Electrical impedance cancer detection method based on frequency domain information enhancement

Abstract

The invention discloses an electrical impedance cancer detection method based on frequency domain information enhancement. The method comprises the following steps: S1, acquiring an electrical impedance signal in the body of a patient through an electrical impedance imaging technology; s2, converting the signal into frequency domain representation by using discrete cosine transform, and performing frequency domain enhancement; s3, inputting the features subjected to frequency domain enhancement into the model, and processing the features subjected to frequency domain enhancement in combination with frequency domain information and a global dependency relationship to obtain a global feature matrix; and S4, inputting the global feature matrix into a classifier for cancer detection, and giving a corresponding detection result. According to the method, the time domain signals are subjected to frequency domain conversion, the features subjected to frequency domain enhancement are input into the network model, the global matrix features are obtained by combining the frequency domain information and the global dependency relationship, the global matrix features are input into the classifier for cancer detection, the robustness of the frequency domain data is improved, and the application effect of the electrical impedance signals in cancer detection is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of cancer detection, and particularly to an impedance cancer detection method based on enhanced frequency domain information. Background Art

[0002] Cancer is one of the main causes of death globally, and its early detection and treatment are crucial for improving the survival rate of patients. Traditional cancer detection methods have certain limitations, including high cost, risk of radiation exposure, and insufficient sensitivity for early detection. Therefore, the development of new cancer detection technologies is of great significance. The impedance cancer detection method based on enhanced frequency domain information provides a new solution for cancer detection due to its non-invasive and simple operation characteristics. With the rapid development of deep learning technology, it can provide accurate cancer diagnosis results.

[0003] In the prior art, the Chinese patent with the publication number CN117481630A discloses a "breast cancer detection method based on bioelectrical impedance analysis". This method extracts the electrical signal data of normal and malignant samples of breast tissue through bioelectrical impedance analysis technology to improve breast cancer detection. However, this method mainly relies on the detection of a single cancer and cannot detect multiple cancers through frequency domain information, making the application of impedance signals in cancer detection have certain limitations.

[0004] Therefore, it is urgent to design an impedance cancer detection method based on enhanced frequency domain information to solve the problems existing in the above prior art. Summary of the Invention

[0005] The purpose of the present invention is to provide an impedance cancer detection method based on enhanced frequency domain information. By performing frequency domain conversion on the time domain signal, the features after frequency domain enhancement are input into the network model, and the global matrix features are obtained by combining the frequency domain information and global dependence relationship. Finally, the classifier is input for cancer detection, effectively improving the robustness to frequency domain data, thereby improving the application effect of impedance signals in cancer detection.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: The first aspect of the present invention provides an impedance cancer detection method based on enhanced frequency domain information, and the method includes the following steps: S1: Obtain the impedance signal in the patient's body through impedance imaging technology and perform preprocessing on it; S2: Perform frequency domain conversion on the preprocessed time domain signal, and use discrete cosine transform to convert the signal into a frequency domain representation and perform frequency domain enhancement; S3: Input the features enhanced in the frequency domain into the improved network model. The improved network model combines frequency domain information and global dependencies, and processes the features enhanced in the frequency domain to obtain the final global feature matrix. S4: Input the final global feature matrix into the classifier. After calculation by the fully connected layer to obtain the output, then convert the output into a probability value through a function, which can determine whether cancer exists and give the corresponding detection result.

[0007] As an embodiment of the present application, the step S1 specifically includes: S11: Through electrical impedance tomography technology, inject current into the human body and record the reflected signals at different electrode positions. These signals reflect the conductivity of different tissues in the body, contain amplitude and phase information in the time domain, and these signals are represented in complex form, including the real part resistance and the imaginary part reactance. The calculation formula is as follows:

[0008] Among them, represents the total impedance changing with time; the resistance value changing with time; represents the reactance value changing with time; is the imaginary unit, indicating the difference between the reactance part and the resistance part; is used to control the weight or influence between reactance and resistance; represents the phase, the phase information of the current or voltage changing with time; S12: Perform preliminary denoising processing on the signal, apply a low-pass filter to remove high-frequency noise and retain the low-frequency components of the signal. The formula of the filtered signal is expressed as:

[0009] Among them, represents the denoised signal; represents the original electrical impedance signal; represents the impulse response of the filter; represents the adjustment parameter of the filter; controls the frequency response of the filter; S13: Remove the baseline drift by subtracting the mean value from the signal to ensure that the mean value of the signal is close to zero. The calculation formula is as follows:

[0010] Among them, represents the signal after removing the baseline drift; represents the denoised signal; represents the mean value of the signal multiplied by the intensity or offset used to control the removal of the baseline drift; S14: Standardize the signal to have zero mean and unit variance to ensure that signals are compared on the same scale. The standardization formula is as follows:

[0011] where, represents the standardized signal; represents the signal after removing baseline drift, represents the mean of the standardized signal; represents the standard deviation of the standardized signal; represents the new coefficient used to control standardization.

