Cancer detection method based on multi-spectrum electrical impedance

Abstract

The invention discloses a cancer detection method based on multi-spectrum electrical impedance. The method comprises the following steps: S1, measuring signal data of normal, benign and malignant samples of cancer tissues by using the multi-spectrum electrical impedance; s2, designing a time sequence block division mechanism, and dividing the acquired multi-spectrum electrical impedance data into time sequence segments in a specific frequency range; s3, constructing a multi-scale signal processing module, performing data filling on the divided time sequence data, and extracting a characteristic mode in a specific frequency range; and S4, according to the characteristics extracted by the multi-scale signal processing module, capturing time sequence and frequency domain interaction characteristics in combination with the network model, and carrying out cancer detection. According to the invention, the accuracy and reliability of cancer detection are improved by designing a time sequence block division mechanism, constructing a multi-scale signal processing module and combining the designed network model.

Description

Technical Field

[0001] The present invention relates to the field of cancer detection, and in particular to a cancer detection method based on multi-spectral electrical impedance. Background Art

[0002] Cancer is one of the most lethal diseases worldwide, causing millions of deaths each year. Since the early symptoms of cancer are relatively hidden, many patients are unable to detect the disease in time in the early stages, causing the disease to develop to the late stage, making treatment difficult and the survival rate low. Therefore, early detection of cancer and timely treatment are of vital importance to improving the survival rate and quality of life of cancer patients. Traditional cancer detection methods include imaging examinations, tissue biopsies, blood tests, urine tests, etc. These methods each have certain advantages and limitations.

[0003] With the rapid development of neural network models, especially the advancement of deep learning technology, the processing of medical images and biological signals has been significantly improved. Neural networks can greatly improve the accuracy and efficiency of cancer detection through automated feature extraction and pattern recognition. In this context, combining multi-spectral electrical impedance technology with neural network models has become a highly promising innovative method.

[0004] In the prior art, a Chinese patent with publication number TW201329433A discloses an "image analysis system and method for cancer cell detection". This method uses an acquisition module to acquire cell images from an image magnification module, surrounds the cell position from the cell image, and analyzes the cell spectrum through multi-spectral color image reproduction to further identify the degree of canceration of the cell. The above scheme mainly identifies the degree of canceration of the cell by analyzing the images between cancer cells. However, only analyzing the images between cancer cells cannot fully exert the detection function, and there are limitations in the accuracy and reliability of cancer detection.

[0005] Therefore, it is urgent to design a cancer detection method based on multi-spectral electrical impedance 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 cancer detection method based on multi-spectral electrical impedance. The electrical signal characteristic data of cancer tissue is collected through multi-spectral electrical impedance technology, and a time series block division mechanism is designed to divide the collected multi-spectral electrical impedance data into time series segments in a specific frequency range. A multi-scale signal processing module is constructed to fill in the incomplete data of the divided time series data, and the characteristic patterns in the specific frequency range are extracted. Based on the extracted features, the interaction characteristics of the time series and frequency domain are captured in combination with the designed network model, so as to detect cancer, thereby improving the accuracy and reliability of cancer detection.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions: The first aspect of the present invention provides a cancer detection method based on multi-spectral electrical impedance, the method comprising the following steps: S1: Use multi-spectral electrical impedance to measure the signal data of normal, benign and malignant samples of cancer tissue. The measured data include conductivity, high-frequency impedance, low-frequency impedance, etc. S2: Design a time series block division mechanism to divide the collected multi-spectrum electrical impedance data into time series segments in a specific frequency range. After the division, the time series signal of each frequency band forms a set containing multiple time series segments; S3: Use a designed network model to build a multi-scale signal processing module, fill in the incomplete data of the divided time series data, extract the characteristic patterns in a specific frequency range and perform feature enhancement; S4: Based on the features extracted by the multi-scale signal processing module and the designed network model, the interaction characteristics of time and frequency domains are captured to detect cancer.

