A method for detecting anti-tumor by bioelectrical impedance based on pattern recognition

Abstract

The present invention discloses a method for detecting anti-tumor by bioelectrical impedance based on pattern recognition, which includes the following steps: S1: Use a bioelectrical impedance analysis instrument to measure the impedance of the tumor area to be detected, and obtain the current, voltage and impedance values at different depths and frequencies; S2: Design a feature extraction module to extract the key features in the impedance data at different frequencies of each channel; S3: Input the extracted key features into a multi-level feature fusion module to fuse the impedance features at different channels and frequencies to form new fused features; S4: Input the fused features into an adaptive classification module and adjust the model parameters by combining an adaptive optimization method. The present invention can automatically process and classify data through a pattern recognition algorithm, and at the same time, by combining a feature extraction module, a multi-level feature fusion module and an adaptive classification module, it can effectively improve the accuracy, sensitivity and reliability of tumor detection.

Description

Technical Field

[0001] The present invention relates to the field of biomedical detection, and particularly to a method for detecting anti-tumor by bioelectrical impedance based on pattern recognition. Background Art

[0002] As a disease that seriously threatens human health, the high incidence and high mortality rate of tumors have become a major challenge faced by global public health. Especially malignant tumors, that is, what we often call cancer, have become one of the main causes of death. According to the data of the World Health Organization, cancer claims millions of lives every year, and most patients are in the advanced stage when they are discovered, missing the best treatment opportunity. Therefore, the early detection and timely treatment of tumors are of crucial significance for improving the cure rate and patient survival rate. At present, common tumor detection methods include CT, MRI, ultrasound, and tissue biopsy, etc. However, these traditional detection methods usually require expensive equipment, complex operations, and long detection cycles, and may bring certain radiation risks or a sense of pain.

[0003] Bioelectrical impedance measurement is a method for analyzing the physical properties and health status of tissues by measuring the impedance characteristics of tissues. The change in impedance can reflect the differences between different types of tissues, especially the significant impedance differences between tumor tissues and normal tissues. By measuring impedance data, early screening and diagnosis of tumors can be achieved. However, traditional impedance detection methods usually rely on manual analysis, with low accuracy and efficiency. With the development of machine learning and pattern recognition technologies, pattern recognition-based analysis methods can significantly improve the accuracy and automation level of tumor detection.

[0004] In the prior art, Chinese Patent No. CN118864354A discloses "a method for detecting multi-modal brain tumors based on unsupervised contrast learning", which combines the feature extraction of MRI and CT images with unsupervised contrast learning technology, improves the detection accuracy of brain tumors, and reduces the dependence on labeled data. However, this method mainly relies on the feature fusion and contrast learning of MRI and CT images, and fails to fully consider the integration of other modal data or additional features.

[0005] Therefore, there is an urgent need to design a method for detecting anti-tumor by bioelectrical impedance based on pattern recognition to solve the problems existing in the above-mentioned prior art. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for detecting anti-tumor by bioelectrical impedance based on pattern recognition. Through pattern recognition algorithms, the data is automatically processed and classified, and at the same time, combined with a feature extraction module, a multi-level feature fusion module, and an adaptive classification module, it can effectively improve the accuracy, sensitivity, and reliability of tumor detection.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect of the present invention, a method for anti-tumor detection of bioelectrical impedance based on pattern recognition is provided, and the method includes the following steps:

[0009] S1: Perform impedance measurement on the tumor region to be detected through a bioelectrical impedance analysis instrument to obtain current, voltage, and impedance values at different depths and different frequencies;

[0010] S2: Design a feature extraction module to extract key features from the impedance data at different frequencies of each channel;

[0011] S3: Input the extracted key features into a multi-level feature fusion module to fuse the impedance features at different channels and different frequencies to form new fused features;

[0012] S4: Input the fused features into an adaptive classification module and combine with an adaptive optimization method to adjust the model parameters and improve the classification accuracy and sensitivity.

[0013] As an embodiment of the present application, the step S1 specifically includes:

[0014] S11: Configure a pair of double electrodes, apply current respectively and measure voltage;

[0015] S12: Introduce the current applied by the pair of double electrodes into the tumor region to be detected respectively, and stimulate the bio-tissue impedance response of this region through current signals within different frequency ranges;

[0016] S13: Use a voltage electrode to measure the voltage signal of the tumor region, and through processing such as signal amplification and filtering, obtain voltage values at different frequencies;

[0017] S14: Each channel respectively calculates the ratio of voltage to the applied current to obtain impedance values at different frequencies and generates a frequency-impedance spectrum.

