to the size and position of abnormal tissues can be found in the surface temperature distribution of biological tissues.

[0004] Therefore, the purpose of the present invention is to provide a method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network. The temperature distribution of the biological tissue model is obtained by solving the partial differential Pennes bioheat transfer equation through the finite element method, and at the same time, it serves as the data basis for training the neural network model. Compared with the ideal spherical shape, the ellipsoidal abnormal tissue contains attitude parameters in addition to position and size. Therefore, multiple training cases can be simulated by changing the size and position of the abnormal tissue to generate sufficient data for training the created neural network. In addition, the present invention also determines the optimal number of neurons in the hidden layer of the neural network model by observing the change of the root mean square error of the generated results, determines the training accuracy of the model using simulation data, and verifies the performance of the model using test data. Therefore, it is of great significance to explore the method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network.

[0005] It introduces the feedforward neural network technology, combines with the bioheat transfer mathematical model and establishes models based on the different heat generations of normal tissues and abnormal tissues. By using the data generated from modeling to train the feedforward neural network to reach the best fitting state, the prediction accuracy of new data is improved. On this basis, the data is flattened so that a matrix composed of multiple data generated by one model becomes a vector. Multiple models are established to obtain a dataset composed of multiple model data. At the same time, the angular parameter of abnormal tissues within normal tissues is increased to improve the prediction of the state of abnormal tissues. The present invention can make the prediction accuracy of the size and position of abnormal tissues in a given proportion of ellipsoidal shapes greater than 98%. It is also found that the prediction accuracy of the attitude angle of abnormal tissues is significantly correlated with the irregularity of abnormal tissues, playing a role in assisting scientific research.

[0006] The technical solution adopted by the present invention to solve its technical problems is as follows:

[0007] A method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network, characterized by comprising the following steps:

[0008] Step S1: Construct a geometric model of biological tissues;

[0009] Step S2: Solve the partial differential Pennes bioheat transfer equation by the finite element method in combination with boundary conditions and initial conditions to obtain the temperature distribution of biological tissues;

[0010] Step S3: Design multiple different parameters to construct multiple different models;

[0011] Step S4: Simulate the multiple models constructed in Step S3, export and organize the simulation data results, and use them as the training set, validation set, and test set respectively according to a certain proportion;

[0012] Step S5: Construct a feedforward neural network, and input the training set and validation set data into the neural network for training and validation;

[0013] Step S6: Use the RMSE error to obtain the optimal number of neurons;

[0014] Step S7: Set the neural network parameters and perform training and adjustment to achieve the best fit;

[0015] Step S8: Use the trained neural network to predict the characteristics of abnormal tissues contained in biological tissues.

[0016] Non-destructive detection of abnormal tissues can be carried out by obtaining the thermal imaging map caused by the infrared radiation on the surface of biological tissues through infrared thermal imaging technology, so as to predict the size and position information of abnormal tissues contained in biological tissues.

[0017] Further, step S1 specifically includes the following steps:

[0018] Step S11: Construct two different biological tissues. The first biological tissue is semi-ellipsoidal, and the second biological tissue is ellipsoidal and embedded inside the first biological tissue.

[0019] Step S12: Set the property parameters of the two biological tissue materials respectively.

[0020] Further, the property parameters of the biological tissue materials in step S12 include the density of the tissue, specific heat capacity, tissue temperature, thermal conductivity, blood perfusion term, metabolic heat production rate, and distributed volume heat source. The property parameters of the two biological tissues are different.

[0021] Further, step S2 specifically includes the following steps:

[0022] Step S21: Divide the geometric model of the entire biological tissue into several finite intervals, combine the boundary conditions of absolute pressure and external temperature, and combine the initial conditions of biological tissue temperature and abnormal tissue temperature.

[0023] Step S22: Solve the partial differential Pennes bioheat transfer equation for several finite intervals to obtain the temperature distribution data on the model surface. The Pennes bioheat transfer equation is expressed as:

[0024]

[0025] where the symbol is the Hamiltonian operator. The subscript i is used to represent the properties of different biological tissues, where "0" represents normal biological tissue and "1" represents abnormal biological tissue; the subscript b is used to describe the thermal properties of blood; the parameters ρ i , c i , T i , k i respectively represent the density of the tissue, specific heat capacity, tissue temperature, thermal conductivity, and arterial temperature; Q b , Q m and Q s respectively represent the blood perfusion term, metabolic heat production rate, and distributed volume heat source.

[0026] Further, in step S3, the multiple different parameters designed include: biological tissue size, abnormal tissue size, abnormal tissue location, and abnormal tissue rotation angle.