[0008] As an embodiment of the present application, the step S1 specifically includes: S11: Select an electrical impedance spectroscopy instrument that can collect multiple spectra to collect cancer time series signals; S12: Conductivity, high-frequency impedance, and low-frequency impedance are extracted based on the electrical characteristics of the tissue for measuring cancer tissue.

[0009] As an embodiment of the present application, step S2 specifically includes: S21: dividing the frequency range collected by the multi-spectrum electrical impedance spectroscopy instrument into three sub-frequency bands, namely, a low-frequency block segment, a medium-frequency block segment, and a high-frequency block segment; S22: For the three frequency bands collected, a fixed length of time is set to 0.5 seconds, and the step length is 50% of the fixed length of time, which is 0.25 seconds, to ensure that there is a certain overlap in the frequency domains of adjacent blocks; S23: Set the acquisition rate , the acquisition rate is set to 1000 Hz, and each window contains 500 data points; S24: The timing signal of each block frequency band continues to be generated, and the divided timing signal of each frequency band forms 3 sets containing multiple timing segments: , ,

[0010] in, is a collection of segmented low-frequency blocks. is a collection of segmented intermediate frequency blocks. is a collection of high-frequency block segments after segmentation. It is the total number of time series divided using the time series block division mechanism.

[0011] As an embodiment of the present application, step S3 specifically includes: S31: construct a three-dimensional matrix and store the data of the three sub-frequency bands in the three-dimensional matrix; S32: Checking the data integrity of the three-dimensional matrix for the element data in the three-dimensional matrix, and filling the data; S33: Extract and enhance features of the padded data.

[0012] As an embodiment of the present application, the step S31 specifically includes: S311: The time series data collected from the multi-spectral electrical impedance instrument is stored in a three-dimensional matrix. The collected time series signals come from low frequency band, medium frequency band and high frequency band. Each frequency band contains multiple time series segments. The three-dimensional matrix The formula is as follows:

[0013] in, represents the constructed three-dimensional matrix, , , Represents data in three dimensions within a three-dimensional matrix; S312: Data in each frequency band , , The specific three-dimensional matrix storage structure is as follows:

[0014] in, is the number of samples, is the time step of each sample, It is the number of frequency bands, including low frequency band, medium frequency band and high frequency band.

[0015] As an embodiment of the present application, step S32 specifically includes: S321: Three-dimensional matrix Perform integrity check on each element in the three-dimensional matrix, traverse all elements in the three-dimensional matrix, and check the three-dimensional matrix Each element in Is it missing? If the data is incomplete, mark the missing data, otherwise the data is complete. The specific formula is as follows:

[0016] in, represents the constructed three-dimensional matrix, It is a conditional statement used to determine whether the data is complete.

[0017] S322: After judging the data integrity and marking the missing data, a method based on elements in a wider adjacent interval is used to supplement the missing values ​​and fill the data. The specific formula is as follows:

[0018] in, is the missing data to be supplemented. is the time step Around The sum is taken over the time steps, that is, Multiple time steps around, It is a distance-based weighting coefficient used to determine the degree of influence of each adjacent time step on the current time step. The weighting function is as follows:

[0019] in, is the distance between the adjacent time step and the actual time step, It is an added error value to ensure the correctness of data calculation.

[0020] As an embodiment of the present application, step S33 specifically includes: S331: combining the data at the current moment with the data at the previous and next moments and using a weighted sum to represent the relationship between the time series data for feature extraction; S332: applying feature enhancement after combining the moment data; S333: To prevent overfitting and improve the generalization ability of the model, regularization is added. The specific formula is as follows:

[0021] in, It is the feature representation of the current moment output after feature extraction by the multi-scale signal processing module; It is an activation function used to map the input to a non-linear space; is the input data matrix; Yes Connect The linear transformation matrix of the hidden state; Indicates at time Hidden state when is the weight matrix, connecting the hidden state of the previous moment with The linear transformation matrix of the current output; It is the standard representation of the hyperbolic tangent function, which performs a nonlinear transformation on the input value, mapping the value to a numerical range. Indicates the state of the unit at a certain moment; It is a hyperparameter of the regularization strength, which controls the influence of the regularization term and prevents overfitting; is a regularization term that sums the squares of all model parameters; It represents the model , The weight parameter of .