[0018] As an embodiment of the present application, the step S14 specifically includes:

[0019] S141: For each frequency point, calculate the impedance value using the known Ohm's law, and its calculation formula is as follows;

[0020]

[0021] Among them, is the resistance component, is the reactive component reactance, is the imaginary unit;

[0022] S142: Extract the amplitude, phase, and impedance characteristic parameters by performing time-domain and frequency-domain analyses on the signal;

[0023] S143: Integrate the impedance values at different frequencies and plot a spectrogram of the impedance versus frequency;

[0024] S144: Use adaptive filtering technology on the frequency-impedance spectrogram, dynamically adjust the filter parameters according to the statistical characteristics of the noise, and suppress different types of noise and interference. The calculation formula is as follows:

[0025]

[0026]

[0027]

[0028]

[0029] where, represents the difference between the true output value and the filter output value at time , is the target value at time , is the filter coefficient, representing the filter parameter at time , is the input signal vector, representing the input data at time , is the gain vector, is the forgetting factor, is the inverse covariance matrix at the previous time, is the updated inverse covariance matrix.

[0030] As an embodiment of the present application, the step S2 specifically includes:

[0031] S21: Perform frequency-domain analysis on the impedance data at different frequencies, convert the time-domain signal into a frequency-domain signal for the data of each channel through a frequency-domain transformation method, and obtain the amplitude and phase information of the impedance data at each frequency point;

[0032] S22: Calculate the amplitude characteristics in the impedance data, including the modulus of the impedance and the phase angle , and the calculation formula is as follows:

[0033]

[0034]

[0035] where, is the resistance, i.e., the real part of the impedance, is the reactance, i.e., the imaginary part of the impedance, is the arctangent function, which can calculate the phase angle based on the real part and the imaginary part of the impedance; the range of the phase angle is ;

[0036] S23: Extract the resistivity, conductivity, reactance and other electrical parameters as features according to the relationship between the impedance amplitude and the phase angle. The calculation formulas are as follows:

[0037]

[0038]

[0039]

[0040] where, is the real part of the impedance, is the cross-sectional area, is the distance between the electrodes, represents the imaginary part of the impedance ;

[0041] S24: Use the weighted variance to extract the key information of the frequency-impedance spectrum. The calculation formula is as follows:

[0042]

[0043]

[0044] where, represents the weighted variance, represents the number of frequency sample points, is the th frequency point, is the weighting factor for each frequency point, used to assign different weights to different frequency points, represents the th impedance value corresponding to the frequency point, is the weighted mean value of the impedance values of all frequency points;

[0045] S25: Use the adaptive sliding window algorithm to extract local features from the impedance data;

[0046] S26: Perform dimensionality reduction on the extracted local features to obtain the key features in the impedance data at different frequencies for each channel.

[0047] As an embodiment of the present application, the step S21 specifically includes:

[0048] S211: After transforming the time-domain signal by a frequency-domain transformation method, the obtained frequency range is divided into n sub-bands, and its calculation formula is as follows:

[0049]

[0050]

[0051] where, represents the complex value of the signal at the frequency and time position in the time-frequency plane, is the time-domain representation of the signal, and is just a sample point of the signal, representing the value of the signal at time , is the window function, represents the time-domain shift of the window function, and its center is located at time, represents the complex modulation of the frequency components, describing how the frequency components of the signal change at different times , is the imaginary unit, is the th frequency of the sub-band, and are the minimum and maximum probabilities respectively, is the number of sub-bands;

[0052] S212: Process each element of the frequency-domain signal data matrix for each sub-band, and calculate the amplitude and phase feature of the sub-band, and its calculation formula is as follows:

[0053]

[0054]

[0055] where, and are the real part and the imaginary part of the signal at the frequency respectively, is the arctangent function, and calculates the angle between the point and the axis;

[0056] S213: For each element of the frequency-domain signal data matrix, calculate the corresponding element of the response matrix, and its calculation formula is as follows:

[0057]

[0058] Among them, is the response matrix element, representing the response characteristic of the signal at frequency . is the gain and phase shift of the function with respect to the signal at frequency . is the complex value representing the original signal at frequency . is the imaginary unit, is the phase of the original signal at frequency . is the phase of the transfer function at frequency . The sum of the phase parts represents the total phase shift of the signal and the system.

[0059] As an embodiment of the present application, step S3 specifically includes:

[0060] S31: Input the key features, and use a cross-channel attention mechanism to calculate the attention weights. The calculation formula is as follows:

[0061]

[0062]

[0063] Among them, represents the similarity between channel and channel . and are the signal representations of the th and th channels respectively. and are the L2 norms of and respectively. represents the attention weight of the th channel. After softmax, the sum of the attention weights is 1. represents the number of channels;

[0064] S32: According to the calculated attention weights, adjust the contribution degree of each channel feature, perform weighted averaging on the features of different channels, and dynamically adjust the influence of each channel according to the attention weights;

[0065] S33: Perform weighted averaging on all the weighted features to form a new fused feature representation.