[0027] Among them, four scales of biological tissues are selected, and the range of the radius of abnormal tissues is 5 mm - 30 mm; the shape of abnormal tissues is set as an ellipsoid, and the ratio of the abc semi-axes is assumed to be 9:8:6. At the same time, 6 abnormal tissue coordinates are randomly generated within the given model range; then, these coordinates are assigned to models with different abnormal tissue radii; and a pair of angular parameters θ and ψ are added to the ellipsoid type, and the two sets of angular parameters are combined into 4 postures; θ and ψ define the rotation angle of the abnormal tissue through a three-dimensional spherical surface. Among them, θ is the rotation of the ellipsoid with the y-axis as the rotation axis from the positive half-axis of the z-axis to the positive half-axis of the x-axis, and ψ is the rotation of the ellipsoid with the z-axis as the rotation axis from the positive half-axis of the x-axis to the positive half-axis of the y-axis.

[0028] Furthermore, in step S4, the data composition of the simulation result output includes the X, Y, and Z coordinates of each point, as well as the temperature value corresponding to that point.

[0029] Furthermore, the feedforward neural network constructed in step S5 refers to selecting the Sigmoid function as the activation function, selecting the Levenberg-MarQuardt algorithm as the training function, and adjusting the weights and biases between neurons through backpropagation; at the same time, the data simulated by the model is flattened and trained in the input neural network.

[0030] Furthermore, the specific expression of RMSE in step S6 is:

[0031]

[0032] Among them, is the predicted value of the neural network, and y k is the actual value.

[0033] Furthermore, the neural network parameters in step S7 are determined by comparing the values of RMSE in step S6 to determine the number of neurons in the hidden layer, so as to obtain the most accurate output result.

[0034] Compared with the prior art, the present invention and its preferred solutions solve the Pennes bioheat transfer equation with specified initial conditions and boundary conditions by the finite element method to obtain a data set of the surface temperature of biological tissues. After flattening the data set, it is input into a feedforward neural network for training, and the root mean square error is used as an evaluation criterion to optimize the number of neurons, so that the fitting effect reaches the best state. The results show that the neural network algorithm has a high prediction accuracy for the volume and coordinates of abnormal tissues. At the same time, the difference in the shape of abnormal tissues has an impact on the prediction accuracy of the attitude angle. The more slender the shape of abnormal tissues, the higher the prediction accuracy of the attitude angle. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments:

[0036] Figure 1 This is the flowchart of the method according to the embodiments of the present invention.

[0037] Figure 2 This is the geometric model diagram of biological tissue in the embodiments of the present invention.

[0038] Figure 3 This is the relationship diagram between the abnormal tissue size and the predicted and actual values of its X, Y, and Z coordinates in the embodiments of the present invention.

[0039] Figure 4 This is the curve diagram of the predicted and actual values of the angles θ and ψ of the abnormal tissue model ratios 9:8:6 and 9:8:4 in the embodiments of the present invention. Detailed implementation manners

[0040] To make the features and advantages of this patent more obvious and understandable, specific embodiments are given below and described in detail as follows:

[0041] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further descriptions of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those of ordinary skill in the technical field to which the present application belongs.

[0042] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary implementation manners according to the present application. As used herein, unless otherwise clearly specified in the context, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "include" and / or "comprise" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0043] As Figure 1 shown, the present invention provides a design process for a method for predicting abnormal tissue characteristics in biological tissue based on a feedforward neural network, including the following steps:

[0044] Step S1: Construct a geometric model of biological tissue, which specifically includes the following steps:

[0045] Step S11: Construct two different biological tissues. The first biological tissue is pear-shaped (i.e., semi-ellipsoidal), and the second biological tissue is ellipsoidal and embedded inside the first biological tissue;

[0046] As a preferred solution of this embodiment, as Figure 2As shown, the biological model is a biological tissue model, which is divided into abnormal tissue and healthy tissue. The healthy tissue is the first biological tissue, with a width, depth, and height of 8 cm, 6.2 cm, and 6.2 cm respectively; the abnormal tissue is the second biological tissue, with a width, depth, and height of 1.17 cm, 1.04 cm, and 0.78 cm respectively. Step S12: Set the material property parameters of the two biological tissues respectively, including the density, specific heat capacity, tissue temperature, thermal conductivity, blood perfusion term, metabolic heat production rate, and distributed volume heat source of the tissue. The tissue property parameters of the two are different.