[0022] As an embodiment of the present application, step S4 specifically includes: S41: sending the features extracted by the multi-scale signal processing module into a network model for capturing the interaction characteristics of time series and frequency domains; S42: After the time series and frequency domain interaction characteristics are processed, the cancer detection results are output through network model processing.

[0023] As an embodiment of the present application, the step S41 specifically includes: S411: The input features extracted by the multi-scale signal processing module include time series signals. The network model processes the time series signals through time series convolution, extracting the time-varying pattern of the signal from the time series data, that is, the time-dependent features of the signal. The time domain features capture the dynamic changes of the signal over time; S412: The time series data is converted to the frequency domain through Fourier transform. The frequency domain features capture the pattern of the signal at different frequencies, which is used to capture the frequency difference between cancer tissue and normal tissue. The time domain features and frequency domain features are fused to capture the interactive characteristics of the time series and frequency domain. The specific formula is expressed as follows:

[0024] in, is the input signal in the time domain, is the frequency domain of the input signal, is the Fourier transform.

[0025] As an embodiment of the present application, the step S42 specifically includes: S421: By deeply fusing the extracted time domain and frequency domain features to form a comprehensive feature representation containing time series and frequency domain information, a comprehensive feature representation is obtained by splicing. The specific formula is as follows:

[0026] in, Represents the comprehensive characteristics of splicing time domain and frequency domain, is the input signal in the time domain, is the frequency domain of the input signal; S422: Then the comprehensive features The input is processed into a fully connected layer to further extract deeper information from the fused features and output a linear transformation formula as follows:

[0027] in, is the output of the fully connected layer, , They are the weights and biases of the convolutional layer respectively; S423: The output of the fully connected layer is converted into a category probability through an activation function, and finally the detection result of the tissue is output. The specific formula is as follows:

[0028] in, It is samples belong to the category The predicted probability, It is samples belong to the category The original score, Yes Category The bias term, is an exponential function used to amplify and normalize the original score. is to index and sum the scores of all categories.

[0029] The beneficial effects of the present invention are: (1) The present invention divides the collected electrical impedance data into three sub-bands according to the frequency range by designing a time series block division mechanism. Each segment corresponds to a specific frequency bandwidth. These time series segments can reflect the electrical impedance characteristics within the frequency range in a concentrated manner, ensuring that the characteristic information of different frequency bands is processed and analyzed independently, and effectively processing different frequency characteristics independently. After the signal data of each sub-band is finely divided and processed, a more detailed characteristic pattern can be extracted, thereby further improving the accuracy of cancer detection.

[0030] (2) The present invention constructs a multi-scale signal processing module by utilizing the designed network model to process and analyze the divided time series data, especially to fill in the incomplete data and extract the characteristic patterns in a specific frequency range; the module can effectively fill in the missing data while maintaining the integrity and accuracy of the data, thereby ensuring that the timing characteristics of the signal are not destroyed; through multi-scale processing, the key features of the signal can be captured at different frequency levels, and the characteristic patterns in the low-frequency, medium-frequency and high-frequency ranges can be accurately extracted.

[0031] (3) The features extracted by the multi-scale signal processing module in the present invention, combined with the designed network model, can effectively capture the interaction characteristics between the time domain and the frequency domain, thereby improving the accuracy of cancer detection. By fusing the changing trend of the time domain signal with the spectral characteristics of the frequency domain signal, the network model can identify the key characteristic patterns of different tissue types, thereby achieving more accurate cancer classification and detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic flow chart of a cancer detection method based on multi-spectral electrical impedance provided in an embodiment of the present invention; Figure 2 A schematic diagram of a timing block division mechanism of a cancer detection method based on multi-spectral electrical impedance provided in an embodiment of the present invention; Figure 3 A schematic diagram of feature extraction of a cancer detection method based on multi-spectral electrical impedance provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0035] In the present invention, unless otherwise clearly specified and limited, the terms "connection", "fixation", etc. should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0036] 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 used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the meaning of "and / or" appearing in the full text includes three parallel schemes. Taking "A and / or B" as an example, it includes scheme A, or scheme B, or a scheme that satisfies both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0037] Reference Figures 1 to 3 In a first aspect, the present invention provides a cancer detection method based on multi-spectral electrical impedance, the method comprising the following steps: S1: Use multi-spectral electrical impedance to measure the signal data of normal, benign and malignant samples of cancer tissue. The measured data include conductivity, high-frequency impedance, low-frequency impedance, etc. As an embodiment of the present application, the step S1 specifically includes: S11: Select an electrical impedance spectroscopy instrument that can collect multiple spectra to collect cancer time series signals; S12: Conductivity, high-frequency impedance, and low-frequency impedance are extracted based on the electrical characteristics of the tissue for measuring cancer tissue.