[0066] As an embodiment of the present application, step S4 specifically includes:

[0067] S41: Input the fused feature data into the adaptive classification module, and select the convolutional long short-term memory network combined with the adaptive optimization method to train the model;

[0068] S42: Tune the hyperparameters of the trained model, and use cross-validation to evaluate the performance of the model under different parameter configurations.

[0069] As an embodiment of the present application, the step S41 specifically includes:

[0070] S411: Input the fused features into the convolutional layer to extract high-level abstract features. The convolutional layer includes 1×1 one-dimensional convolution, ReLU activation function, 3×3 one-dimensional convolution, ReLU activation function, and pooling layer;

[0071] S412: The features processed by the convolutional layer are fed into the convolutional long short-term memory network module to further extract the temporal features in the sequence data;

[0072] S413: Flatten the features after passing through the convolutional long short-term memory network module and then summarize to obtain the temporal information, and finally feed it into the fully connected layer;

[0073] S414: To ensure the robustness of the model, use the He method for model initialization to avoid the problems of gradient disappearance or gradient explosion;

[0074] S415: Use the adaptive optimization algorithm to dynamically adjust the model parameters and optimize the classification accuracy. Its calculation formula is as follows:

[0075]

[0076]

[0077] where, is the exponential decay average of the square of the gradient, is the current gradient, is the current parameter, is the updated parameter, is the learning rate, is the decay factor, set to 0.9, is a constant used to prevent division by zero;

[0078] S416: Use a smaller learning rate at the beginning of training, gradually increase the learning rate to avoid slow convergence, and then gradually decrease the learning rate as the number of training rounds increases to avoid premature convergence or oscillation near the local minimum;

[0079] S417: Introduce the ensemble learning strategy of gradient boosting trees. Using the method of weighted average, generate a preliminary prediction result by gradually training weak learners and combining their outputs.

[0080] S418: In the real-time adaptive adjustment and online learning stage, based on the newly input data and feedback, continue to update the model parameters to improve the classification accuracy. The calculation formula is as follows:

[0081]

[0082] where, are the updated model parameters, are the model parameters at the current time , is the learning rate, which controls the step size of model parameter update and determines the amplitude of each adjustment. is the loss function, are the currently received samples and labels, represents the gradient of the loss function with respect to the model parameters , indicating the rate of change of the loss value with respect to the model parameters.

[0083] The beneficial effects of the present invention are as follows:

[0084] (1) The present invention combines bioelectrical impedance analysis technology with pattern recognition methods, can accurately extract the electrical signal characteristics of tumor tissues at different frequencies, effectively distinguish normal tissues from tumor tissues, further distinguish between benign and malignant tumors, and in some cases can assist in distinguishing the specific types of tumors. By integrating feature extraction and intelligent pattern recognition methods, the present invention greatly improves the accuracy of tumor detection and reduces the incidence of misdiagnosis and missed diagnosis.

[0085] (2) By designing a feature extraction module, the present invention can extract highly distinguishable key features from the impedance data of each channel and at different frequencies. By performing time-domain and frequency-domain analysis on the electrical signals, electrical property parameters such as conductivity, reactance, and resistance are extracted. These features reflect the electrical properties of tissues and can effectively distinguish normal tissues from tumor tissues. During the extraction process, the module can select the most representative features according to the response characteristics at different frequencies, remove noise and redundant information, so as to ensure that the extracted features have strong discriminative ability.

[0086] (3) By designing a multi-level feature fusion module, the present invention can effectively integrate information from different depths, frequencies, and channels, fuse it into a more comprehensive and accurate representation, further remove redundant and noise information, and extract global features with high discriminative ability; these fused features can more accurately reflect the electrical properties of breast tissue, providing a more precise feature representation for subsequent classification tasks; by fusing information at different levels, the differences between tumors and normal tissues can be captured more comprehensively, enhancing the robustness and accuracy of the classifier. Through multi-level fusion, the present invention can effectively capture the electrical features of tumor regions, reduce noise interference, and enhance the discriminative ability of the model.

