A human tissue identification and classification method, system, computer equipment and medium

Abstract

The present invention provides a method, system, computer equipment and medium for identifying and classifying human tissues. The method comprises obtaining a dielectric property data set of human tissues, and dividing the dielectric property data set into a training set and a validation set according to a preset ratio; the dielectric property data set comprises dielectric property data and corresponding classification labels; a deep learning neural network is constructed according to the dispersion frequency characteristics of the training set; the deep learning neural network is trained according to the training set to obtain a human tissue identification and classification model; the dielectric property data to be identified in the validation set is input into the human tissue identification and classification model for identification and classification to obtain corresponding classification results. In actual human tissue identification and classification applications, the present invention effectively solves the technical defects of existing human tissue identification and classification, and not only has strong practicality, objectivity and repeatability, but also has high learning efficiency and classification accuracy.

Description

Technical Field

[0001] The present invention relates to the field of dielectric property measurement of biological tissues, and in particular to a method, system, computer equipment and storage medium for identifying and classifying human tissues based on dielectric properties. Background Art

[0002] The dielectric properties of human tissue, including conductivity and relative dielectric constant, are the basic physical properties of human tissue. A large number of studies have shown that when a tissue is diseased, the value of the dielectric properties will also change accordingly, so the dielectric property values ​​reflect the characteristic markers of the tissue to a certain extent. Obtaining the dielectric properties of human tissue can be used to identify and classify tissues, and provide information on whether the tissue is normal tissue, diseased tissue, and what type of diseased tissue it is, which can be applied to the diagnosis of clinical diseases.

[0003] In the process of extracting clinical diagnostic information using dielectric properties, the identification and classification of human tissues based on dielectric properties is a key link. At present, the existing method is to measure the dielectric properties of the human body, and then compare the measured values ​​of the dielectric properties with the literature values, and judge whether the measured tissue has a lesion and what kind of lesion it has based on the comparison results. However, this method has the following shortcomings:

[0004] First, the literature values ​​used for comparison are relatively limited, which reduces the practicality of existing methods. Existing methods need to use literature values ​​for comparison. Although the literature values ​​of the dielectric properties of healthy human tissues are relatively complete, and the dielectric properties of most healthy tissues have been reported in the literature, there are fewer literature reports on the dielectric properties of diseased tissues, and the types of lesions are limited. For example, the dielectric properties of liver tumor tissues are only reported in the literature with three simple classifications of malignant, benign, and normal, while there are no literature values ​​available for the dielectric properties corresponding to tumor grading, tumor type, etc. Therefore, existing methods can only be used for a few diseases to achieve relatively simple identification and classification.

[0005] Second, the individual differences in the dielectric properties of human tissues lead to low accuracy of existing methods. In actual clinical practice, even if the tissues are the same, the dielectric properties of different patients are not exactly the same. The dielectric property values ​​show a certain fluctuation range in different individuals and have individual differences. Existing methods directly use literature values ​​for comparison, which cannot take into account the impact of individual differences, resulting in low accuracy of identification and classification results.

[0006] Third, the existing methods are highly subjective and have low repeatability. The existing methods require the implementer to compare the measured values ​​of the dielectric properties with the values ​​in the literature to determine what kind of pathological tissue the tissue to be identified and classified should be classified as. The implementation of the method is highly subjective. In addition, since the identification and classification of tissues usually rely on the implementer's experience, different implementers often get different results, so the existing methods have low repeatability.

[0007] Therefore, there is an urgent need to provide a method for identifying and classifying human tissues to overcome the shortcomings of existing methods. Summary of the invention

[0008] The purpose of the present invention is to provide a method for identifying and classifying human tissues. A dielectric property data set is established through a novel data collection method, and is used to train a deep learning convolutional neural network constructed based on the characteristics of the dielectric property data, so as to obtain a stable human tissue identification and classification model for human tissue identification, so as to ensure strong practicality, objectivity, repeatability, as well as high accuracy and learning efficiency.

[0009] In order to achieve the above objectives, it is necessary to provide a human tissue identification and classification method, system, computer device and storage medium in response to the above technical problems.

[0010] In a first aspect, an embodiment of the present invention provides a method for identifying and classifying human tissues, the method comprising the following steps:

[0011] Acquire a dielectric property data set of human tissue, and divide the dielectric property data set into a training set and a validation set according to a preset ratio; the dielectric property data set includes dielectric property data and corresponding classification labels;

[0012] Constructing a deep learning neural network according to the dispersion frequency characteristics of the training set;

[0013] Training the deep learning neural network according to the training set to obtain a human tissue identification and classification model;

[0014] The dielectric property data to be identified in the verification set is input into the human tissue identification and classification model for identification and classification to obtain corresponding classification results.

[0015] Furthermore, the step of obtaining a dielectric property data set of human tissue includes:

[0016] Obtaining a human tissue sample, and determining a part to be measured based on the human tissue sample; the human tissue sample includes a resected tumor and surrounding tissue of the tumor;

[0017] Using an open-end coaxial method, repeatedly measuring the dielectric properties of each to-be-measured part of the human tissue sample, obtaining reflection coefficients corresponding to a preset number of groups, and calculating the corresponding dielectric property data based on the reflection coefficients;

[0018] Each part to be measured of the human tissue sample is made into a corresponding pathological section to be measured, and each pathological section to be measured is identified and classified according to histopathology to obtain the classification label of each part to be measured.

[0019] Furthermore, the step of calculating the corresponding dielectric characteristic data according to the reflection coefficient includes:

[0020] Calibration determines the characteristic parameters of the open-ended coaxial probe;

[0021] According to the characteristic parameters of the open-ended coaxial probe and the reflection coefficients of the various parts to be measured of the human tissue sample, the initial dielectric characteristic data corresponding to the preset number of groups are calculated; the initial dielectric characteristic data is expressed as:

[0022]

[0023]

[0024] Where f and ρ represent the measurement frequency and the measured reflection coefficient, respectively; λ1, λ2 and λ3 represent the characteristic parameters of the open-ended coaxial probe; Re(·) and Im(·) represent the real part operation and the imaginary part operation, respectively; ε r and σ represent the relative permittivity and conductivity of the part to be measured, respectively;

[0025] The initial dielectric property data of each to-be-measured part of the human tissue sample are preprocessed by using mean filtering and low-pass filtering in sequence to obtain the corresponding dielectric property data.