[0038] Specifically, the present invention avoids invasive operations that may be involved in traditional inspection methods by utilizing multi-frequency electrical impedance to measure signal data of normal, benign and malignant samples of cancer tissue. This non-invasive detection method not only reduces the patient's pain and psychological burden, but also greatly reduces the health risks caused by radiation or trauma.

[0039] S2: Design a time series block division mechanism to divide the collected multi-spectrum electrical impedance data into time series segments in a specific frequency range. After the division, the time series signal of each frequency band forms a set containing multiple time series segments; S3: Use a designed network model to build a multi-scale signal processing module, fill in the incomplete data of the divided time series data, extract the characteristic patterns in a specific frequency range and perform feature enhancement; S4: Based on the features extracted by the multi-scale signal processing module and the designed network model, the interaction characteristics of time and frequency domains are captured to detect cancer.

[0040] Specifically, the present invention utilizes multi-frequency electrical impedance to measure the signal data of normal, benign and malignant samples of cancer tissue, designs a time series block division mechanism, divides the collected multi-spectrum electrical impedance data into time series segments in a specific frequency range, and uses a designed network model to construct a multi-scale signal processing module, fills the divided time series data with incomplete data, extracts characteristic patterns in a specific frequency range, and captures the interaction characteristics of the time and frequency domains based on the extracted features in combination with the designed network model, thereby detecting cancer and improving the accuracy of cancer detection.

[0041] As an embodiment of the present application, step S2 specifically includes: S21: dividing the frequency range collected by the multi-spectrum electrical impedance spectroscopy instrument into three sub-frequency bands, namely, a low-frequency block segment, a medium-frequency block segment, and a high-frequency block segment; S22: For the three frequency bands collected, a fixed length of time is set to 0.5 seconds, and the step length is 50% of the fixed length of time, which is 0.25 seconds, to ensure that there is a certain overlap in the frequency domains of adjacent blocks; S23: Set the acquisition rate , the acquisition rate is set to 1000 Hz, and each window contains 500 data points; S24: The timing signal of each block frequency band continues to be generated, and the divided timing signal of each frequency band forms 3 sets containing multiple timing segments: , ,

[0042] in, is a collection of segmented low-frequency blocks. is a collection of segmented intermediate frequency blocks. is a collection of high-frequency block segments after segmentation. It is the total number of time series divided using the time series block division mechanism.

[0043] Specifically, the present invention divides the collected multi-spectrum electrical impedance data into several time series segments according to different frequency ranges through a time series block division mechanism, and each segment corresponds to a specific frequency bandwidth. These time series segments can reflect the electrical impedance characteristics within the frequency range, ensure that the characteristic information of different frequency bands is processed and analyzed independently, and after the signal data of each sub-band is finely divided and processed, a more detailed characteristic pattern can be extracted, thereby improving the representation accuracy of the data. Through this division method, the conductivity changes and impedance responses of cancer tissues at different frequencies can be better captured, providing a more accurate basis for subsequent feature extraction and enhancement.

[0044] As an embodiment of the present application, step S3 specifically includes: S31: construct a three-dimensional matrix and store the data of the three sub-frequency bands in the three-dimensional matrix; S32: Checking the data integrity of the three-dimensional matrix for the element data in the three-dimensional matrix, and filling the data; S33: Extract and enhance features of the padded data.