[0087] (4) By designing an adaptive classification module, the present invention can automatically adjust hyperparameters such as the learning rate and weight decay according to the characteristics of different tumor samples, continuously optimize the performance of the model through training and validation data. Through the ensemble learning method of gradient boosting trees, the model can improve the accuracy and stability of classification from multiple perspectives. Each type of tumor (benign, malignant, or different types of tumors) has its unique electrical characteristics, and the classification module can effectively distinguish different types of tissues based on these characteristics. Through this method, not only can normal tissues and tumor tissues be effectively distinguished, but also they can be further refined into categories such as benign tumors and malignant tumors, providing more accurate tumor detection and diagnosis information, and providing a reliable basis for subsequent treatment decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a schematic flow chart of a bioelectrical impedance anti-tumor detection method based on pattern recognition provided in an embodiment of the present invention;

[0089] Figure 2 It is a schematic flow chart of a feature extraction module of a bioelectrical impedance anti-tumor detection method based on pattern recognition provided in an embodiment of the present invention;

[0090] Figure 3 It is a schematic flow chart of a multi-level feature fusion module of a bioelectrical impedance anti-tumor detection method based on pattern recognition provided in an embodiment of the present invention;

[0091] Figure 4 It is a schematic diagram of an adaptive classification module of a bioelectrical impedance anti-tumor detection method based on pattern recognition provided in an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0092] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0093] It should be noted that all directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative position relationship and movement conditions between components in a specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indications will also change accordingly.

[0094] In the present invention, unless otherwise clearly defined and limited, terms such as "connection" and "fixation" should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and can be the communication inside two components or the interaction relationship between two components, unless otherwise clearly limited. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0095] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one such feature. In addition, the meaning of "and / or" appearing throughout the text includes three parallel solutions. Taking "A and / or B" as an example, it includes solution A, solution B, or the solution where A and B are satisfied simultaneously. In addition, the technical solutions between the embodiments can be combined with each other, but it must be based on the fact that those of ordinary skill in the art can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the protection scope required by the present invention.

[0096] Referring to Figures 1 to 4 , the first aspect of the present invention provides a method for detecting anti-tumor by bioelectrical impedance based on pattern recognition. The method includes the following steps:

[0097] S1: Use a bioelectrical impedance analysis instrument to measure the impedance of the tumor area to be detected, and obtain the current, voltage, and impedance values at different depths and different frequencies;

[0098] S2: Design a feature extraction module to extract the key features from the impedance data at different frequencies of each channel;

[0099] S3: Input the extracted key features into the multi-level feature fusion module to fuse the impedance features at different channels and frequencies to form new fused features.

[0100] S4: Input the fused features into the adaptive classification module and combine with the adaptive optimization method to adjust the model parameters and improve the classification accuracy and sensitivity.

[0101] The present invention combines bioelectrical impedance analysis technology with pattern recognition method, which can accurately extract the electrical signal features of tumor tissues at different frequencies, effectively distinguish normal tissues from tumor tissues, further distinguish benign tumors from malignant tumors, and in some cases can assist in distinguishing the specific types of tumors. By fusing feature extraction and intelligent pattern recognition methods, the present invention greatly improves the accuracy of tumor detection, reduces the incidence of misdiagnosis and missed diagnosis. The pattern recognition method adopted by the present invention has good interpretability, can provide a transparent decision-making process for doctors, and assist clinicians to make more accurate diagnostic decisions.

[0102] As an embodiment of the present application, step S1 specifically includes:

[0103] S11: Configure double electrode pairs, apply currents respectively and measure voltages.

[0104] S12: Introduce the currents applied by the double electrode pairs into the tumor area to be detected respectively, and stimulate the bio-tissue impedance response of this area through current signals in different frequency ranges.

[0105] S13: Use voltage electrodes to measure the voltage signals of the tumor area, and through processing such as signal amplification and filtering, obtain voltage values at different frequencies.

[0106] S14: Each channel calculates the ratio of voltage to applied current respectively to obtain impedance values at different frequencies and generate a frequency-impedance spectrum.

[0107] As an embodiment of the present application, step S14 specifically includes:

[0108] S141: For each frequency point, calculate the impedance value using the known Ohm's law, and its calculation formula is as follows;

[0109]

[0110] Where, is the resistance component, is the reactive component reactance, is the imaginary unit;

[0111] S142: Through time-domain and frequency-domain analysis of the signal, extract its amplitude, phase and impedance characteristic parameters.

[0112] S143: Integrate the impedance values at different frequencies and plot a spectrogram of the impedance versus frequency;

[0113] S144: Use adaptive filtering technology on the frequency-impedance spectrum, dynamically adjust the filter parameters according to the statistical characteristics of the noise, and suppress different types of noise and interference. The calculation formula is as follows:

[0114]

[0115]

[0116]

[0117]

[0118] Wherein, represents the difference between the true output value and the filter output value at time , is the target value at time , is the filter coefficient, representing the filter parameter at time , is the input signal vector, representing the input data at time , is the gain vector, is the forgetting factor, is the inverse covariance matrix at the previous moment, is the updated inverse covariance matrix.