[0012] As an embodiment of the present application, the specific steps of using the discrete cosine transform to convert the signal into a frequency-domain representation in step S2 include: Perform a discrete cosine transform on the time-domain signal. The discrete cosine transform can effectively extract the frequency components of the signal and convert the time-domain signal into a frequency-domain signal. The specific formula is as follows:

[0013] where, is the transformed frequency-domain signal; is the standardized time-domain signal; is the time-domain signal weighting coefficient; is the frequency coefficient, representing the discrete cosine transform of the signal at different frequencies; is the length of the signal; is the transformed frequency-domain index; is the index of the time-domain signal; is the newly added coefficient; used to adjust the overall scaling ratio of the final output amplitude to control the overall result.

[0014] As an embodiment of the present application, the specific steps of performing frequency-domain enhancement in step S2 include: In the frequency domain, the low-frequency part of the signal contains useful biological information. Enhancing the low-frequency part in the frequency domain can further strengthen the effective features of the signal. The formula for the enhanced frequency-domain signal is as follows:

[0015] where, represents the enhanced frequency-domain signal; represents the overall gain factor multiplied by the amplitude part of the frequency-domain signal at a frequency; is the gain function used to adjust the gain of different frequency components; is the phase part corresponding to the frequency; is the phase offset parameter used to adjust the signal phase and avoid phase distortion caused by gain adjustment.

[0016] As an embodiment of the present application, step S3 specifically includes: S31: Input the signal enhanced in the frequency domain into a network model for processing the frequency-domain signal; S32: Extract frequency-domain features through the network model, and obtain a global feature matrix after completing global dependence modeling.

[0017] As an embodiment of the present application, the extraction of frequency-domain features in step S32 specifically includes: S3211: Use four layers of convolution to extract four levels of frequency-domain features from the signal enhanced in the frequency domain Each layer of convolution can gradually extract more complex frequency-domain features to help the model better understand the pattern of the signal. The specific formula is as follows:

[0018]

[0019]

[0020]

[0021] Wherein, is the output after the first layer of convolution operation, is the first layer of 1x1 convolution kernel, is the signal enhanced in the frequency domain, is the bias term of the first layer of convolution, controlling the offset of the model; is the output after the second layer of convolution operation, is the second layer of 2x2 convolution kernel, is the bias term of the second layer of convolution, controlling the offset of the model; is the output after the third layer of convolution operation, is the third layer of 3x3 convolution kernel, is the bias term of the third layer of convolution, controlling the offset of the model; is the output after the fourth layer of convolution operation, is the fourth layer of 4x4 convolution kernel, is the bias term of the fourth layer of convolution, controlling the offset of the model; S3212: After four layers of convolution, the obtained feature map is the output of the deepest layer. To further improve the expression ability of the model and reduce information redundancy, a pooling-based feature aggregation method is used to integrate these frequency-domain features. The formula is as follows:

[0022] Wherein, is the aggregated feature map; It is the feature map output after four - layer convolution.

[0023] As an embodiment of the present application, after completing the global - dependency modeling in step S32, the obtained global - feature matrix specifically includes: S3221: Input the pooled features into the global - dependency modeling layer. In this layer, the self - attention mechanism is used to calculate the dependency relationship between features. The formula is as follows:

[0024] Where, are the pooled features; represent the weight matrices of query, key, and value, which are learning parameters obtained through training; represent the bias terms of query, key, and value; S3222: Use a self - attention mechanism to calculate the correlation between the query and the key. By taking the dot - product of the query and the key, the attention weight of each feature is obtained. The specific formula is as follows:

[0025] Where, represents the attention weight, which contains the dependency relationship between features. Each feature is weighted according to the correlation with other features; represents the query vector of the input feature at the current time step; is the vector obtained after transposition; represents the dimension of the key vector, which is used for normalization, is an additional bias term; S3223: After calculating the attention scores, perform a weighted sum of multiple attention weights and the value matrix. The formula is as follows:

[0026] Where, is the global - feature matrix after weighted sum; are the calculated attention weights; is the value matrix.