[0026] Furthermore, the deep learning neural network includes an input layer, a dielectric feature extraction layer, a dielectric feature fusion layer and an output layer in sequence; the dielectric feature fusion layer includes a plurality of fully connected layers;

[0027] The step of constructing a deep learning neural network according to the dispersion frequency characteristics of the training set includes:

[0028] According to the relative dielectric constant and conductivity of the dielectric characteristic data in the training set, the dielectric feature extraction layer is set; the dielectric feature extraction layer includes a relative dielectric constant feature extraction layer and a conductivity feature extraction layer;

[0029] According to the dispersion frequency band of the dispersion frequency characteristic, the relative dielectric constant feature extraction layer and the conductivity feature extraction layer are respectively set; the relative dielectric constant feature extraction layer and the conductivity feature extraction layer both include a frequency band feature extraction layer and a frequency band feature fusion layer; the frequency band feature extraction layer includes a plurality of dispersion frequency band feature extraction layers of different frequency bands; the dispersion frequency band feature extraction layer includes a first convolution layer and a first pooling layer; the frequency band feature fusion layer includes a second convolution layer and a second pooling layer.

[0030] Furthermore, the step of inputting the dielectric property data to be identified in the verification set into the human tissue identification and classification model for identification and classification to obtain the corresponding classification result includes:

[0031] Determining whether the measurement frequency of the dielectric property data to be identified is consistent with the measurement frequency of the dielectric property data in the training set;

[0032] If the measurement frequency of the dielectric property data to be identified is consistent with the measurement frequency of the dielectric property data in the training set, the human tissue identification and classification model is used to identify and classify the dielectric property data to be identified to obtain the classification result. Otherwise, the dielectric property data to be identified is interpolated and fitted by the Cole-Cole model to obtain normalized dielectric property data to be identified, and the human tissue identification and classification model is used to identify and classify the normalized dielectric property data to be identified to obtain the classification result.

[0033] Furthermore, the step of performing interpolation fitting on the dielectric property data to be identified by using a Cole-Cole model to obtain normalized dielectric property data to be identified includes:

[0034] The Cole-Cole model is fitted using a nonlinear optimization algorithm to obtain a Cole-Cole interpolation model; the Cole-Cole model is expressed as:

[0035]

[0036] In the formula, ε * represents the complex dielectric constant; σ and ε r Respectively represent the measured conductivity and relative dielectric constant; f represents the measurement frequency; ε0 represents the vacuum dielectric constant; j represents the sign of the complex imaginary part; σ s Indicates DC conductivity; ε ∞ Indicates the relative dielectric constant when the frequency is infinite; Δε n , τ n and α n They represent the relative permittivity increment, average relaxation time and relaxation time distribution coefficient corresponding to the nth dispersion frequency band, and n takes the values ​​of 1, 2, 3 and 4;

[0037] The normalized dielectric property data to be identified are obtained according to the Cole-Cole interpolation model and the measurement frequency of the dielectric property data to be identified.

[0038] Furthermore, the step of fitting the Cole-Cole model using a nonlinear optimization algorithm to determine the undetermined coefficients of the Cole-Cole model includes:

[0039] According to the dispersion frequency characteristics of the dielectric properties of human tissue, the Cole-Cole model is expanded according to the prior information to obtain the Cole-Cole frequency band model corresponding to each dispersion frequency characteristic;

[0040] Each Cole-Cole frequency band model is fitted separately to obtain the corresponding Cole-Cole frequency band interpolation model;

[0041] The Cole-Cole interpolation model is obtained according to each Cole-Cole frequency band interpolation model.

[0042] In a second aspect, an embodiment of the present invention provides a human tissue identification and classification system, the system comprising:

[0043] An acquisition module, used to acquire a dielectric property data set of human tissue, and divide the dielectric property data set into a training set and a validation set according to a preset ratio; the dielectric property data set includes dielectric property data and corresponding classification labels;

[0044] A modeling module, used for constructing a deep learning neural network according to the dispersion frequency characteristics of the training set;

[0045] A training module, used to train the deep learning neural network according to the training set to obtain a human tissue identification and classification model;

[0046] The classification module is used to input the dielectric property data to be identified in the verification set into the human tissue identification and classification model for identification and classification to obtain corresponding classification results.

[0047] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0048] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0049] The present application provides a method, system, computer device and storage medium for human tissue identification and classification based on dielectric properties. Through the method, the dielectric property data of different parts of human tissue samples are measured by the open-end coaxial method, and the dielectric property data of each measured part is determined according to the histopathology. The corresponding classification label of the pathological slice is constructed to construct a dielectric property data set of human tissue, and after the training set and the validation set are divided according to the preset ratio, a deep learning neural network including an input layer, a dielectric feature extraction layer, a dielectric feature fusion layer and an output layer is constructed according to the dispersion frequency characteristics of the dielectric property data in the training set, and the deep learning neural network is trained according to the training set to obtain a human tissue identification and classification model, and the dielectric property data to be identified in the validation set is input into the human tissue identification and classification model for identification and classification to obtain the corresponding classification result. Compared with the prior art, the human tissue identification and classification method effectively solves the technical defects of the existing human tissue identification and classification in the actual human tissue identification and classification application, and not only has strong practicality, objectivity and repeatability, but also has high accuracy and learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a schematic diagram of an application scenario of a human tissue identification and classification method based on dielectric properties in an embodiment of the present invention;

