Disclosure of Invention
The invention aims to provide a bioelectrical impedance tumor detection method based on pattern recognition, which is characterized in that data are automatically processed and classified through a pattern recognition algorithm, and meanwhile, the accuracy, the sensitivity and the reliability of tumor detection can be effectively improved by combining a feature extraction module, a multi-level feature fusion module and an adaptive classification module.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
The first aspect of the invention provides a bioelectrical impedance tumor detection method based on pattern recognition, which comprises the following steps:
S1, carrying out electrical impedance measurement on a tumor area to be detected through a bioelectrical impedance analysis instrument to obtain current, voltage and impedance values at different depths and different frequencies;
S2, designing a feature extraction module to extract key features in the electrical impedance data of each channel under different frequencies;
s3, inputting the extracted key features into a multi-level feature fusion module, and fusing the electrical impedance features of different channels and different frequencies to form new fusion features;
and S4, inputting the fusion characteristics to an adaptive classification module, and adjusting model parameters by combining an adaptive optimization method, so as to improve classification accuracy and sensitivity.
As an embodiment of the present application, the step S1 specifically includes:
s11, configuring a double electrode pair, respectively applying current and measuring voltage;
S12, respectively introducing currents applied by the double electrode pairs into a tumor area to be detected, and exciting the impedance response of biological tissues of the area through current signals in different frequency ranges;
S13, measuring voltage signals of a tumor area by using a voltage electrode, and obtaining voltage values under different frequencies through signal amplification, filtering and other treatments;
And S14, each channel obtains impedance values under different frequencies by calculating the ratio of the voltage to the applied current, and generates a frequency-impedance spectrum.
As an embodiment of the present application, the step S14 specifically includes:
s141, calculating an impedance value by using a known ohm law for each frequency point, wherein a calculation formula is as follows;
wherein,As a component of the electrical resistance,Is the reactance of the reactive component,Is an imaginary unit;
S142, extracting amplitude, phase and impedance characteristic parameters of the signal by carrying out time domain and frequency domain analysis on the signal;
s143, integrating impedance values at different frequencies, and drawing a spectrogram of impedance changing along with the frequency;
S144, using an adaptive filtering technology for the frequency-impedance spectrum, dynamically adjusting filter parameters according to the statistical characteristics of noise, and inhibiting different types of noise and interference, wherein the calculation formula is as follows:
wherein,Is shown at the momentThe difference between the true output value of (c) and the filter output value,Is at the momentIs set to be a target value of (c),For the filter coefficients, expressed at time instantIs used for the filtering of the filter parameters,For input signal vectors, expressed at time instantsIs used for the input data of the (a),As a vector of the gain,As a forgetting factor,For the inverse covariance matrix of the previous instant,Is the updated inverse covariance matrix.
As an embodiment of the present application, the step S2 specifically includes:
s21, carrying out frequency domain analysis on the electrical impedance data under different frequencies, and converting a time domain signal into a frequency domain signal by a frequency domain transformation method on the data of each channel to obtain amplitude and phase information of the electrical impedance data on each frequency point;
s22, calculating amplitude characteristics in the electrical impedance data, including the modulus of the electrical impedanceAnd phase angleThe calculation formula is as follows:
wherein,For the resistance, i.e. the real part of the electrical impedance,The reactance, i.e. the imaginary part of the electrical impedance,Is an arctangent function, can be based on the real part of the electrical impedanceAnd imaginary partTo calculate the phase angleIn the range ofBetween them;
S23, extracting the resistivity according to the relation between the electrical impedance amplitude and the phase angleConductivity ofReactance ofThe electrical parameters are characterized by the following calculation formula:
wherein,Is the real part of the electrical impedance,Is the cross-sectional area,Is the distance between the electrodes and,Representing the imaginary part of an electrical impedance;
S24, extracting key information of the frequency-impedance spectrum by using the weighted variance, wherein the calculation formula is as follows:
wherein,The weighted variance is represented as such,Representing the number of frequency sample points,Is the firstThe frequency points of the frequency spectrum are selected,Is a weighting factor for each frequency bin, for giving different weights to different frequency bins,Represent the firstThe impedance value corresponding to the frequency point,Is the weighted average of the impedance values of all the frequency points;
s25, carrying out local feature extraction on the electrical impedance data by using an adaptive sliding window algorithm;
S26, performing dimension reduction processing on each extracted local feature to obtain key features in the electrical impedance data of each channel under different frequencies.