[0027] As an embodiment of the present application, in step S4, input the final global - feature matrix into the classifier, and the output obtained through the calculation of the fully - connected layer specifically includes: Take the global - feature matrix obtained through weighted sum as the input of the classifier. Through a flattening operation, input it into a classifier, and the output is obtained through the calculation of the fully - connected layer. The formula is as follows:

[0028]

[0029] Among them, is a flattening operation, and the features after multi-scale fusion are flattened into a one-dimensional vector; is the output of the fully connected layer; is the weight matrix of the fully connected layer; is the weight matrix of the fully connected layer; is the bias term.

[0030] As an embodiment of the present application, in step S4, the output is converted into a probability value through a function, and it can be judged whether there is cancer, and the corresponding detection results specifically include: The output of the fully connected layer is converted into a probability value through a function for classification, and the probability of each category is calculated. The specific calculation formula is:

[0031] Among them, is the probability value of the th class; the calculation method is based on the adjustment of given features and parameters; and are the modified input features; and are adjustable parameters used to control the influence of features on the final result; represents the number of categories.

[0032] The beneficial effects of the present invention are: (1) The present invention obtains the electrical signal data of cancer tissues through electrical impedance tomography technology, preprocesses and enhances the impedance signals in the frequency domain, including denoising, baseline drift correction, and normalization steps, effectively improving the quality of the signals, and can efficiently extract the features in the impedance signals, thereby excavating potential cancer features. Compared with traditional cancer examination methods, the present invention does not require invasive operations such as puncture or surgery, and realizes non-invasive cancer detection.

[0033] (2) The present invention converts the preprocessed time-domain signal into a frequency-domain representation by using the discrete cosine transform and enhances the frequency-domain features, improving the effect of frequency-domain feature extraction. The enhanced frequency-domain features are input into the network model for feature extraction. The frequency-domain features are gradually extracted through four layers of convolution. After four layers of convolution, a pooling operation is used to aggregate the features, enhancing the model's expressive ability and reducing information redundancy. Then, the pooled features are input into the global dependence modeling layer, and the self-attention mechanism is adopted to calculate the dependence relationship between the features, enhancing the model's attention to key information. Finally, by calculating the similarity between the query and the key and weighted summing with the value matrix, the final global feature matrix is obtained for further cancer detection and classification tasks. Through the above steps, the robustness to frequency-domain data can be effectively improved, thereby enhancing the application effect of impedance signals in cancer detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 FIG. is a schematic diagram of a signal collection method for an impedance cancer detection method based on frequency-domain information enhancement provided in an embodiment of the present invention; Figure 2 FIG. is a schematic flowchart of a signal processing method for an impedance cancer detection method based on frequency-domain information enhancement provided in an embodiment of the present invention; Figure 3 FIG. is a schematic flowchart of a signal feature extraction process for an impedance cancer detection method based on frequency-domain information enhancement provided in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the 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.

[0036] It should be noted that all the 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 drawings). If the specific posture changes, the directional indications will also change accordingly.

[0037] In the present invention, unless otherwise clearly specified and defined, terms such as "connection" and "fixation" shall 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 it can be the communication inside two components or the interaction relationship between two components, unless otherwise clearly defined. 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 circumstances.

[0038] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, such descriptions of "first", "second", etc. are only for descriptive purposes and should not be construed 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 scenarios. Taking "A and / or B" as an example, it includes scenario A, scenario B, or the scenario where both A and B are satisfied simultaneously. In addition, the technical solutions between various embodiments can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0039] Referring to Figures 1 to 3 , the first aspect of the present invention provides a method for impedance cancer detection based on frequency-domain information enhancement, and the method includes the following steps: S1: Obtain the impedance signal in the patient's body through impedance imaging technology and preprocess it; S2: Perform frequency-domain conversion on the preprocessed time-domain signal, and use the discrete cosine transform to convert the signal into a frequency-domain representation and perform frequency-domain enhancement; S3: Input the features after frequency-domain enhancement into an improved network model, and the improved network model combines frequency-domain information and global dependence relationship to process the features after frequency-domain enhancement to obtain a final global feature matrix; S4: Input the final global feature matrix into a classifier, calculate the output through a fully connected layer, and then convert the output into a probability value through a function, which can determine whether cancer exists and give corresponding detection results.