[0051] Figure 2 is a schematic flow chart of a method for identifying and classifying human tissue based on dielectric properties in an embodiment of the present invention;

[0052] Figure 3 yes Figure 2 A schematic diagram of a flow chart of obtaining a dielectric property data set of human tissue in step S11;

[0053] Figure 4 is a schematic diagram of the structure of a deep learning neural network in an embodiment of the present invention;

[0054] Figure 5 yes Figure 2 Step S12 is a flow chart of constructing a deep learning neural network according to the dispersion frequency characteristics of the training set;

[0055] Figure 6 yes Figure 2 Step S14 is a flow chart of performing identification and classification according to the human tissue identification and classification model to obtain the classification result;

[0056] Figure 7 is a schematic structural diagram of a human tissue identification and classification system based on dielectric properties in an embodiment of the present invention;

[0057] Figure 8 It is a diagram of the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical scheme and beneficial effects of the present application clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the embodiments described below are part of the embodiments of the present invention and are only used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] The human tissue identification and classification method based on dielectric properties provided by the present invention can be applied to Figure 1 The terminal or server shown. The terminal may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, and portable wearable devices, and the server may be implemented as an independent server or a server cluster consisting of multiple servers. The server may adopt the human tissue identification and classification method based on dielectric properties of the present invention to directly obtain the corresponding classification results according to the acquired human tissue samples, and use them for subsequent research and analysis by the server, or send them to the terminal for research and use by the terminal user.

[0060] In one embodiment, Figure 2 As shown, a method for identifying and classifying human tissues is provided, comprising the following steps:

[0061] S11, obtaining a dielectric property data set of human tissue, and dividing the dielectric property data set into a training set and a validation set according to a preset ratio; the dielectric property data set includes dielectric property data and corresponding classification labels;

[0062] Among them, human tissues can be various organ tissues removed during solid tumor surgery, such as liver tumors, breasts, glands and peripheral lymph nodes, etc. This embodiment does not impose any restrictions on the specific source of human sample tissues, that is, the tissue identification method can be applied to the identification and classification of different tissues without affecting the corresponding classification effect. Since deep neural network learning is to be based on the dielectric property data set of human tissue, the dielectric property data set to be constructed must include the dielectric property data of each tissue and the corresponding classification label. Therefore, this embodiment uses the open-end coaxial method and the principle of tissue pathology to obtain the dielectric property data and classification label corresponding to each sample tissue, and determines the ratio of the training set and the verification set for deep neural network learning according to actual application needs, and divides the obtained dielectric property data set into a training set and a verification set according to the preset ratio for subsequent training and verification of the deep neural network. Specifically, Figure 3 As shown, the step S11 of acquiring a dielectric property data set of human tissue includes:

[0063] S111, obtaining a human tissue sample, and determining a part to be measured according to the human tissue sample; the human tissue sample includes a resected tumor and surrounding tissue of the tumor;

[0064] Among them, as mentioned above, human tissue samples can be determined according to actual application needs. After obtaining the tumor and surrounding tissue removed by the surgeon in tumor surgery, different tissue parts are selected for measurement according to the type of tissue removed. For example, if the liver tumor resection tissue is obtained, the parts to be measured can be the center of the tumor, the edge of the tumor, the tissue edema, the normal tissue at the distal end of the tumor, etc. It should be noted that the parts to be measured given here are only exemplary descriptions and are not intended to be a limitation of the specific scope. The parts to be measured of other tissue types can be determined according to actual needs.

[0065] S112, using an open-end coaxial method to repeatedly measure the dielectric properties of each to-be-measured part of the human tissue sample, to obtain reflection coefficients corresponding to a preset number of groups, and to calculate the corresponding dielectric property data based on the reflection coefficients;

[0066] Among them, after each part to be measured of the human sample tissue is determined according to the above method, a network analyzer (wideband network analyzer) or other related dielectric property measurement equipment can be used to measure each part to be measured using the open-end coaxial method to obtain the corresponding reflection coefficient. Each part to be measured is measured no less than 3 times, and the reflection coefficients of all parts to be measured are no less than 100 groups. It should be noted that each group measures the reflection coefficients of the above-mentioned different parts to be measured, and each group of measurement data should have the same measurement frequency. For example, in the process of multiple surgical measurements, a total of 200 groups of data are obtained, and the frequency of each group of measurement data is between 1kHz and 5GHz. After repeatedly measuring multiple reflection coefficients of each part to be measured according to the above method, the corresponding relative dielectric constant and conductivity can be obtained using the calculation formula of dielectric properties. Specifically, the step of calculating the corresponding dielectric property data based on the reflection coefficient includes:

[0067] The characteristic parameters of the open-ended coaxial probe are determined by calibration; wherein the specific calibration method for determining the characteristic parameters is: firstly, the open-ended coaxial probe is connected to an open circuit, a short circuit and a known load (or a standard liquid, such as deionized water), respectively, and three corresponding groups of reflection coefficients are measured respectively, and then the characteristic parameters of the open-ended coaxial probe are calculated based on the three groups of reflection coefficients;

[0068] According to the characteristic parameters of the open-ended coaxial probe and the reflection coefficients of the various parts to be measured of the human tissue sample, the initial dielectric characteristic data corresponding to the preset number of groups are calculated; the initial dielectric characteristic data is expressed as:

[0069]

[0070]

[0071] Where f and ρ represent the measurement frequency and the measured reflection coefficient, respectively; λ1, λ2 and λ3 represent the characteristic parameters of the open-ended coaxial probe; Re(·) and Im(·) represent the real part operation and the imaginary part operation, respectively; ε r and σ represent the relative permittivity and conductivity of the part to be measured, respectively;

[0072] The initial dielectric property data of each to-be-measured part of the human tissue sample are preprocessed by using mean filtering and low-pass filtering in sequence to obtain the corresponding dielectric property data.