As an embodiment of the present application, the step S21 specifically includes:
S211, after the time domain signal is transformed by a frequency domain transformation method, the obtained frequency range is divided into n frequency sub-bands, and the calculation formula is as follows:
wherein,Representing the frequency of a signal in the time-frequency planeAnd time ofThe complex value of the position is used to determine,Is a time domain representation of the signalOnly one sample point of the signal, representing the signal at the momentIs used as a reference to the value of (a),As a function of the window(s),Representing the time-domain movement of the window function with its center atAt the moment of time of day,Representing complex modulation of frequency components, describes how the frequency components of the signal are at different timesThe change is made up to the above-mentioned,Is an imaginary unit of number and is,Is the firstThe frequencies of the sub-bands,AndThe minimum and maximum probabilities respectively are given,Is the number of sub-bands;
s212, processing each element of the frequency domain signal data matrix of each sub-band, and calculating the amplitude of the sub-bandAnd phase characteristicsThe calculation formula is as follows:
wherein,AndRespectively are signalsAt the frequency ofThe real and imaginary parts of the point(s),Is an arctangent function, calculated as a pointAnd (3) withAn included angle of the shaft;
s213, for each element of the frequency domain signal data matrix, calculating the corresponding element of the response matrix, wherein the calculation formula is as follows:
wherein,In response to the matrix element, the signal is represented at frequencyThe response characteristics of the device,Is a function of frequencyThe gain and phase offset of the signal are compared,Is indicative of the original signal at frequencyThe complex number of which is defined by the number,Is an imaginary unit of number and is,Is the original signal at frequencyThe phase at which the phase is to be shifted,Is the transfer function at frequencySum of phase, phase partRepresenting the total phase offset of the signal and the system.
As an embodiment of the present application, the step S3 specifically includes:
S31, inputting the key features, and calculating the attention weight by using a cross-channel attention mechanism, wherein the calculation formula is as follows:
wherein,Representation channelAnd a channelIs used for the degree of similarity of (c) to (c),AndRespectively the firstAnd (d)The signal representation of the individual channels is used,AndRespectively areAndIs set to be a normal number of L2 of (c),Represent the firstThe attention weights of the channels are summed to 1 after soft maximization,Representing the number of channels;
S32, according to the calculated attention weight, the contribution degree of each channel characteristic is adjusted, the characteristics of different channels are weighted and averaged, and according to the attention weight, the influence of each channel is dynamically adjusted;
And S33, carrying out weighted average on all the weighted features to form a new fusion feature representation.
As an embodiment of the present application, the step S4 specifically includes:
S41, inputting the fused characteristic data into an adaptive classification module, and selecting a convolution long-short-time memory network to train a model by combining an adaptive optimization method;
and S42, performing super-parameter tuning on the model obtained through training, and evaluating the performance of the model under different parameter configurations by adopting cross verification.
As an embodiment of the present application, the step S41 specifically includes:
S411, inputting the fusion features into a convolution layer to extract high-level abstract features, wherein the convolution layer comprises a1 multiplied by 1 one-dimensional convolution, a ReLU activation function, a3 multiplied by 3 one-dimensional convolution, a ReLU activation function and a pooling layer;
s412, the characteristics processed by the convolution layer are transmitted into a convolution long-short time memory network module, and the time sequence characteristics in the sequence data are further extracted;
s413, flattening the features after the convolution long-short-term memory network module, summarizing to obtain time sequence information, and finally transmitting the time sequence information into a full-connection layer;
s414, in order to ensure the robustness of the model, initializing the model by adopting a He method, so as to avoid the problem of gradient disappearance or gradient explosion;
S415, adopting an adaptive optimization algorithm to dynamically adjust model parameters and optimize classification accuracy, wherein the calculation formula is as follows:
wherein,Is the exponential decay average of the square of the gradient,Is the current gradient of the gradient,Is the current parameter of the current value,Is a parameter that has been updated and is then used,Is the rate of learning to be performed,For the attenuation factor, set to 0.9,Is a constant to prevent division by zero;
S416, using a smaller learning rate in the initial training stage, avoiding too slow convergence by gradually increasing the learning rate, and gradually reducing the learning rate along with the increase of training rounds so as to avoid premature convergence or oscillation near a local minimum;
S417, introducing an integrated learning strategy of a gradient lifting tree, and generating a preliminary prediction result by gradually training weak learners and combining the outputs of the weak learners by using a weighted average method;
S418, in the stage of real-time self-adaptive adjustment and online learning, based on the newly input data and feedback, continuously updating the model parameters to improve the classification accuracy, wherein the calculation formula is as follows:
wherein,Is the model parameters after the update of the model parameters,Is the current timeIs used for the model parameters of the model (a),Is learning rate, controls the step length of model parameter updating, decides the amplitude of each adjustment,Is the loss function of the device,For the currently received samples and tags,Representing a loss functionFor model parametersAnd the gradient of (c) represents the rate of change of the loss value with respect to the model parameters.