[0040] Specifically, the present invention obtains the electrical impedance signal in the patient's body through electrical impedance tomography technology, preprocesses the signal, and through a discrete cosine transform, converts the time-domain signal into a frequency-domain representation. After enhancing the frequency-domain features, it is input into the improved network model. The improved network model combines the frequency-domain information and global dependence relationship to process the features after frequency-domain enhancement to obtain the final global feature matrix; the final global feature matrix is input into the classifier, and the output is calculated through the fully connected layer, and then the output is converted into a probability value through a function, which can determine whether cancer exists and give the corresponding detection result, providing higher-precision cancer detection support.

[0041] As Figure 2 shown, as an embodiment of the present application, step S1 specifically includes: S11: Through electrical impedance tomography technology, current is injected into the human body and the reflected signals are recorded at different electrode positions. These signals reflect the conductivity of different tissues in the body and contain the amplitude and phase information in the time domain. Usually, these signals are represented in complex form, including the real part resistance and the imaginary part reactance. The calculation formula is as follows:

[0042] Where, represents the total impedance varying with time; the resistance value varying with time; represents the reactance value varying with time; is the imaginary unit, representing the difference between the reactance part and the resistance part; is used to control the weight or influence between the reactance and the resistance; represents the phase, the phase information of the current or voltage varying with time; S12: Since the electrical impedance signal is easily interfered by noise, it is necessary to perform preliminary denoising processing on the signal. A low-pass filter is applied to remove high-frequency noise and retain the low-frequency components of the signal. The formula of the filtered signal is expressed as:

[0043] Where, represents the denoised signal; represents the original electrical impedance signal; represents the impulse response of the filter; represents the adjustment parameter of the filter; controls the frequency response of the filter; S13: The electrical impedance signal may have baseline drift due to equipment errors or external interference, affecting the analysis result; the baseline drift is removed by subtracting the mean value from the signal to ensure that the mean value of the signal is close to zero. The calculation formula is as follows:

[0044] Among them, represents the signal after low-pass filtering and removing baseline drift; represents the denoised signal; represents the mean value of the signal multiplied by the intensity or offset used to control the removal of baseline drift; S14: To avoid the influence of the amplitude difference between different time points of the signal on subsequent analysis, it is necessary to standardize the signal so that it has a zero mean and unit variance, ensuring that the signals are compared on the same scale. The standardization formula is as follows:

[0045] Among them, represents the standardized signal; represents the signal after removing baseline drift, represents the mean value of the standardized signal; represents the standard deviation of the standardized signal; represents the new coefficient used to control the standardization.

[0046] Specifically, the present invention obtains the electrical signal data of cancer tissues through electrical impedance tomography technology, and then preprocesses the obtained impedance signals, including denoising, baseline drift correction, and standardization steps, effectively improving the quality of the signals, being able to efficiently extract the features in the impedance signals, thereby uncovering potential cancer features; compared with traditional cancer examination methods, the present invention does not require invasive operations such as puncture or surgery, achieving non-invasive cancer detection.

[0047] As an embodiment of the present application, the use of discrete cosine transform to convert the signal into a frequency domain representation in step S2 specifically includes: Performing a discrete cosine transform on the time domain signal. The discrete cosine transform can effectively extract the frequency components of the signal and convert the time domain signal into a frequency domain signal. The specific formula is as follows:

[0048] Among them, is the transformed frequency domain signal; is the standardized time domain signal; is the time domain signal weighting coefficient; is the frequency coefficient, representing the discrete cosine transform of the signal at different frequencies; is the length of the signal; is the transformed frequency domain index; is the index of the time domain signal; is the newly added coefficient; used to adjust the amplitude of the final output to control the overall scaling ratio of the overall result.

[0049] As an embodiment of the present application, the frequency domain enhancement in step S2 specifically includes: In the frequency domain, the low-frequency part of the signal usually contains useful biological information. Enhancing the low-frequency part in the frequency domain can further strengthen the effective features of the signal. The formula for the enhanced frequency domain signal is as follows:

[0050] Where: represents the enhanced frequency domain signal; represents the overall gain factor multiplied by the amplitude part of the frequency domain signal at a frequency; is the gain function, which is used to adjust the gain of different frequency components. The gain of the low-frequency part is higher, while the gain of the high-frequency part is lower; is the phase part corresponding to the frequency; is the phase offset parameter, which is used to adjust the signal phase to avoid phase distortion caused by gain adjustment.