[0045] As an embodiment of the present application, the step S31 specifically includes: S311: Define the structure of the matrix. The time series data collected from the multi-spectrum electrical impedance instrument needs to be stored in a three-dimensional matrix. The collected time series signals come from low frequency band, medium frequency band and high frequency band. Each frequency band contains multiple time series segments. Define a three-dimensional matrix To represent these data, the three-dimensional matrix The formula is as follows:

[0046] in, represents the constructed three-dimensional matrix, , , Represents data in three dimensions within a three-dimensional matrix; S312: Data in each frequency band , , The specific three-dimensional matrix storage structure is as follows:

[0047] in, is the number of samples, is the time step of each sample, It is the number of frequency bands, including low frequency band, medium frequency band and high frequency band.

[0048] As an embodiment of the present application, step S32 specifically includes: S321: Three-dimensional matrix Perform integrity check on each element in the three-dimensional matrix, traverse all elements in the three-dimensional matrix, and check the three-dimensional matrix Each element in Is it missing? If the data is incomplete, mark the missing data, otherwise the data is complete. The specific formula is as follows:

[0049] in, represents the constructed three-dimensional matrix, It is a conditional statement used to determine whether the data is complete.

[0050] S322: After judging data integrity and marking missing data, use a method to fill in missing values based on elements within a broader adjacent interval to process the data. The specific formula is as follows:

[0051] Among them, is the missing data to be filled in, is the time step around the sum within a number of time steps around, that is, select a number of time steps around,

[0052] Among them, is the distance between the adjacent time step and the actual time step, is an added error value to ensure the correctness of data operations.

[0053] After the data is supplemented as necessary, the formed matrix is sent into a multi-scale signal processing module. The multi-scale signal processing module has the ability to process multi-dimensional data and can flexibly respond. In the multi-scale signal processing module, the data will undergo transformations at multiple scales, and these transformations are similar to disassembling information into particles of different thicknesses in order to capture the characteristics of the data from various levels.

[0054] As an embodiment of the present application, the step S33 specifically includes: S331: Combine the data at the current moment with the data at the previous and next moments and use a weighted sum to represent the relationship between time series data for feature extraction; S332: Apply feature enhancement after combining the data at the moment; S333: Prevent overfitting and improve the generalization ability of the model by adding regularization. The specific formula is as follows:

[0055] Among them, is the feature representation of the current moment output after feature extraction by the multi-scale signal processing module; is an activation function used to map the input to a non-linear space; is the input data matrix; is the connection the linear transformation matrix of the hidden state; represents the hidden state at the moment ; is the weight matrix, a linear transformation matrix that connects the hidden state at the previous moment with the current output; is the standard representation of the hyperbolic tangent function, which non-linearly transforms the input value and maps the value to a numerical interval range, helping the model learn complex non-linear features. represents the cell state at a certain moment; is a hyperparameter of the regularization strength, which controls the influence of the regularization term and prevents overfitting; is a regularization term that sums the squares of all model parameters; represents in the model and weight parameters.

[0056] Specifically, the present invention uses the multi-scale signal processing module constructed by the designed network model to process the divided time-series data through the network. Fill in the missing data in the case of incomplete data to ensure the integrity and quality of the data. For the signal in each specific frequency range, extract the impedance feature pattern related to cancer.

[0057] As an embodiment of the present application, the step S4 specifically includes: S41: Send the features extracted by the multi-scale signal processing module into a network model for capturing the interaction characteristics of time series and frequency domain; S42: After the interaction characteristics of time series and frequency domain are processed and passed through the network model, output the cancer detection result.

[0058] As an embodiment of the present application, the step S41 specifically includes: S411: The input features extracted by the multi-scale signal processing module contain time series signals. The network model processes the time series signals through temporal convolution to extract the patterns of signal changes over time from the time series data, that is, the time-dependent features of the signal. The time domain features capture the dynamic changes of the signal over time; S412: Convert the time series data to the frequency domain through Fourier transform. The frequency domain features capture the patterns of the signal at different frequencies, which are used to capture the frequency differences between cancerous tissues and normal tissues. Fuse the time domain features and the frequency domain features to capture the interaction characteristics of time series and frequency domain; The specific formula is as follows:

[0059] where, is the time domain in the input signal, is the frequency domain of the input signal, is the Fourier transform.