The beneficial effects of the invention are as follows:
(1) The invention utilizes bioelectrical impedance analysis technology in combination with pattern recognition method, can accurately extract the electrical signal characteristics of tumor tissues under different frequencies, effectively distinguish normal tissues from tumor tissues, further distinguish benign and malignant tumors, and can assist in distinguishing specific types of tumors under certain conditions. By fusing the feature extraction and the intelligent pattern recognition method, the invention greatly improves the accuracy of tumor detection and reduces the occurrence rate of misdiagnosis and missed diagnosis.
(2) According to the invention, by designing the feature extraction module, key features with high identification degree can be extracted from the electrical impedance data of each channel and different frequencies, and electrical characteristic parameters such as conductivity, reactance, resistance and the like are extracted by carrying out time domain and frequency domain analysis on the electrical signals, the features reflect the electrical properties of tissues and can effectively distinguish normal tissues from tumor tissues, and in the extraction process, the module can select the most representative features according to the response characteristics of different frequencies and remove noise and redundant information, so that the extracted features have stronger discrimination capability.
(3) The invention can integrate information from different depths, frequencies and channels effectively by designing a multi-level feature fusion module, integrate the information into a more comprehensive and accurate representation, further remove redundant and noise information, extract global features with high discrimination capability, enable the fused features to reflect the electrical characteristics of breast tissues more accurately, provide more accurate feature representation for subsequent classification tasks, capture the difference between tumor and normal tissues more comprehensively by fusing information of different levels, improve the robustness and accuracy of a classifier, and effectively capture the electrical features of tumor regions, reduce noise interference and enhance the discrimination capability of models by multi-level fusion.
(4) According to the invention, by designing the self-adaptive classification module, super parameters such as learning rate, weight attenuation and the like can be automatically adjusted according to the characteristics of different tumor samples, the performance of the model is continuously optimized through training and verifying data, the accuracy and stability of classification can be improved from multiple angles through the integrated learning method of the gradient lifting tree, each type of tumor (benign, malignant or different types of tumors) has unique electrical characteristics, and the classification module can effectively distinguish each type of tissue according to the characteristics.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that all directional indicators (such as up, down, left, right, front, and rear are used in the embodiments of the present invention) are merely for explaining the relative positional relationship, movement conditions, and the like between the components in a certain specific posture (as shown in the drawings), and if the specific posture is changed, the directional indicators are changed accordingly.
In the present invention, unless explicitly specified and limited otherwise, the terms "connected," "fixed," and the like are to be construed broadly, and for example, "fixed" may be fixedly connected, detachably connected, or integrally formed, mechanically connected, electrically connected, directly connected, indirectly connected via an intervening medium, or in communication between two elements or in an interaction relationship between two elements, unless otherwise explicitly specified. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
In addition, if there is a description of "first", "second", etc. in the embodiments of the present invention, the description of "first", "second", etc. is for descriptive purposes only and is not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In addition, the meaning of "and/or" as it appears throughout includes three parallel schemes, for example "A and/or B", including the A scheme, or the B scheme, or the scheme where A and B are satisfied simultaneously. In addition, the technical solutions of the embodiments may be combined with each other, but it is necessary to base that the technical solutions can be realized by those skilled in the art, and when the technical solutions are contradictory or cannot be realized, the combination of the technical solutions should be considered to be absent and not within the scope of protection claimed in the present invention.
Referring to fig. 1 to 4, a first aspect of the present invention provides a bioelectrical impedance tumor detection method based on pattern recognition, the method comprising the steps of:
S1, carrying out electrical impedance measurement on a tumor area to be detected through a bioelectrical impedance analysis instrument to obtain current, voltage and impedance values at different depths and different frequencies;
S2, designing a feature extraction module to extract key features in the electrical impedance data of each channel under different frequencies;
s3, inputting the extracted key features into a multi-level feature fusion module, and fusing the electrical impedance features of different channels and different frequencies to form new fusion features;
and S4, inputting the fusion characteristics to an adaptive classification module, and adjusting model parameters by combining an adaptive optimization method, so as to improve classification accuracy and sensitivity.