[0073] Among them, mean filtering and low-pass filtering are used to smooth and denoise the dielectric characteristic data. Specifically, if the dielectric characteristic data obtained by the above measurement are repeated to obtain 5 sets of data at each part to be measured, the 5 sets of repeated measurement data are first mean filtered, and then the low-pass filter is used to smooth the mean filtered data. The expression of the corresponding filter is:

[0074]

[0075] Where, the standard deviation d is 1, the filter kernel size is 5, and the filter matrix is: [0.0545, 0.2442, 0.4026, 0.2442, 0.0545]. The filter matrix is ​​used to calculate the conductivity σ and the relative dielectric constant ε. r Perform convolution to achieve smooth filtering and eliminate the interference of random noise in measurement.

[0076] It should be noted that the filter matrix given above is only an exemplary description. In practical applications, the standard deviation of the filter expression and the setting of the filter kernel size directly determine the degree of smoothing, which can be determined according to the size of the measured random noise. After the standard deviation and the filter kernel size are determined, the corresponding filter matrix is ​​also determined. Although the conductivity and relative dielectric constant have different waveforms, they are both smooth curves. Using a Gaussian smoothing filter matrix (low-pass filtering) can eliminate noise jitter on the smooth curve. In principle, conductivity and relative dielectric constant can be pre-processed using different filter matrices (filters), but in order to ensure that the method is simple and easy to use, this embodiment chooses to use the same filter matrix to process conductivity and relative dielectric constant while meeting actual needs.

[0077] S113, making each to-be-measured part of the human tissue sample into a corresponding to-be-measured pathological slice, and identifying and classifying each to-be-measured pathological slice according to histopathology to obtain the classification label of each to-be-measured part.

[0078] Among them, the slice tissue used to make the pathological slices to be measured must be consistent with the parts to be measured used for the dielectric property data obtained by measurement, that is, the classification label (tissue site classification) corresponding to each dielectric property data obtained by the above measurement is given through tissue pathology. For example, the labels of liver tumor tissue are: liver cancer tissue, benign lesion tissue, edema tissue, and normal tissue. After the classification label of each dielectric property data is set, the creation of the dielectric property data set of human tissue is completed. After that, it can be randomly divided into training set and verification set according to the preset ratio. If there are 200 groups of measurement data in the dielectric property data set and the preset ratio is 8:2, the training set is 160 groups of data and the verification set is 40 groups of data. After completing the construction of the deep learning neural network, the deep learning neural network can be trained based on the training set using the python language, and tested and verified based on the verification set.

[0079] S12. Constructing a deep learning neural network according to the dispersion frequency characteristics of the training set;

[0080] Among them, the dispersion frequency characteristics are the α, β, δ and γ dispersions of the dielectric properties of human tissue. In different dispersion frequency bands, the factors that affect the dielectric property values ​​are different: in the α dispersion band, the polarization of ions in the cells has a major influence on the dielectric properties; in the β dispersion band, the polarization charge of the cell membrane has a major influence on the dielectric properties; in the δ dispersion band, the polarization of water bound to the protein has a major influence on the dielectric properties; in the γ dispersion band, the polarization of free water molecules has a major influence on the dielectric properties. Based on this objective fact, this embodiment establishes a deep learning neural network that sequentially includes an input layer 1, a dielectric feature extraction layer 2, a dielectric feature fusion layer 3 and an output layer 4 by setting groupings of different frequency bands for the convolution layer and the pooling layer to perform corresponding feature extraction according to the frequency distribution range of the input dielectric property data, and the dielectric feature fusion layer 3 includes multiple fully connected layers, such as Figure 4 After the dielectric property data enters the deep learning neural network through the input layer, it is directly assigned to the corresponding convolution layer group according to the frequency of the data for corresponding feature extraction. Specifically, Figure 5 As shown, the step S12 of constructing a deep learning neural network according to the dispersion frequency characteristics of the training set includes:

[0081] S121, setting the dielectric feature extraction layer according to the relative dielectric constant and conductivity of the dielectric characteristic data in the training set; the dielectric feature extraction layer includes a relative dielectric constant feature extraction layer and a conductivity feature extraction layer;

[0082] Among them, the dielectric characteristics are divided into relative dielectric constant characteristics and conductivity characteristics according to the classification of dielectric property data. In principle, the extraction methods of the corresponding relative dielectric constant characteristics and conductivity characteristics can be determined to be consistent according to needs. In order to ensure the simplicity and consistency of the method application, the dielectric feature extraction layer 2 in the deep learning neural network is divided into a relative dielectric constant feature extraction layer 21 and a conductivity feature 22 in this embodiment. The relative dielectric constant feature extraction layer 21 and the conductivity feature extraction layer 22 are both set to have exactly the same structure according to the following method steps according to different dispersion frequency bands of the dispersion frequency characteristics. The only difference between the two is that the corresponding processed data is different.

[0083] S122. According to the dispersion frequency band of the dispersion frequency characteristic, the relative dielectric constant feature extraction layer and the conductivity feature extraction layer are respectively set; the relative dielectric constant feature extraction layer and the conductivity feature extraction layer both include a frequency band feature extraction layer and a frequency band feature fusion layer; the frequency band feature extraction layer includes a plurality of dispersion frequency band feature extraction layers of different frequency bands; the dispersion frequency band feature extraction layer includes a first convolution layer and a first pooling layer; the frequency band feature fusion layer includes a second convolution layer and a second pooling layer.