[0119] Specifically, the present invention uses a multi-frequency bioelectrical impedance analyzer to obtain the electrical signal data of the detected tumor area. By measuring and analyzing the current, voltage, and impedance values at different depths and different frequencies, important electrical parameters such as the conductivity, resistance, and capacitance of the tumor tissue can be extracted. These electrical parameters have significant differences between tumor tissue and normal tissue. Because of the abnormal proliferation of tumor cells and the change of tissue structure, it usually leads to changes in electrical properties. Therefore, compared with traditional tumor diagnosis methods (such as puncture, CT, or MRI examination), the present invention does not require invasive operations, reduces the discomfort and risks of patients. This method collects electrical signals through external electrodes and has the advantages of simplicity, non-invasiveness, and no radiation, and is suitable for wide application in health screening and early tumor detection.

[0120] As an embodiment of the present application, the step S2 specifically includes:

[0121] S21: Perform frequency-domain analysis on the impedance data at different frequencies. For the data of each channel, convert the time-domain signal into a frequency-domain signal through a frequency-domain transformation method to obtain the amplitude and phase information of the impedance data at each frequency point;

[0122] S22: Calculate the amplitude characteristics in the impedance data, including the modulus of the impedance and the phase angle , and their calculation formulas are as follows:

[0123]

[0124]

[0125] Among them, is the resistance, that is, the real part of the impedance, is the reactance, that is, the imaginary part of the impedance, is the arctangent function, which can calculate the phase angle according to the real part and the imaginary part of the impedance, and its range is between ;

[0126] S23: Extract resistivity , conductivity , reactance and other electrical parameters as features, and their calculation formulas are as follows:

[0127]

[0128]

[0129]

[0130] Among them, is the real part of the impedance, is the cross-sectional area, is the distance between the electrodes, represents the imaginary part of the impedance ;

[0131] S24: Use weighted variance to extract the key information of the frequency-impedance spectrum, and its calculation formula is as follows:

[0132]

[0133]

[0134] Among them, represents the weighted variance, represents the number of frequency sample points, is the th frequency point, is the weighting factor for each frequency point, used to assign different weights to different frequency points, represents the th impedance value corresponding to the frequency point, is the weighted average value of the impedance values of all frequency points;

[0135] S25: Use the adaptive sliding window algorithm to extract local features from the impedance data;

[0136] S26: Perform dimensionality reduction on the extracted local features to obtain the key features in the impedance data at different frequencies for each channel.

[0137] As an embodiment of the present application, the step S21 specifically includes:

[0138] S211: After transforming the time-domain signal through the frequency-domain transformation method, divide the obtained frequency range into n sub-bands, and its calculation formula is as follows:

[0139]

[0140]

[0141] where, represents the complex value of the frequency and time positions of the signal in the time-frequency plane, is the time-domain representation of the signal, and is just a sample point of the signal, representing the value of the signal at the moment , is the window function, represents the time-domain shift of the window function, and its center is located at moment, represents the complex modulation of the frequency component, describing how the frequency components of the signal change at different times , is the imaginary unit, is the th frequency of the sub-band, and are the minimum and maximum probabilities respectively, is the number of sub-bands;

[0142] S212: Process each element of the frequency-domain signal data matrix for each sub-band, and calculate the amplitude and phase feature of the sub-band, and its calculation formula is as follows:

[0143]

[0144]

[0145] Among them, and are the real part and the imaginary part of the signal at the frequency respectively, is the arctangent function, calculating the angle between the point and axis;

[0146] S213: For each element of the frequency-domain signal data matrix, calculate the corresponding element of the response matrix, and its calculation formula is as follows:

[0147]

[0148] Among them, is the response matrix element, representing the response characteristic of the signal at the frequency respectively, is the gain and phase shift of the function to the signal at the frequency respectively, is the complex value representing the original signal at the frequency respectively, is the imaginary unit, is the phase of the original signal at the frequency respectively, is the phase of the transfer function at the frequency respectively, and the sum of the phase parts represents the total phase shift of the signal and the system.

[0149] The present invention utilizes bioelectrical impedance analysis technology to extract the impedance characteristics of tumor tissues at different frequencies by performing frequency-domain analysis on electrical signal data. By processing the electrical signal data in different frequency bands, this method can comprehensively understand the differences in the responses of tumor tissues to each frequency band. This helps to identify which frequency bands have high discrimination and importance between tumor tissues and normal tissues.

[0150] Specifically, the present invention can extract highly distinguishable key features from the impedance data of each channel and different frequencies by designing a feature extraction module; this module extracts electrical property parameters such as conductivity, reactance, and resistance by performing time-domain and frequency-domain analysis on electrical signals; these features reflect the electrical properties of tissues and can effectively distinguish normal tissues from tumor tissues; during the extraction process, the module can select the most representative features according to the response characteristics of different frequencies, removing noise and redundant information, so as to ensure that the extracted features have strong discriminative ability; these features will provide strong data support for subsequent classification tasks.