The invention utilizes bioelectrical impedance analysis technology in combination with pattern recognition method, can accurately extract the electrical signal characteristics of tumor tissues under different frequencies, effectively distinguish normal tissues from tumor tissues, further distinguish benign and malignant tumors, and can assist in distinguishing specific types of tumors under certain conditions, by fusing the feature extraction and the intelligent pattern recognition method, the invention greatly improves the accuracy of tumor detection, reduces the occurrence rate of misdiagnosis and missed diagnosis, and the pattern recognition method adopted by the invention has better interpretability, can provide a transparent decision process for doctors and assists the clinicians to make more accurate diagnosis decisions.
As an embodiment of the present application, the step S1 specifically includes:
s11, configuring a double electrode pair, respectively applying current and measuring voltage;
S12, respectively introducing currents applied by the double electrode pairs into a tumor area to be detected, and exciting the impedance response of biological tissues of the area through current signals in different frequency ranges;
S13, measuring voltage signals of a tumor area by using a voltage electrode, and obtaining voltage values under different frequencies through signal amplification, filtering and other treatments;
And S14, each channel obtains impedance values under different frequencies by calculating the ratio of the voltage to the applied current, and generates a frequency-impedance spectrum.
As an embodiment of the present application, the step S14 specifically includes:
s141, calculating an impedance value by using a known ohm law for each frequency point, wherein a calculation formula is as follows;
wherein,As a component of the electrical resistance,Is the reactance of the reactive component,Is an imaginary unit;
S142, extracting amplitude, phase and impedance characteristic parameters of the signal by carrying out time domain and frequency domain analysis on the signal;
s143, integrating impedance values at different frequencies, and drawing a spectrogram of impedance changing along with the frequency;
S144, using an adaptive filtering technology for the frequency-impedance spectrum, dynamically adjusting filter parameters according to the statistical characteristics of noise, and inhibiting different types of noise and interference, wherein the calculation formula is as follows:
wherein,Is shown at the momentThe difference between the true output value of (c) and the filter output value,Is at the momentIs set to be a target value of (c),For the filter coefficients, expressed at time instantIs used for the filtering of the filter parameters,For input signal vectors, expressed at time instantsIs used for the input data of the (a),As a vector of the gain,As a forgetting factor,For the inverse covariance matrix of the previous instant,Is the updated inverse covariance matrix.
Specifically, the invention obtains the electrical signal data of the detected tumor area by utilizing the multi-frequency bioelectrical impedance analyzer, and can extract important electrical parameters such as conductivity, resistance, capacitance and the like of tumor tissues by measuring and analyzing the current, voltage and impedance values at different depths and different frequencies. These electrical parameters vary significantly between tumor tissue and normal tissue, as abnormal proliferation of tumor cells and changes in tissue architecture typically result in changes in electrical characteristics. Therefore, compared with the traditional tumor diagnosis method (such as puncture, CT or MRI examination), the method does not need invasive operation, reduces discomfort and risk of a patient, has the advantages of simplicity, convenience, no wound, no radiation and the like by collecting the electric signals through the external electrode, and is suitable for being widely applied to health screening and early tumor detection.
As an embodiment of the present application, the step S2 specifically includes:
s21, carrying out frequency domain analysis on the electrical impedance data under different frequencies, and converting a time domain signal into a frequency domain signal by a frequency domain transformation method on the data of each channel to obtain amplitude and phase information of the electrical impedance data on each frequency point;
s22, calculating amplitude characteristics in the electrical impedance data, including the modulus of the electrical impedanceAnd phase angleThe calculation formula is as follows:
wherein,For the resistance, i.e. the real part of the electrical impedance,For reactance, i.e. the imaginary part of the electrical impedance,Is an arctangent function, can be based on the real part of the electrical impedanceAnd imaginary partTo calculate the phase angleIn the range ofBetween them;
S23, extracting the resistivity according to the relation between the electrical impedance amplitude and the phase angleConductivity ofReactance ofThe electrical parameters are characterized by the following calculation formula:
wherein,Is the real part of the electrical impedance,Is the cross-sectional area,Is the distance between the electrodes and,Representing the imaginary part of an electrical impedance;
S24, extracting key information of the frequency-impedance spectrum by using the weighted variance, wherein the calculation formula is as follows:
wherein,The weighted variance is represented as such,Representing the number of frequency sample points,Is the firstThe frequency points of the frequency spectrum are selected,Is a weighting factor for each frequency bin, for giving different weights to different frequency bins,Represent the firstThe impedance value corresponding to the frequency point,Is the weighted average of the impedance values of all the frequency points;
s25, carrying out local feature extraction on the electrical impedance data by using an adaptive sliding window algorithm;
S26, performing dimension reduction processing on each extracted local feature to obtain key features in the electrical impedance data of each channel under different frequencies.