[0084] Among them, the dispersion frequency band is divided into the following conventional methods: the relative dielectric constant / conductivity data below 100kHz belongs to the α dispersion frequency band; the relative dielectric constant / conductivity data of 100kHz to 10MHz belongs to the β dispersion frequency band, the relative dielectric constant / conductivity data of 10MHz to 1GHz belongs to the δ dispersion frequency band, and the relative dielectric constant / conductivity data above 1GH belongs to the γ dispersion frequency band. Correspondingly, the frequency band feature extraction layer 211 of the relative dielectric constant feature extraction layer 21 and the frequency band feature extraction layer 221 of the conductivity feature extraction layer 22 both include four dispersion band feature extraction layers: an α dispersion band feature extraction layer, a β dispersion band feature extraction layer, a δ dispersion band feature extraction layer, and a γ dispersion band feature extraction layer, and the corresponding four dispersion band feature extraction layers all include a first convolution layer and a first pooling layer. After the dielectric property data is input into the deep learning neural network, the corresponding dispersion band features are first extracted through the kernel function in the first convolution layer of different dispersion band feature extraction layers, and then output to the corresponding first pooling layer through the ReLU activation function to achieve feature data compression. At the same time, the frequency band feature fusion layer 212 of the relative dielectric constant feature extraction layer 21 and the frequency band feature fusion layer 222 of the conductivity feature extraction layer 22 both include a second convolution layer and a second pooling layer, and the frequency band feature fusion layer (212 / 222) is used to fuse the data features extracted by the four dispersion frequency band feature extraction layers through a concatenate function, and then perform corresponding convolution operations to obtain the corresponding full-band data features. It should be noted that the convolution kernels used in the first convolution layer and the second convolution layer are of different sizes, and the convolution kernel used in the first convolution layer should be larger than the convolution kernel used in the second convolution layer. For example, the four dispersion frequency band feature extraction layers in the frequency band feature extraction layer 211 of the relative permittivity feature extraction layer 21 and the frequency band feature extraction layer 221 of the conductivity feature extraction layer 22 all use convolution kernels of the same size of 7×1, and the frequency band feature fusion layer 212 of the relative permittivity feature extraction layer 21 and the frequency band feature fusion layer 222 of the conductivity feature extraction layer 22 all use convolution kernels of the same size of 3×1. The dielectric feature fusion layer 3 includes multiple fully connected layers for fusing the dielectric constant features and the conductivity features. The specific number of fully connected layers can be determined according to actual needs in principle. In order to improve the learning efficiency while ensuring the fusion effect, the dielectric feature fusion layer 3 is preferably set to two fully connected layers, and after the dielectric features are mapped to the sample classification label space through the two fully connected layers, the final classification result is obtained through the Softmax function of the output layer 4.

[0085] In this embodiment, a deep learning neural network including an input layer 1, a dielectric feature extraction layer 2, a dielectric feature fusion layer 3 and an output layer 4 is set according to the dielectric property data in the training set, and a dielectric feature extraction layer including a relative dielectric constant feature extraction layer and a conductivity feature extraction layer is set according to the relative dielectric constant and the conductivity, and a model construction method of a relative dielectric constant feature extraction layer and a conductivity feature extraction layer, both of which include a frequency band feature extraction layer and a frequency band feature fusion layer, is set according to the dispersion frequency band of the dispersion frequency characteristics, which effectively ensures the practicality of the model, the effectiveness and comprehensiveness of the feature extraction, and further ensures the practicality and accuracy of the deep learning neural network in identifying and classifying human tissues.

[0086] S13, training the deep learning neural network according to the training set to obtain a human tissue identification and classification model;

[0087] Among them, after the deep learning neural network is constructed, the specific process of the human tissue identification and classification model obtained by training with the training set data is as follows. After each training set data is input into the deep learning neural network, the relative dielectric constant and conductivity are first input into the corresponding relative dielectric constant feature extraction layer and conductivity feature extraction layer for feature extraction processing, and the dispersion frequency band feature extraction layers corresponding to the dielectric constant feature extraction layer and the conductivity feature extraction layer in different dispersion feature groupings are processed in sequence by kernel function, ReLU activation function and pooling to obtain different dispersion frequency band features, and then sent to the corresponding frequency band feature fusion layer and processed in sequence by the second convolution layer and the second pooling layer. The corresponding relative dielectric constant characteristics and conductivity characteristics are obtained; then the dielectric feature fusion layer including multiple fully connected layers is entered to fuse the relative dielectric constant characteristics and conductivity characteristics, and mapped to the corresponding sample classification label space; then it is sent to the output layer, and the final classification result is obtained through the Softmax function; finally, the loss function is calculated according to the corresponding classification result and the original classification label, and the parameters of the deep learning neural network are updated according to the preset training rules until the model converges to obtain a stable human tissue identification and classification model, which can be directly used for subsequent human tissue identification and classification. The specific loss function and parameter updating method can be implemented by existing technology, which will not be repeated here.

[0088] S14, inputting the dielectric property data to be identified in the verification set into the human tissue identification and classification model for identification and classification to obtain corresponding classification results.

[0089] Among them, the dielectric property data depends on the measurement frequency, and the dielectric property data at different measurement frequencies are different. In addition, since the human tissue identification and classification model is trained based on the training set data, in order to ensure the accuracy of identification and classification, when using the human tissue identification and classification model to classify the validation set or other human tissues to be identified and classified, it is necessary to first detect whether the measurement frequency of the human tissue data to be identified and classified is consistent with the measurement frequency in the training set. When the measurement frequency of the dielectric property data of the tissue to be identified is inconsistent with the measurement frequency of the training data set, it is necessary to perform corresponding normalization processing first, and then use the human tissue identification and classification model for identification and classification. Specifically, Figure 6 As shown, the step S14 of inputting the dielectric property data to be identified in the verification set into the human tissue identification and classification model for identification and classification to obtain the corresponding classification result includes:

[0090] S141, determining whether the measurement frequency of the dielectric property data to be identified is consistent with the measurement frequency of the dielectric property data in the training set;

[0091] S142. If the measurement frequency of the dielectric property data to be identified is consistent with the measurement frequency of the dielectric property data in the training set, the human tissue identification and classification model is used to identify and classify the dielectric property data to be identified to obtain the classification result. Otherwise, the dielectric property data to be identified is interpolated and fitted by the Cole-Cole model to obtain normalized dielectric property data to be identified, and the human tissue identification and classification model is used to identify and classify the normalized dielectric property data to be identified to obtain the classification result.