[0051] Specifically, after preliminary denoising, baseline drift removal, and normalization processing, a clear impedance signal is obtained. A discrete cosine transform is performed on the normalized time domain signal to extract frequency components, and the frequency domain signal is enhanced to further improve the effective features in the signal, thereby uncovering potential cancer features.

[0052] As an embodiment of the present application, step S3 specifically includes: S31: Input the signal enhanced in the frequency domain into the network model for processing the frequency domain signal; S32: Extract frequency domain features through the network model, and obtain the global feature matrix after completing global dependence modeling.

[0053] As Figure 3 shown, as an embodiment of the present application, the extraction of frequency domain features in step S32 specifically includes: S3211: Use four layers of convolution to extract four levels of frequency domain features from the signal enhanced in the frequency domain . Each layer of convolution can gradually extract more complex frequency domain features to help the model better understand the pattern of the signal. The specific formula is as follows:

[0054]

[0055]

[0056]

[0057] Where: is the output after the first layer of convolution operation, is the first - layer 1x1 convolutional kernel, is the signal after frequency - domain enhancement, is the bias term of the first - layer convolution, which controls the offset of the model; is the output after the second - layer convolution operation, is the second - layer 2x2 convolutional kernel, is the bias term of the second - layer convolution, which controls the offset of the model; is the output after the third - layer convolution operation, is the third - layer 3x3 convolutional kernel, is the bias term of the third - layer convolution, which controls the offset of the model; is the output after the fourth - layer convolution operation, is the fourth - layer 4x4 convolutional kernel, is the bias term of the fourth - layer convolution, which controls the offset of the model; S3212: After four - layer convolution, the obtained feature map is the output of the deepest layer. To further improve the expression ability of the model and reduce information redundancy, a pooling - based feature aggregation method is used to integrate these frequency - domain features. The formula is as follows:

[0058] Among them, is the aggregated feature map; is the feature map output after four - layer convolution.

[0059] As an embodiment of the present application, the specific process of obtaining the global feature matrix after completing global - dependence modeling in step S32 includes: S3221: Input the pooled features into the global - dependence modeling layer. In this layer, a self - attention mechanism is used to calculate the dependence relationship between features. The formula is as follows:

[0060] Among them, is the pooled feature; represents the weight matrices of query, key, and value, which are learning parameters obtained through training; represents the bias terms of query, key, and value; S3222: Use a self - attention mechanism to calculate the correlation between the query and the key. By taking the dot - product of the query and the key, the attention weight of each feature can be obtained. The specific formula is as follows:

[0061] Among them, represents the attention weight, which contains the dependence relationship between features. Each feature is weighted according to the correlation with other features; Represents the query vector of the input feature at the current time step; is The column vector obtained after transposition, which changes from a column vector to a row vector; Represents the dimension of the key vector, used for normalization, Is an additional bias term, and the calculated similarity can be adjusted, making the feature have a greater influence during calculation; S3223: After calculating the attention scores, next, multiple attention weights need to be weighted and summed with the value matrix, and the formula is as follows:

[0062] Among them, Is the global feature matrix after weighted summation; Is the attention weight obtained through calculation; Is the value matrix.

[0063] Specifically, the frequency-domain enhanced signal is input into the network model and normalized to ensure that the data is further processed on the same scale. Then, the network model extracts the frequency-domain features, layer by layer extracting more complex features to help the model better understand the pattern of the signal. Each layer of convolution gradually extracts the frequency-domain features. After four layers of convolution, a pooling operation is used to aggregate the features, enhancing the model's expressive ability and reducing information redundancy. Next, the pooled features are input into the global dependency modeling layer, and the self-attention mechanism is used to calculate the dependency relationship between the features, enhancing the model's attention to key information. By calculating the similarity between the query and the key and weighting and summing with the value matrix, the final global feature matrix is obtained for further cancer detection and classification tasks.

[0064] As an embodiment of the present application, step S4 inputs the final global feature matrix into the classifier, and the output obtained through the calculation of the fully connected layer specifically includes: Taking the global feature matrix obtained through weighted summation as the input of the classifier, through a flattening operation, it is input into a classifier, and the output is obtained through the calculation of the fully connected layer. The formula is as follows:

[0065]

[0066] Among them, Is a flattening operation, flattening the multi-scale fused features into a one-dimensional vector, preparing to enter the fully connected layer; Is the weight matrix of the fully connected layer; Is the output of the fully connected layer; Is the weight matrix of the fully connected layer; Is the bias term.