As an embodiment of the present application, the step S21 specifically includes:
S211, after the time domain signal is transformed by a frequency domain transformation method, the obtained frequency range is divided into n frequency sub-bands, and the calculation formula is as follows:
wherein,Representing the frequency of a signal in the time-frequency planeAnd time ofThe complex value of the position is used to determine,Is a time domain representation of the signalOnly one sample point of the signal, representing the signal at the momentIs used as a reference to the value of (a),As a function of the window(s),Representing the time-domain movement of the window function with its center atAt the moment of time of day,Representing complex modulation of frequency components, describes how the frequency components of the signal are at different timesThe change is made up to the above-mentioned,Is an imaginary unit of number and is,Is the firstThe frequencies of the sub-bands,AndThe minimum and maximum probabilities respectively are given,Is the number of sub-bands;
s212, processing each element of the frequency domain signal data matrix of each sub-band, and calculating the amplitude of the sub-bandAnd phase characteristicsThe calculation formula is as follows:
wherein,AndRespectively are signalsAt the frequency ofThe real and imaginary parts of the point(s),Is an arctangent function, calculated as a pointAnd (3) withAn included angle of the shaft;
s213, for each element of the frequency domain signal data matrix, calculating the corresponding element of the response matrix, wherein the calculation formula is as follows:
wherein,In response to the matrix element, the signal is represented at frequencyThe response characteristics of the device,Is a function of frequencyThe gain and phase offset of the signal are compared,Is indicative of the original signal at frequencyThe complex number of which is defined by the number,Is an imaginary unit of number and is,Is the original signal at frequencyThe phase at which the phase is to be shifted,Is the transfer function at frequencySum of phase, phase partRepresenting the total phase offset of the signal and the system.
The invention utilizes bioelectrical impedance analysis technology to extract the electrical impedance characteristics of tumor tissues under different frequencies by carrying out frequency domain analysis on the electrical signal data. The method can comprehensively understand the difference of the response of each frequency band to the tumor tissue by processing the electric signal data of different frequency bands. This helps to identify which frequency bands have a higher degree of differentiation and importance between tumor tissue and normal tissue.
The method comprises the steps of designing a characteristic extraction module, extracting key characteristics with high identification degree from electrical impedance data of each channel and different frequencies, extracting electrical characteristic parameters such as conductivity, reactance, resistance and the like by carrying out time domain and frequency domain analysis on an electrical signal, reflecting the electrical properties of tissues and effectively distinguishing normal tissues from tumor tissues, selecting the most representative characteristics according to response characteristics of different frequencies in the extraction process, removing noise and redundant information, and accordingly ensuring that the extracted characteristics have strong discrimination capability, and providing powerful data support for subsequent classification tasks.
As an embodiment of the present application, the step S3 specifically includes:
S31, inputting the key features, and calculating the attention weight by using a cross-channel attention mechanism, wherein the calculation formula is as follows:
wherein,Representation channelAnd a channelIs used for the degree of similarity of (c) to (c),AndRespectively the firstAnd (d)The signal representation of the individual channels is used,AndRespectively areAndIs set to be a normal number of L2 of (c),Represent the firstThe attention weights of the channels are summed to 1 after soft maximization,Representing the number of channels;
S32, according to the calculated attention weight, the contribution degree of each channel characteristic is adjusted, the characteristics of different channels are weighted and averaged, and according to the attention weight, the influence of each channel is dynamically adjusted;
And S33, carrying out weighted average on all the weighted features to form a new fusion feature representation.
The invention can more accurately reflect the electrical characteristics of breast tissues and provide more accurate characteristic representation for subsequent classification tasks, can more comprehensively capture the difference between tumors and normal tissues and improve the robustness and accuracy of a classifier by integrating the information from different channels and frequency ranges, can integrate the multidimensional characteristics into more comprehensive and accurate representation by integrating the information from different channels and frequency ranges, can stably work even facing different individuals and different types of tumors, reduces the influence of the difference between samples on detection results, can effectively capture the electrical characteristics of tumor areas, reduces noise interference and enhances the discrimination capability of a model by multi-level fusion.