[0092] Among them, the interpolation fitting of the dielectric property data to be identified is to first fit the Cole-Cole model (fourth-order model) with existing data, determine the undetermined coefficients in the model, and obtain the Cole-Cole interpolation model that can be used for the interpolation of the dielectric property data to be identified, and then input the measurement frequency of the dielectric property data to be identified into the Cole-Cole interpolation model to obtain the normalized dielectric property data to be identified, specifically including:

[0093] The Cole-Cole model is fitted using a nonlinear optimization algorithm to obtain a Cole-Cole interpolation model; the Cole-Cole model is expressed as:

[0094]

[0095] In the formula, ε * represents the complex dielectric constant; σ and ε r Respectively represent the measured conductivity and relative dielectric constant; f represents the measurement frequency; ε0 represents the vacuum dielectric constant; j represents the sign of the complex imaginary part; σ sIndicates DC conductivity; ε ∞ Indicates the relative dielectric constant when the frequency is infinite; Δε n , τ n and α n They represent the relative permittivity increment, average relaxation time and relaxation time distribution coefficient corresponding to the nth dispersion frequency band, and n takes the values ​​of 1, 2, 3 and 4;

[0096] Among them, the Cole-Cole model can be implemented by using the existing technology using a nonlinear optimization algorithm fitting method. In order to improve the accuracy of the fourth-order Cole-Cole model fitting and thereby ensure the accuracy of the normalized dielectric property data to be identified, this embodiment preferably expands the fourth-order Cole-Cole model into four independent Cole-Cole frequency band models based on the dispersion frequency characteristics of the dielectric properties of human tissue, and fits them respectively to determine the corresponding Cole-Cole frequency band interpolation models, and then obtains the final Cole-Cole interpolation model. Specifically, the steps of fitting the Cole-Cole model using a nonlinear optimization algorithm and determining the undetermined coefficients of the Cole-Cole model include:

[0097] According to the dispersion frequency characteristics of the dielectric properties of human tissue, the Cole-Cole model is expanded according to the prior information to obtain the Cole-Cole frequency band model corresponding to each dispersion frequency characteristic;

[0098] Each Cole-Cole frequency band model is fitted separately to obtain the corresponding Cole-Cole frequency band interpolation model;

[0099] The Cole-Cole interpolation model is obtained according to each Cole-Cole frequency band interpolation model.

[0100] Specifically, taking the model in the α dispersion band as an example, the steps of fitting the Cole-Cole band model to obtain the Cole-Cole band interpolation model are introduced in detail.

[0101] Determine the model parameter σ in the α dispersion band with a frequency f range of 100Hz to 100kHz s , Δε1, τ1 and α1, the corresponding Cole-Cole frequency band model is:

[0102]

[0103] Set the corresponding cost function F1 to:

[0104]

[0105] The summation range is all measurement points within the frequency range f.

[0106] After the cost function F1 is set, a nonlinear iterative optimization algorithm is used to minimize the F1 value. Specifically, the steepest descent method is used to implement iterative optimization: ① Set σ s The initial values ​​of Δε1, τ1 and α1 are preferably parameters that are close to those of biological tissues and have been reported in research, σ s =0.7, Δε1=56, τ1=8.38×10 -12 , and α1 = 0.1; ② Solve the gradient direction based on the expression of the cost function F1 ③ Use one-dimensional exact search method to determine the step size λ; ④ Use and Update σ separately s , Δε1, τ1 and α1, and repeat ②③④ until the preset termination condition is met, and the final σ s , Δε1, τ1, α1, and the corresponding Cole-Cole band interpolation model in the α dispersion band is obtained.

[0107] Similarly, the above method steps are used to obtain the Cole-Cole frequency band interpolation models corresponding to the β dispersion frequency band with a frequency f range of 100kHz to 10MHz, the δ dispersion frequency with a frequency f range of 10MHz to 1GHz, and the γ dispersion frequency with a frequency f range of 1GHz to 20GHz, respectively. Then, based on the Cole-Cole frequency band interpolation models obtained under the four dispersion frequencies, the final Cole-Cole interpolation model is aggregated to obtain the normalized dielectric property data to be identified using the following method steps.

[0108] Based on the objective fact that the dielectric properties of biological tissues have dispersion frequency characteristics, this embodiment independently fits the Cole-Cole model expressions of various orders corresponding to different dispersion frequency bands, thereby reducing the fitting complexity and improving the accuracy of the fitting results. In addition, during the iterative optimization process, the initial values ​​are set using parameter information that is close to the biological tissues and has been reported in research, so that the optimal solution can be obtained through fewer iterations, thereby improving the fitting efficiency.

[0109] The normalized dielectric property data to be identified are obtained according to the Cole-Cole interpolation model and the measurement frequency of the dielectric property data to be identified.

[0110] After the normalized dielectric property data to be identified is obtained by adopting the above method steps, it can be input into the trained human tissue identification and classification model for identification and classification to obtain the corresponding classification result.

[0111] The embodiment of the present application establishes a dielectric property data set by measuring the dielectric property data of different parts of a human tissue sample and determining the corresponding dielectric property data classification labels based on pathological histology, and is used to train a stable human tissue identification and classification model for human tissue identification based on a deep learning convolutional neural network constructed based on the dispersion frequency characteristics of the dielectric property data. In addition, when using the human tissue identification and classification model to identify and classify dielectric property data to be identified that is inconsistent with the measurement frequency of the training set, a fourth-order Cole-Cole interpolation model is obtained by pre-fitting each order Cole-Cole model corresponding to different dispersion frequency bands for frequency normalization, and then input into the human tissue identification and classification model for identification and classification. The method not only has strong practicality, objectivity and repeatability, but also effectively guarantees high learning efficiency and classification accuracy.