[0151] As an embodiment of the present application, step S3 specifically includes:

[0152] S31: Input the key features, and use a cross-channel attention mechanism to calculate the attention weights. The calculation formula is as follows:

[0153]

[0154]

[0155] where represents the similarity between channel and channel . and are the signal representations of the -th and -th channels respectively. and are the L2 norms of and respectively. represents the attention weight of the -th channel. After softmax, the sum of the attention weights is 1. represents the number of channels;

[0156] S32: According to the calculated attention weights, adjust the contribution degrees of the features of each channel, perform weighted average on the features of different channels, and dynamically adjust the influence of each channel according to the attention weights;

[0157] S33: Perform weighted average on all the weighted features to form a new fused feature representation.

[0158] Specifically, by designing a multi-level feature fusion module, the present invention can fuse the impedance features at different channels and different frequencies; through a multi-scale feature fusion strategy, the information from different depths, frequencies, and channels is effectively integrated together, further removing redundant and noise information and extracting global features with high discriminative ability; these fused features can more accurately reflect the electrical properties of breast tissues and provide a more accurate feature representation for subsequent classification tasks; by fusing information at different levels, the present invention can more comprehensively capture the differences between tumors and normal tissues, improving the robustness and accuracy of the classifier; this module fuses multi-dimensional features into a more comprehensive and accurate representation by integrating information from different channels and frequency ranges. Even when facing different individuals and different types of tumors, the system can work stably, reducing the influence of differences between samples on the detection results; through multi-level fusion, the present invention can effectively capture the electrical features of tumor regions, reduce noise interference, and enhance the discriminative ability of the model.

[0159] As an embodiment of the present application, step S4 specifically includes:

[0160] S41: Input the fused feature data into the adaptive classification module, and select the convolutional long short-term memory network combined with the adaptive optimization method to train the model;

[0161] S42: Perform hyperparameter tuning on the trained model, and use cross-validation to evaluate the performance of the model under different parameter configurations.

[0162] As an embodiment of the present application, step S41 specifically includes:

[0163] S411: Input the fused features into the convolutional layer to extract high-level abstract features. The convolutional layer includes a 1×1 one-dimensional convolution, a ReLU activation function, a 3×3 one-dimensional convolution, a ReLU activation function, and a pooling layer;

[0164] S412: The features processed by the convolutional layer are passed into the convolutional long short-term memory network module to further extract the temporal features in the sequence data;

[0165] S413: Flatten the features after passing through the convolutional long short-term memory network module and then summarize them to obtain temporal information, and finally pass them into the fully connected layer;

[0166] S414: To ensure the robustness of the model, use the He method for model initialization to avoid the problems of gradient disappearance or gradient explosion;

[0167] S415: Use an adaptive optimization algorithm to dynamically adjust the model parameters and optimize the classification accuracy. Its calculation formula is as follows:

[0168]

[0169]

[0170] Where, is the exponential decay average of the square of the gradient, is the current gradient, is the current parameter, is the updated parameter, is the learning rate, is the decay factor, set to 0.9, is a constant used to prevent division by zero;

[0171] S416: Use a smaller learning rate at the beginning of training, and gradually increase the learning rate to avoid slow convergence. As the number of training rounds increases, gradually decrease the learning rate to avoid premature convergence or oscillation near the local minimum;

[0172] S417: Introduce the ensemble learning strategy of gradient boosting trees. Using the method of weighted average, generate preliminary prediction results by gradually training weak learners (decision trees) and combining their outputs.

[0173] S415: In the real-time adaptive adjustment and online learning stage, based on the newly input data and feedback, continue to update the model parameters to improve the classification accuracy. The calculation formula is as follows:

[0174]

[0175] where, are the updated model parameters, are the model parameters at the current moment , is the learning rate, which controls the step size of model parameter update and determines the amplitude of each adjustment. is the loss function, is the currently received sample and label, represents the gradient of the loss function with respect to the model parameters , indicating the rate of change of the loss value with respect to the model parameters.

[0176] The adaptive classification module will automatically adjust hyperparameters such as the learning rate and weight decay according to the characteristics of different tumor samples, and continuously optimize the performance of the model through training and validation data; through the ensemble learning method of gradient boosting trees, the model can improve the classification accuracy and stability from multiple perspectives; each type of tumor (benign, malignant, or different types of tumors) will have its unique electrical characteristics, and the classification module can effectively distinguish different types of tissues based on these characteristics; through this method, not only can normal tissues and tumor tissues be effectively distinguished, but also they can be further refined into categories such as benign tumors and malignant tumors, providing more accurate tumor detection and diagnosis information and providing a reliable basis for subsequent treatment decisions.