[0067] As an embodiment of the present application, in step S4, the output is converted into a probability value through a function, and it can be determined whether cancer exists, and the corresponding detection results specifically include: The output of the fully connected layer is converted into a probability value through a function for classification, and the probability of each category is calculated. The specific calculation formula is:

[0068] where is the probability value of the th category; the calculation method is based on the adjustment of given features and parameters; and are the modified input features; and are adjustable parameters used to control the influence of features on the final result; represents the number of categories.

[0069] Specifically, the global feature matrix obtained by weighted summation is input into the classifier. After flattening, it is input into the fully connected layer for calculation to obtain the output of the classifier. Then, the output of the fully connected layer is converted into the probability value of each category through a function, and the probability of each category is calculated based on the adjustment of features and adjustable parameters, thereby realizing the classification detection of cancer.

[0070] 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, but also should 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. An electrical impedance cancer detection method based on frequency domain information enhancement, characterized in that: The method comprises the following steps: S1: Obtain the electrical impedance signal in the patient's body through electrical impedance imaging technology and pre-process it; S2: Perform frequency domain conversion on the preprocessed time domain signal, use discrete cosine transform to convert the signal into frequency domain representation and perform frequency domain enhancement; S3: Inputting the features enhanced in the frequency domain into an improved network model, wherein the improved network model combines the frequency domain information and the global dependency relationship, and processes the features enhanced in the frequency domain to obtain a final global feature matrix; S4: The final global feature matrix is ​​input into the classifier, and the output is calculated by the fully connected layer. The output is then converted into a probability value through a function, which can determine whether cancer exists and give the corresponding test results.

2. The electrical impedance cancer detection method based on frequency domain information enhancement according to claim 1 is characterized in that: The step S1 specifically includes: S11: Through electrical impedance imaging technology, current is injected into the human body and reflected signals are recorded at different electrode positions. These signals reflect the conductivity of different tissues in the body and contain amplitude and phase information in the time domain. These signals are expressed in complex form, including real resistance and imaginary reactance. The calculation formula is as follows: in, represents the total impedance that varies with time; Resistance value changing with time; Indicates the reactance value that varies with time; It is an imaginary unit, which indicates the difference between the reactance part and the resistance part; Used to control the weight or influence between reactance and resistance; Indicates the phase, the phase information of the current or voltage changing with time; S12: Perform preliminary denoising on the signal and apply a low-pass filter to remove high-frequency noise and retain the low-frequency components of the signal. The formula of the filtered signal is expressed as: in, represents the denoised signal; Represents the original electrical impedance signal; represents the impulse response of the filter; Indicates the adjustment parameters of the filter; controls the frequency response of the filter; S13: Remove the baseline drift by subtracting the mean from the signal to ensure that the mean of the signal is close to zero. The calculation formula is as follows: in, represents the signal after removing the baseline drift; represents the signal after denoising; represents the mean of the signal multiplied by the intensity or offset used to control for baseline drift removal; S14: Standardize the signal to make it have zero mean and unit variance to ensure that the signals are compared on the same scale. The standardization formula is as follows: in, represents the normalized signal; represents the signal after removing the baseline drift, represents the mean of the normalized signal; represents the standard deviation of the normalized signal; Represents new coefficients that control normalization.

3. The electrical impedance cancer detection method based on frequency domain information enhancement according to claim 2 is characterized in that: The step S2 of using discrete cosine transform to convert the signal into frequency domain representation specifically includes: A discrete cosine transform is performed on the time domain signal. The discrete cosine transform can effectively extract the frequency component of the signal and convert the time domain signal into a frequency domain signal. The specific formula is as follows: in, is the transformed frequency domain signal; is the normalized time domain signal; is the time domain signal weighting coefficient; is the frequency coefficient, which represents the discrete cosine transform of the signal at different frequencies; is the length of the signal; is the transformed frequency domain index; is the index of the time domain signal; is a newly added factor; it is used to adjust the magnitude of the final output to control the overall scaling of the overall result.