[0112] It should be noted that although the steps in the above flowchart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0113] In one embodiment, Figure 7 As shown, a human tissue identification and classification system is provided, the system comprising:

[0114] Acquisition module 1 is used to acquire a dielectric property data set of human tissue, and divide the dielectric property data set into a training set and a validation set according to a preset ratio; the dielectric property data set includes dielectric property data and corresponding classification labels;

[0115] Modeling module 2, used to construct a deep learning neural network according to the dispersion frequency characteristics of the training set;

[0116] A training module 3 is used to train the deep learning neural network according to the training set to obtain a human tissue identification and classification model;

[0117] The classification module 4 is used to input the dielectric property data to be identified in the verification set into the human tissue identification and classification model for identification and classification to obtain corresponding classification results.

[0118] For the specific definition of the human tissue identification and classification system, please refer to the definition of the human tissue identification and classification method above, which will not be repeated here. Each module in the above human tissue identification and classification system can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0119] Figure 8 FIG. 1 shows an internal structure diagram of a computer device in an embodiment, and the computer device may specifically be a terminal or a server. Figure 8 As shown, the computer device includes a processor, a memory, a network interface, a display and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a human tissue identification and classification method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0120] It can be understood by those skilled in the art that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the present application scheme, and does not constitute a limitation on the computer device to which the present application scheme is applied. The specific computing device may include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0121] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0122] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0123] In summary, the embodiments of the present invention provide a method, system, computer device and storage medium for identifying and classifying human tissues. The method for identifying and classifying human tissues realizes the construction of a dielectric property data set of human tissues by measuring dielectric property data of different parts of human tissue samples by adopting an open-end coaxial method, and determining the classification labels corresponding to pathological sections of each measured part according to tissue pathology, and after dividing the training set and the validation set according to a preset ratio, a deep learning neural network including an input layer, a dielectric feature extraction layer, a dielectric feature fusion layer and an output layer is constructed according to the dispersion frequency characteristics of the dielectric properties of human tissues, and the deep learning neural network is trained according to the training set to obtain a human tissue identification and classification model, and when using the human tissue identification and classification model to identify and classify dielectric property data to be identified that is inconsistent with the measurement frequency of the training set, a fourth-order Cole-Cole interpolation model is obtained by pre-fitting each order Cole-Cole model corresponding to different dispersion frequency bands, and then input into the human tissue identification and classification model for identification and classification to obtain a corresponding classification result. Technical solution. This human tissue identification and classification method effectively solves the technical defects of existing human tissue identification and classification in actual human tissue identification and classification applications. It not only has strong practicality, objectivity and repeatability, but also effectively guarantees high learning efficiency and classification accuracy.

[0124] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0125] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A method for identifying and classifying human tissues, characterized in that: The method comprises the following steps: Acquire a dielectric property data set of human tissue, and divide the dielectric property data set into a training set and a validation set according to a preset ratio; the dielectric property data set includes dielectric property data and corresponding classification labels; Constructing a deep learning neural network according to the dispersion frequency characteristics of the training set; Training the deep learning neural network according to the training set to obtain a human tissue identification and classification model; Inputting the dielectric property data to be identified in the verification set into the human tissue identification and classification model for identification and classification to obtain corresponding classification results; The deep learning neural network includes an input layer, a dielectric feature extraction layer, a dielectric feature fusion layer and an output layer in sequence; the dielectric feature fusion layer includes a plurality of fully connected layers; The step of constructing a deep learning neural network according to the dispersion frequency characteristics of the training set includes: According to the relative dielectric constant and conductivity of the dielectric characteristic data in the training set, the dielectric feature extraction layer is set; the dielectric feature extraction layer includes a relative dielectric constant feature extraction layer and a conductivity feature extraction layer; According to the dispersion frequency band of the dispersion frequency characteristic, the relative dielectric constant feature extraction layer and the conductivity feature extraction layer are respectively set; the relative dielectric constant feature extraction layer and the conductivity feature extraction layer both include a frequency band feature extraction layer and a frequency band feature fusion layer; the frequency band feature extraction layer includes a plurality of dispersion frequency band feature extraction layers of different frequency bands; the dispersion frequency band feature extraction layer includes a first convolution layer and a first pooling layer; the frequency band feature fusion layer includes a second convolution layer and a second pooling layer; The step of inputting the dielectric property data to be identified in the verification set into the human tissue identification and classification model for identification and classification to obtain the corresponding classification result comprises: Determining whether the measurement frequency of the dielectric property data to be identified is consistent with the measurement frequency of the dielectric property data in the training set; If the measurement frequency of the dielectric property data to be identified is consistent with the measurement frequency of the dielectric property data in the training set, the human tissue identification and classification model is used to identify and classify the dielectric property data to be identified to obtain the classification result; otherwise, the dielectric property data to be identified is interpolated and fitted by the Cole-Cole model to obtain normalized dielectric property data to be identified, and the human tissue identification and classification model is used to identify and classify the normalized dielectric property data to be identified to obtain the classification result; The step of performing interpolation fitting on the dielectric property data to be identified by using the Cole-Cole model to obtain normalized dielectric property data to be identified comprises: The Cole-Cole model is fitted using a nonlinear optimization algorithm to obtain a Cole-Cole interpolation model; the Cole-Cole model is expressed as: In the formula, represents the complex dielectric constant; and Respectively represent the measured conductivity and relative dielectric constant; f represents the measurement frequency; represents the vacuum dielectric constant; j represents the sign of the complex imaginary part; It represents the DC conductivity; Represents the relative dielectric constant when the frequency is infinite; , and They represent the relative permittivity increment, average relaxation time and relaxation time distribution coefficient corresponding to the nth dispersion frequency band, and n takes the values ​​of 1, 2, 3 and 4; Obtaining the normalized dielectric property data to be identified according to the Cole-Cole interpolation model and the measurement frequency of the dielectric property data to be identified; The step of fitting the Cole-Cole model using a nonlinear optimization algorithm to determine the undetermined coefficients of the Cole-Cole model comprises: According to the dispersion frequency characteristics of the dielectric properties of human tissue, the Cole-Cole model is expanded according to the prior information to obtain the Cole-Cole frequency band model corresponding to each dispersion frequency characteristic; Each Cole-Cole frequency band model is fitted separately to obtain the corresponding Cole-Cole frequency band interpolation model; The Cole-Cole interpolation model is obtained according to each Cole-Cole frequency band interpolation model.