[0177] The present invention uses bioelectrical impedance analysis to obtain the electrical signal data of the tumor area, and through the feature extraction module, deeply analyzes the data, and can extract effective impedance features from different depths and frequency ranges; by designing a multi-level feature fusion module, the present invention fuses the signal features from different channels and frequencies, effectively improving the data expression ability and discrimination, thereby further improving the accuracy and reliability of tumor detection; finally, combined with the adaptive classification module and optimization method, the model is adjusted and optimized to ensure good adaptability and high classification performance of the model for different tumor types. The present invention has the advantages of high-precision signal acquisition, accurate feature extraction, strong information fusion ability, and adaptive classification, and can provide effective support for early tumor detection and diagnosis, and has broad clinical application prospects.

[0178] The above description is only some preferred embodiments of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) disclosed in the embodiments of the present disclosure that have similar functions.

Claims

1. A method for detecting anti-tumor by bioelectrical impedance based on pattern recognition, characterized in that, The method includes the following steps: S1: Use a bioelectrical impedance analysis instrument to measure the impedance of the tumor area to be detected, and obtain the current, voltage, and impedance values at different depths and frequencies; S2: Design a feature extraction module to extract the key features from the impedance data at different frequencies of each channel; S3: Input the extracted key features into a multi-level feature fusion module to fuse the impedance features at different channels and frequencies to form new fused features; S4: Input the fused features into an adaptive classification module, and combine an adaptive optimization method to adjust the model parameters to improve the classification accuracy and sensitivity; The specific steps of S2 include: S21: Perform frequency-domain analysis on the impedance data at different frequencies. For the data of each channel, convert the time-domain signal into a frequency-domain signal through a frequency-domain transformation method to obtain the amplitude and phase information of the impedance data at each frequency point; S22: Calculate the amplitude characteristics in the impedance data, including the modulus of the impedance and the phase angle , and their calculation formulas are as follows: Among them, is the resistance, that is, the real part of the impedance, is the reactance, that is, the imaginary part of the impedance, is the arctangent function, which can calculate the phase angle according to the real part and the imaginary part of the impedance, and its range is between S23: Extract the resistivity according to the relationship between the impedance magnitude and the phase angle , conductivity , reactance as electrical parameters as features, and their calculation formulas are as follows: Among them, is the real part of the impedance, is the cross-sectional area, is the distance between the electrodes, represents the imaginary part of the impedance ; S24: Use weighted variance to extract the key information of the frequency-impedance spectrum. The calculation formula is as follows: Among them, represents the weighted variance, represents the number of frequency sample points, is the th frequency point, is the weighting factor for each frequency point, used to assign different weights to different frequency points, represents the th impedance value corresponding to the frequency point, is the weighted mean of the impedance values of all frequency points; S25: Use an adaptive sliding window algorithm to extract local features from the impedance data; S26: Perform dimensionality reduction processing on the extracted local features to obtain the key features in the impedance data at different frequencies of each channel.

2. The method for detecting anti-tumor by bioelectrical impedance based on pattern recognition according to claim 1, wherein, The specific steps of S1 include: S11: Configure a pair of double electrodes, apply current respectively and measure the voltage; S12: Introduce the current applied by the pair of double electrodes into the tumor area to be detected respectively, and stimulate the bio-tissue impedance response of this area through current signals in different frequency ranges; S13: Use a voltage electrode to measure the voltage signal of the tumor area, and through processing such as signal amplification and filtering, obtain the voltage values at different frequencies; S14: Each channel calculates the ratio of the voltage to the applied current respectively to obtain the impedance values at different frequencies, and generates a frequency-impedance spectrum.

3. The method for detecting anti-tumor by bioelectrical impedance based on pattern recognition according to claim 2, wherein, The specific steps of S14 include: S141: For each frequency point, use the known Ohm's law to calculate the impedance value. The calculation formula is as follows; Among them, is the resistance component, is the reactance of the reactive component, is the imaginary unit; S142: Through time-domain and frequency-domain analysis of the signal, extract its amplitude, phase, and impedance characteristic parameters; S143: Integrate the impedance values at different frequencies and draw a spectrum diagram of the impedance changing with frequency; S144: Use adaptive filtering technology for the frequency-impedance spectrum, dynamically adjust the filter parameters according to the statistical characteristics of the noise, and suppress different types of noise and interference. The calculation formula is as follows: wherein, represents the difference between the true output value and the filter output value at time , is the target value at time , is the filter coefficient, representing the filter parameter at time , is the input signal vector, representing the input data at time , is the gain vector, is the forgetting factor, is the inverse covariance matrix at the previous time, is the updated inverse covariance matrix.