[0047] Step S2: Solve the partial differential Pennes bioheat transfer equation by the finite element method combined with boundary conditions and initial conditions to obtain the temperature distribution of biological tissues, which specifically includes the following steps:

[0048] Step S21: Divide the entire geometric model into several finite intervals, combine the boundary condition that the absolute pressure is 1 standard atmospheric pressure, the external temperature is 298.15 K, and combine the initial conditions that the biological tissue temperature is 310.38 K and the abnormal tissue temperature is 313.16 K;

[0049] Step S22: Solve the partial differential Pennes bioheat transfer equation for several finite intervals to obtain the temperature distribution data on the surface of the model. The Pennes bioheat transfer equation is expressed as:

[0050]

[0051] Among them, the symbol is the Hamiltonian operator. The subscript i is used to represent the properties of different biological tissues, where "0" represents biological tissue and "1" represents abnormal tissue. The subscript b is used to describe the thermal properties of blood. The parameters ρ i , c i , T i , k i represent the density, specific heat capacity, tissue temperature, and thermal conductivity of the tissue respectively. In addition, Q n , Q m and Q s represent the blood perfusion term, metabolic heat production rate, and distributed volume heat source respectively.

[0052] Step S3: Design multiple different parameters, such as the size of the biological tissue, the size of the abnormal tissue, the location of the abnormal tissue, and the rotation angle of the abnormal tissue, to construct multiple different models. In this embodiment, 4 scales of the biological tissue are selected, which are (8, 6.2, 6.2), (8, 6.8, 6.8), (7.2, 6.2, 6.2), and (7.2, 6.8, 6.8) respectively. The range of the radius of the abnormal tissue is 5mm - 30mm. The shape of the abnormal tissue is set as an ellipsoid, and the ratio of its abc semi-axes is assumed to be 9:8:6. At the same time, 6 coordinates of the abnormal tissue are randomly generated within the given model range, which are (25, 24, 45), (40, 102, 90), (38, 112, 62), (31, 43, 85), (46, 61, 58), and (29, 67, 36). Then, these coordinates are assigned to the models with different abnormal tissue radii. The abnormal tissue model of the present invention considers the ellipsoidal shape, so a pair of angle parameters θ and ψ are added. Among them, θ takes values of 25, 56, 74, and 120, and ψ takes values of 34, 80, 22, and 105. The two sets of angle parameters are combined into 4 postures. θ and ψ define the rotation angle of the abnormal tissue through a three-dimensional spherical surface. Among them, θ is the rotation of the ellipsoid around the y-axis from the positive half-axis of the z-axis to the positive half-axis of the x-axis, and ψ is the rotation of the ellipsoid around the z-axis from the positive half-axis of the x-axis to the positive half-axis of the y-axis.

[0053] Step S4: Simulate multiple models, export and organize the simulation data results, and use them as the training set, validation set, and test set according to the ratio of 70%:15%:15% respectively.

[0054] Step S5: Construct a feedforward neural network, input the training set and validation set data into the neural network for training and validation. Select the Sigmoid function as the activation function, select the Levenberg-MarQuardt algorithm as the training function, and adjust the weights and biases between neurons through backpropagation. At the same time, flatten the data simulated by the model and input it into the neural network at the input end for training.

[0055] Step S6: Use the RMSE error to statistically obtain the optimal number of neurons;

[0056] The specific expression of RMSE is:

[0057]

[0058] Among them, is the predicted value of the neural network, and y k is the actual value.

[0059] Step S7: Set appropriate neural network parameters to obtain good training results. In step S6, compare the magnitudes of the RMSE values to determine the number of neurons in the hidden layer, so as to obtain the most accurate output result.

[0060] Step S8: Input the test set data into the neural network, analyze and summarize the prediction results of the neural network, and apply them to the prediction of abnormal tissue characteristics in biological tissues.

[0061] As Figure 3 shown, obtain the relationship between the predicted values and the actual values of the size and position X, Y, Z coordinates of the abnormal tissue, judge the accuracy of the neural network prediction, and the prediction results reach a very high accuracy rate, and the regression slopes are all very close to the ideal value of "1";

[0062] As Figure 4 shown, obtain the relationship between the predicted values and the actual values of the abnormal tissue angles θ, ψ and the abnormal tissue angles θ, ψ after changing the abc ratio (9:8:4) of the abnormal tissue model, and judge the accuracy of the neural network prediction. The regression slope error of the original abc ratio prediction result is relatively large, but it still has certain reference value; after changing the abc ratio to (9:8:4), the prediction results show that although the slope of the regression line of the angle ψ decreases, the slope of the regression line of the angle θ increases by about 10%. Therefore, it can be concluded that the difference in the shape of the abnormal tissue has an impact on the prediction accuracy of the pose angle, and the more slender the shape of the abnormal tissue, the higher the prediction accuracy of the pose angle.