[0060] As an embodiment of the present application, the step S42 specifically includes: S421: By deeply fusing the extracted time domain and frequency domain features to form a comprehensive feature representation containing time series and frequency domain information, the fusion strategy can be achieved by splicing to obtain a comprehensive feature representation. The specific formula is as follows:

[0061] in, Represents the comprehensive characteristics of splicing time domain and frequency domain, is the input signal in the time domain, is the frequency domain of the input signal; S422: Then the comprehensive features The input is processed into a fully connected layer to further extract deeper information from the fused features and output a linear transformation formula as follows:

[0062] in, is the output of the fully connected layer, , They are the weights and biases of the convolutional layer respectively; S423: After the output of the fully connected layer network, the output of the fully connected layer is converted into a category probability through an activation function, and finally the detection result of the organization is output. The specific formula is as follows:

[0063] in, It is samples belong to the category The predicted probability, It is samples belong to the category The original score, Yes Category The bias term, is an exponential function used to amplify and normalize the original score. is the exponentiation and summation of the scores for all classes (plus the bias term).

[0064] Specifically, the present invention designs a network model based on the features extracted by the multi-scale signal processing module, and combines the interactive characteristics of the time and frequency domains through a deep learning architecture; through the fusion of the two, the network can identify the special patterns of cancer tissues under the interaction of the time and frequency domains, thereby achieving more accurate cancer detection; combined with the interactive characteristics of the time and frequency domains, the model can efficiently distinguish between normal, benign and malignant samples, further improving the accuracy and robustness of cancer detection.

[0065] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. 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 a specific combination of the above technical features, but 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 above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A cancer detection method based on multi-spectral electrical impedance, characterized in that: The method comprises the following steps: S1: Use multi-spectral electrical impedance to measure the signal data of normal, benign and malignant samples of cancer tissue. The measured data include conductivity, high-frequency impedance, low-frequency impedance, etc. S2: Design a time series block division mechanism to divide the collected multi-spectrum electrical impedance data into time series segments in a specific frequency range. After the division, the time series signal of each frequency band forms a set containing multiple time series segments; S3: Use a designed network model to build a multi-scale signal processing module, fill in the incomplete data of the divided time series data, extract the characteristic patterns in a specific frequency range and perform feature enhancement; S4: Based on the features extracted by the multi-scale signal processing module and the designed network model, the interaction characteristics of time and frequency domains are captured to detect cancer.

2. A cancer detection method based on multi-spectral electrical impedance according to claim 1, characterized in that: The step S1 specifically includes: S11: Select an electrical impedance spectroscopy instrument that can collect multiple spectra to collect cancer time series signals; S12: Conductivity, high-frequency impedance, and low-frequency impedance are extracted based on the electrical characteristics of the tissue for measuring cancer tissue.

3. The cancer detection method based on multi-spectral electrical impedance according to claim 1, characterized in that: The step S2 specifically includes: S21: dividing the frequency range collected by the multi-spectrum electrical impedance spectroscopy instrument into three sub-frequency bands, namely, a low-frequency block segment, a medium-frequency block segment, and a high-frequency block segment; S22: For the three frequency bands collected, a fixed length of time is set to 0.5 seconds, and the step length is 50% of the fixed length of time, which is 0.25 seconds, to ensure that there is a certain overlap in the frequency domains of adjacent blocks; S23: Set the acquisition rate , the acquisition rate is set to 1000 Hz, and each window contains 500 data points; S24: The timing signal of each block frequency band continues to be generated, and the divided timing signal of each frequency band forms 3 sets containing multiple timing segments: , , in, is a collection of segmented low-frequency blocks. is a collection of segmented intermediate frequency blocks. is a collection of high-frequency block segments after segmentation. It is the total number of time series divided using the time series block division mechanism.