4. The electrical impedance cancer detection method based on frequency domain information enhancement according to claim 3 is characterized in that: The frequency domain enhancement in step S2 specifically includes: In the frequency domain, the low-frequency part of the signal contains useful biological information. Enhancing the low-frequency part in the frequency domain can further enhance the effective characteristics of the signal. The formula of the enhanced frequency domain signal is as follows: in, represents the enhanced frequency domain signal; It represents the overall gain factor multiplied by the amplitude part of the frequency domain signal in frequency; is a gain function, which is used to adjust the gain of different frequency components; is the phase part corresponding to the frequency; It is a phase offset parameter used to adjust the signal phase to avoid phase distortion caused by gain adjustment.

5. The electrical impedance cancer detection method based on frequency domain information enhancement as described in claim 4, characterized in that: The step S3 specifically includes: S31: inputting the frequency domain enhanced signal into the network model to process the frequency domain signal; S32: Extract frequency domain features through the network model, and obtain the global feature matrix after completing global dependency modeling.

6. The electrical impedance cancer detection method based on frequency domain information enhancement as described in claim 5, characterized in that: The extraction of frequency domain features in step S32 specifically includes: S3211: Signal enhanced from frequency domain using four-layer convolution Four levels of frequency domain features are extracted from the convolution. Each layer of convolution can gradually extract more complex frequency domain features to help the model better understand the signal pattern. The specific formula is as follows: in, is the output after the first layer of convolution operation, is the first layer 1x1 convolution kernel, is the signal after frequency domain enhancement, It is the bias term of the first convolution layer, which controls the offset of the model; is the output after the second layer of convolution operation, is the 2x2 convolution kernel of the second layer, It is the bias term of the second layer of convolution, which controls the offset of the model; is the output after the third layer of convolution operation, is the 3x3 convolution kernel of the third layer, It is the bias term of the third layer of convolution, which controls the offset of the model; is the output after the 4th convolution operation. is the 4x4 convolution kernel of the 4th layer, It is the bias term of the 4th convolution layer, which controls the offset of the model; S3212: After four layers of convolution, the feature map obtained is the deepest output. In order to further improve the expressiveness of the model and reduce information redundancy, a pooling-based feature aggregation method is used to integrate these frequency domain features. The formula is as follows: in, is the aggregated feature; It is the feature map output after four layers of convolution.

7. The electrical impedance cancer detection method based on frequency domain information enhancement as claimed in claim 6, characterized in that: After completing the global dependency modeling in step S32, the global feature matrix is ​​obtained, which specifically includes: S3221: The aggregated features are input into the global dependency modeling layer, where the self-attention mechanism is used to calculate the dependency between features. The formula is as follows: in, is the feature after pooling; Weight matrices representing queries, keys, and values, which are learned parameters obtained through training; Bias items representing queries, keys, and values; S3222: A self-attention mechanism is used to calculate the correlation between the query and the key. The attention weight of each feature is obtained by the dot product of the query and the key. The specific formula is as follows: in, Represents the attention weight, which includes the dependencies between features. Each feature is weighted according to the relevance of other features. Represents the query vector of the input features at the current time step; is the vector obtained after transposing; represents the dimension of the key vector, used for normalization, is an additional bias term; S3223: After the attention score is calculated, multiple attention weights and value matrices are weighted and summed. The formula is as follows: in, is the global feature matrix after weighted summation; is the calculated attention weight; is a matrix of values.

8. The electrical impedance cancer detection method based on frequency domain information enhancement as claimed in claim 7, characterized in that: In step S4, the final global feature matrix is ​​input into the classifier, and the output obtained by the fully connected layer calculation specifically includes: The global feature matrix obtained by weighted summation is used as the input of the classifier, and is input into a classifier through a flattening operation. The output is calculated through the fully connected layer. The formula is as follows: in, It is a flattening operation, and the multi-scale fused features are flattened into a one-dimensional vector; is the weight matrix of the fully connected layer; is the weight matrix of the fully connected layer; is the bias term; is the output of the fully connected layer.

9. The electrical impedance cancer detection method based on frequency domain information enhancement as claimed in claim 8, characterized in that: In step S4, the output is converted into a probability value through a function, so as to determine whether cancer exists and give corresponding detection results, which specifically include: The output of the fully connected layer is converted into a probability value through a function for classification, and the probability of each category is obtained by calculation. The specific calculation formula is: in, It is The probability value of a class; the calculation method is based on the adjustment of given features and parameters; and is the modified input feature; and It is an adjustable parameter used to control the impact of the feature on the final result; Indicates the number of categories.

Images