4. A method for detecting anti-tumor by bioelectric resistance based on pattern recognition according to claim 1, characterized in that The specific steps of S21 include: S211: After the time-domain signal is transformed through a frequency-domain transformation method, the obtained frequency range is divided into n sub-frequency bands. The calculation formula is as follows: Among them, represents the frequency of the signal in the time-frequency plane and time of the complex value of the position, is the time-domain representation of the signal, while is just a sample point of the signal, representing the signal at the moment value, is the window function, represents the time-domain shift of the window function, and its center is located at moment, represents the complex modulation of the frequency component, describing how the frequency components of the signal change at different times on, is the imaginary unit, is the frequency of the th sub-band, and are the minimum and maximum probabilities respectively, is the number of sub-bands; S212: Process each element of the frequency-domain signal data matrix for each sub-band, and calculate the amplitude of the sub-band and phase characteristics , and its calculation formula is as follows: Among them, and are respectively the real part and the imaginary part of the signal at the frequency ; is the arctangent function, calculating the angle between the point and axis; S213: For each element of the frequency-domain signal data matrix, calculate the corresponding element of the response matrix. The calculation formula is as follows: Among them, is the response matrix element, representing the response characteristic of the signal at frequency ; is the gain and phase shift of the function with respect to the signal at frequency ; is the complex value representing the original signal at frequency ; is the imaginary unit; is the phase of the original signal at frequency ; is the phase of the transfer function at frequency , and the sum of the phase parts represents the total phase shift of the signal and the system.

5. A method for detecting anti-tumor of bioelectric resistance based on pattern recognition according to claim 1, characterized in that, The specific steps of S3 include: S31: Input the key features, and use a cross-channel attention mechanism to calculate the attention weights. The calculation formula is as follows: Among them, represents the similarity of channels and channels The similarity of and are respectively the signal representations of the and the th channels, and are respectively and The L2 norms of represents the attention weight of the th channel. After softmax, the sum of the attention weights is 1, represents the number of channels; S32: According to the calculated attention weights, adjust the contribution degree of each channel feature, perform weighted averaging on the features of different channels, and dynamically adjust the influence of each channel according to the attention weights; S33: Perform weighted averaging on all weighted features to form a new fused feature representation.

6. The method for detecting anti-tumor by bioelectrical impedance based on pattern recognition according to claim 5, wherein, The specific steps of S4 include: S41: Input the fused feature into the adaptive classification module, and select a convolutional long short-term memory network combined with an adaptive optimization method to train the model; S42: Tune the hyperparameters of the trained model, and use cross-validation to evaluate the performance of the model under different parameter configurations.

7. A method for detecting anti-tumor by bioelectrical impedance based on pattern recognition according to claim 6, characterized in that, The specific steps of S41 include: S411: Input the fused feature into the convolutional layer to extract high-level abstract features. The convolutional layer includes a 1×1 one-dimensional convolution, a ReLU activation function, a 3×3 one-dimensional convolution, a ReLU activation function, and a pooling layer; S412: The features processed by the convolutional layer are fed into the convolutional long short-term memory network module to further extract the temporal features in the sequence data; S413: Flatten the features after passing through the convolutional long short-term memory network module and then summarize to obtain temporal information, and finally feed it into the fully connected layer; S414: To ensure the robustness of the model, use the He method for model initialization to avoid the problems of gradient disappearance or gradient explosion; S415: Use an adaptive optimization algorithm to dynamically adjust the model parameters and optimize the classification accuracy. The calculation formula is as follows: wherein, is the exponentially weighted moving average of the square of the gradient, is the current gradient, is the current parameter, is the updated parameter, is the learning rate, is the decay factor, set to 0.9, is a constant used to prevent division by zero; S416: Use a smaller learning rate at the beginning of training, and gradually increase the learning rate to avoid slow convergence. As the number of training rounds increases, gradually decrease the learning rate to avoid premature convergence or oscillation near the local minimum; S417: Introduce the ensemble learning strategy of gradient boosting trees, use the method of weighted averaging, and generate a preliminary prediction result by gradually training weak learners and combining their outputs; S418: In the real-time adaptive adjustment and online learning stage, continue to update the model parameters based on the newly input data and feedback to improve the classification accuracy. The calculation formula is as follows: Among them, is the updated model parameter, is the model parameter at the current moment ; is the learning rate, which controls the step size of the model parameter update and determines the amplitude of each adjustment. is the loss function, is the currently received sample and label, denotes the loss function with respect to the model parameter gradient, representing the rate of change of the loss value with respect to the model parameter. It should be noted that there seems to be a semicolon missing in the original Chinese text at the end of line 6. The above translation has been adjusted accordingly for better logical coherence.

Images