[0063] The above is only a preferred embodiment of the present invention, and it is not a limitation of the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still belong to the protection scope of the technical solution of the present invention.

[0064] This patent is not limited to the above best implementation manner. Anyone inspired by this patent can obtain various other forms of methods for predicting abnormal tissue characteristics in biological tissues based on feedforward neural networks. All equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope covered by this patent.

Claims

1. A method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network, characterized in that, it includes the following steps: Step S1: Construct a geometric model of biological tissue; Step S2: Solve the partial differential Pennes bioheat transfer equation by the finite element method combined with boundary conditions and initial conditions to obtain the temperature distribution of biological tissue; Step S3: Design multiple different parameters to construct multiple different models; Step S4: Simulate the multiple models constructed in Step S3, export and organize the simulation data results, and use them as the training set, validation set, and test set respectively according to a certain ratio; Step S5: Construct a feedforward neural network, and input the training set and validation set data into the neural network for training and validation; Step S6: Use the RMSE error to obtain the optimal number of neurons; Step S7: Set the neural network parameters and perform training and adjustment to achieve the best fit; Step S8: Use the trained neural network to predict the characteristics of abnormal tissues contained in biological tissues; Step S2 specifically includes the following steps: Step S21: Divide the geometric model of the entire biological tissue into several finite intervals, combine the boundary condition of absolute pressure and external temperature, and combine the initial conditions of biological tissue temperature and abnormal tissue temperature; Step S22: Solve the partial differential Pennes bioheat transfer equation for several finite intervals to obtain the temperature distribution data on the model surface. The Pennes bioheat transfer equation is expressed as: Among them, the symbol is the Hamiltonian operator. The subscript i is used to represent the properties of different biological tissues, and its values include 0 and 1. "0" represents normal biological tissues, and "1" represents abnormal biological tissues. The subscript b is used to describe the thermal properties of blood. The parameter ρ i , c i , T i , k i represent the density, specific heat capacity, tissue temperature, and thermal conductivity of the tissue, respectively. Q b , Q m and Q s represent the blood perfusion term, metabolic heat production rate, and distributed volume heat source, respectively. In Step S3, the multiple different parameters designed include: the size of biological tissue, the size of abnormal tissue, the position of abnormal tissue, and the rotation angle of abnormal tissue; Among them, 4 scales of sizes are selected for biological tissue, and the range of the radius of abnormal tissue is 5 mm - 30 mm; the shape of abnormal tissue is set as an ellipsoid, and the ratio of its abc semi-axes is assumed to be 9:8:

6. At the same time, 6 abnormal tissue coordinates are randomly generated within the given model range; then, these coordinates are assigned to models with different abnormal tissue radii; and for the ellipsoidal shape, a pair of angle parameters θ and ψ are added, and the two sets of angle parameters are combined into 4 postures; θ and ψ define the rotation angle of abnormal tissue through a three-dimensional spherical surface. Among them, θ is the rotation of the ellipsoid around the y-axis from the positive half-axis of the z-axis to the positive half-axis of the x-axis, and ψ is the rotation of the ellipsoid around the z-axis from the positive half-axis of the x-axis to the positive half-axis of the y-axis.

2. The method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network according to claim 1, characterized in that: Step S1 specifically includes the following steps: Step S11: Construct two different biological tissues, the first biological tissue is semi-ellipsoidal, and the second biological tissue is ellipsoidal and embedded inside the first biological tissue; Step S12: Set the material property parameters of the two biological tissues respectively.

3. The method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network according to claim 2, characterized in that, the biological tissue material property parameters in Step S12 include the density of the tissue, specific heat capacity, tissue temperature, thermal conductivity, blood perfusion term, metabolic heat generation rate, and distributed volume heat source, and the two biological tissue property parameters are different.

4. The method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network according to claim 1, characterized in that: In step S4, the data composition of the simulation result output includes the X, Y, and Z coordinates of each point, and the temperature value corresponding to that point.

5. The method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network according to claim 1, characterized in that, The feedforward neural network constructed in step S5 refers to selecting the Sigmoid function as the activation function, selecting the Levenberg-MarQuardt algorithm as the training function, and adjusting the weights and biases between neurons through backpropagation; at the same time, the data simulated by the model is flattened and trained in the input end neural network.

6. The method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network according to claim 1, characterized in that, The specific expression of RMSE in step S6 is: Among them, is the predicted value of the neural network, and y k is the actual value.

7. The method for predicting the characteristics of abnormal tissues contained in biological tissues based on a feedforward neural network according to claim 1, characterized in that, The neural network parameters in step S7 are obtained by comparing the values of RMSE in step S6 to determine the number of hidden layer neurons, so as to obtain the most accurate output result.

Images