As an embodiment of the present application, the step S4 specifically includes:
S41, inputting the fused characteristic data into an adaptive classification module, and selecting a convolution long-short-time memory network to train a model by combining an adaptive optimization method;
and S42, performing super-parameter tuning on the model obtained through training, and evaluating the performance of the model under different parameter configurations by adopting cross verification.
As an embodiment of the present application, the step S41 specifically includes:
S411, inputting the fusion features into a convolution layer to extract high-level abstract features, wherein the convolution layer comprises a1 multiplied by 1 one-dimensional convolution, a ReLU activation function, a3 multiplied by 3 one-dimensional convolution, a ReLU activation function and a pooling layer;
s412, the characteristics processed by the convolution layer are transmitted into a convolution long-short time memory network module, and the time sequence characteristics in the sequence data are further extracted;
s413, flattening the features after the convolution long-short-term memory network module, summarizing to obtain time sequence information, and finally transmitting the time sequence information into a full-connection layer;
s414, in order to ensure the robustness of the model, initializing the model by adopting a He method, so as to avoid the problem of gradient disappearance or gradient explosion;
S415, adopting an adaptive optimization algorithm to dynamically adjust model parameters and optimize classification accuracy, wherein the calculation formula is as follows:
wherein,Is the exponential decay average of the square of the gradient,Is the current gradient of the gradient,Is the current parameter of the current value,Is a parameter that has been updated and is then used,Is the rate of learning to be performed,For the attenuation factor, set to 0.9,Is a constant to prevent division by zero;
S416, using a smaller learning rate in the initial training stage, avoiding too slow convergence by gradually increasing the learning rate, and gradually reducing the learning rate along with the increase of training rounds so as to avoid premature convergence or oscillation near a local minimum;
S417, introducing an integrated learning strategy of a gradient lifting tree, and generating a preliminary prediction result by gradually training weak learners (decision trees) and combining the outputs of the weak learners (decision trees) by using a weighted average method;
S415, in the stage of real-time self-adaptive adjustment and online learning, based on the newly input data and feedback, continuously updating the model parameters to improve the classification accuracy, wherein the calculation formula is as follows:
wherein,Is the model parameters after the update of the model parameters,Is the current timeIs used for the model parameters of the model (a),Is learning rate, controls the step length of model parameter updating, decides the amplitude of each adjustment,Is the loss function of the device,For the currently received samples and tags,Representing a loss functionFor model parametersAnd the gradient of (c) represents the rate of change of the loss value with respect to the model parameters.
The self-adaptive classification module can automatically adjust super parameters such as learning rate, weight attenuation and the like according to the characteristics of different tumor samples, continuously optimize the performance of the model through training and verifying data, improve the classification precision and stability of the model from multiple angles through an integrated learning method of gradient lifting trees, effectively distinguish various types of tissues according to the characteristics by the classification module, effectively distinguish normal tissues and tumor tissues, further refine the normal tissues and the tumor tissues into benign tumors, malignant tumors and the like, provide more accurate tumor detection and diagnosis information and provide reliable basis for subsequent treatment decisions.
The invention utilizes bioelectrical impedance analysis to acquire the electrical signal data of a tumor area, carries out deep analysis on the data through a characteristic extraction module, can extract effective electrical impedance characteristics from different depth and frequency ranges, fuses the signal characteristics from different channels and frequencies through a multi-level characteristic fusion module, effectively improves the expression capability and discrimination of the data, further improves the accuracy and reliability of tumor detection, and finally adjusts and optimizes the model by combining an adaptive classification module and an optimization method, thereby ensuring good adaptability and high classification performance of the model on different tumor types. The invention has the advantages of high-precision signal acquisition, accurate feature extraction, strong information fusion capability and self-adaptive classification, can provide effective support for early detection and diagnosis of tumors, and has wide clinical application prospect.
The foregoing description is only of the preferred embodiments of the present disclosure and description of the principles of the technology being employed. It will be appreciated by those skilled in the art that the scope of the invention in the embodiments of the present disclosure is not limited to the specific combination of the above technical features, but encompasses other technical features formed by any combination of the above technical features or their equivalents without departing from the spirit of the invention. Such as the above-described features, are mutually substituted with (but not limited to) the features having similar functions disclosed in the embodiments of the present disclosure.