2. The human tissue identification and classification method according to claim 1, characterized in that: The step of obtaining a dielectric property data set of human tissue comprises: Obtaining a human tissue sample, and determining a part to be measured based on the human tissue sample; the human tissue sample includes a resected tumor and surrounding tissue of the tumor; Using an open-end coaxial method, repeatedly measuring the dielectric properties of each to-be-measured part of the human tissue sample, obtaining reflection coefficients corresponding to a preset number of groups, and calculating the corresponding dielectric property data based on the reflection coefficients; Each part to be measured of the human tissue sample is made into a corresponding pathological section to be measured, and each pathological section to be measured is identified and classified according to histopathology to obtain the classification label of each part to be measured.

3. The human tissue identification and classification method according to claim 2, characterized in that: The step of calculating the corresponding dielectric characteristic data according to the reflection coefficient comprises: Calibration determines the characteristic parameters of the open-ended coaxial probe; According to the characteristic parameters of the open-ended coaxial probe and the reflection coefficients of the various parts to be measured of the human tissue sample, the initial dielectric characteristic data corresponding to the preset number of groups are calculated; the initial dielectric characteristic data is expressed as: In the formula, f and Respectively represent the measurement frequency and the measured reflection coefficient; , and All represent characteristic parameters of open-ended coaxial probes; and Respectively represent the real part operation and the imaginary part operation; and Respectively represent the relative dielectric constant and conductivity of the part to be measured; The initial dielectric property data of each to-be-measured part of the human tissue sample are preprocessed by using mean filtering and low-pass filtering in sequence to obtain the corresponding dielectric property data.

4. A human tissue identification and classification system, characterized in that: The system comprises: An acquisition module, used to acquire a dielectric property data set of human tissue, and divide the dielectric property data set into a training set and a validation set according to a preset ratio; the dielectric property data set includes dielectric property data and corresponding classification labels; A modeling module, used for constructing a deep learning neural network according to the dispersion frequency characteristics of the training set; A training module, used to train the deep learning neural network according to the training set to obtain a human tissue identification and classification model; A classification module, used for inputting the dielectric property data to be identified in the verification set into the human tissue identification and classification model for identification and classification, and obtaining corresponding classification results; The deep learning neural network includes an input layer, a dielectric feature extraction layer, a dielectric feature fusion layer and an output layer in sequence; the dielectric feature fusion layer includes a plurality of fully connected layers; The step of constructing a deep learning neural network according to the dispersion frequency characteristics of the training set includes: According to the relative dielectric constant and conductivity of the dielectric characteristic data in the training set, the dielectric feature extraction layer is set; the dielectric feature extraction layer includes a relative dielectric constant feature extraction layer and a conductivity feature extraction layer; According to the dispersion frequency band of the dispersion frequency characteristic, the relative dielectric constant feature extraction layer and the conductivity feature extraction layer are respectively set; the relative dielectric constant feature extraction layer and the conductivity feature extraction layer both include a frequency band feature extraction layer and a frequency band feature fusion layer; the frequency band feature extraction layer includes a plurality of dispersion frequency band feature extraction layers of different frequency bands; the dispersion frequency band feature extraction layer includes a first convolution layer and a first pooling layer; the frequency band feature fusion layer includes a second convolution layer and a second pooling layer; The step of inputting the dielectric property data to be identified in the validation set into the human tissue identification and classification model for identification and classification to obtain corresponding classification results includes: Determining whether the measurement frequency of the dielectric property data to be identified is consistent with the measurement frequency of the dielectric property data in the training set; If the measurement frequency of the dielectric property data to be identified is consistent with the measurement frequency of the dielectric property data in the training set, the human tissue identification and classification model is used to identify and classify the dielectric property data to be identified to obtain the classification result; otherwise, the dielectric property data to be identified is interpolated and fitted by the Cole-Cole model to obtain normalized dielectric property data to be identified, and the human tissue identification and classification model is used to identify and classify the normalized dielectric property data to be identified to obtain the classification result; The interpolation fitting of the dielectric property data to be identified by the Cole-Cole model to obtain normalized dielectric property data to be identified includes: The Cole-Cole model is fitted using a nonlinear optimization algorithm to obtain a Cole-Cole interpolation model; the Cole-Cole model is expressed as: In the formula, represents the complex dielectric constant; and Respectively represent the measured conductivity and relative dielectric constant; f represents the measurement frequency; represents the vacuum dielectric constant; j represents the sign of the complex imaginary part; It represents the DC conductivity; Represents the relative dielectric constant when the frequency is infinite; , and They represent the relative permittivity increment, average relaxation time and relaxation time distribution coefficient corresponding to the nth dispersion frequency band, and n takes the values ​​of 1, 2, 3 and 4; Obtaining the normalized dielectric property data to be identified according to the Cole-Cole interpolation model and the measurement frequency of the dielectric property data to be identified; The adopting of a nonlinear optimization algorithm to fit the Cole-Cole model to determine the undetermined coefficients of the Cole-Cole model includes: According to the dispersion frequency characteristics of the dielectric properties of human tissue, the Cole-Cole model is expanded according to the prior information to obtain the Cole-Cole frequency band model corresponding to each dispersion frequency characteristic; Each Cole-Cole frequency band model is fitted separately to obtain the corresponding Cole-Cole frequency band interpolation model; The Cole-Cole interpolation model is obtained according to each Cole-Cole frequency band interpolation model.

5. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images