4. The cancer detection method based on multi-spectral electrical impedance according to claim 3, characterized in that: The step S3 specifically includes: S31: construct a three-dimensional matrix and store the data of the three sub-frequency bands in the three-dimensional matrix; S32: Checking the data integrity of the three-dimensional matrix for the element data in the three-dimensional matrix, and filling the data; S33: Extract and enhance features of the padded data.

5. The cancer detection method based on multi-spectral electrical impedance according to claim 4, characterized in that: The step S31 specifically includes: S311: The time series data collected from the multi-spectral electrical impedance instrument is stored in a three-dimensional matrix. The collected time series signals come from low frequency band, medium frequency band and high frequency band. Each frequency band contains multiple time series segments. The three-dimensional matrix The formula is as follows: in, represents the constructed three-dimensional matrix, , , Represents data in three dimensions within a three-dimensional matrix; S312: Data in each frequency band , , The specific three-dimensional matrix storage structure is as follows: in, is the number of samples, is the time step of each sample, It is the number of frequency bands, including low frequency band, medium frequency band and high frequency band.

6. The cancer detection method based on multi-spectral electrical impedance according to claim 5, characterized in that: The step S32 specifically includes: S321: Three-dimensional matrix Perform integrity check on each element in the three-dimensional matrix, traverse all elements in the three-dimensional matrix, and check the three-dimensional matrix Each element in Is it missing? If the data is incomplete, mark the missing data, otherwise the data is complete. The specific formula is as follows: in, represents the constructed three-dimensional matrix, It is a conditional statement used to determine whether the data is complete. S322: After judging the data integrity and marking the missing data, a method based on elements in a wider adjacent interval is used to supplement the missing values ​​and fill the data. The specific formula is as follows: in, is the missing data to be supplemented. is the time step Around The sum is taken over the time steps, that is, Multiple time steps around, It is a distance-based weighting coefficient used to determine the degree of influence of each adjacent time step on the current time step. The weighting function is as follows: in, is the distance between the adjacent time step and the actual time step, It is an added error value to ensure the correctness of data calculation.

7. The cancer detection method based on multi-spectral electrical impedance according to claim 5, characterized in that: The step S33 specifically includes: S331: combining the data at the current moment with the data at the previous and next moments and using a weighted sum to represent the relationship between the time series data for feature extraction; S332: applying feature enhancement after combining the moment data; S333: To prevent overfitting and improve the generalization ability of the model, regularization is added. The specific formula is as follows: in, It is the feature representation of the current moment output after feature extraction by the multi-scale signal processing module; It is an activation function used to map the input to a non-linear space; is the input data matrix; Yes Connect The linear transformation matrix of the hidden state; Indicates at time Hidden state when is the weight matrix, connecting the hidden state of the previous moment with The linear transformation matrix of the current output; It is the standard representation of the hyperbolic tangent function, which performs a nonlinear transformation on the input value, mapping the value to a numerical range. Indicates the state of the unit at a certain moment; It is a hyperparameter of the regularization strength, which controls the influence of the regularization term and prevents overfitting; is a regularization term that sums the squares of all model parameters; It represents the model , The weight parameter of .

8. The cancer detection method based on multi-spectral electrical impedance according to claim 1, characterized in that: The step S4 specifically includes: S41: sending the features extracted by the multi-scale signal processing module into a network model for capturing the interaction characteristics of time series and frequency domains; S42: After the time series and frequency domain interaction characteristics are processed, the cancer detection results are output through network model processing.

9. The cancer detection method based on multi-spectral electrical impedance according to claim 8, characterized in that: The step S41 specifically includes: S411: The input features extracted by the multi-scale signal processing module include time series signals. The network model processes the time series signals through time series convolution, extracting the time-varying pattern of the signal from the time series data, that is, the time-dependent features of the signal. The time domain features capture the dynamic changes of the signal over time; S412: The time series data is converted to the frequency domain through Fourier transform. The frequency domain features capture the pattern of the signal at different frequencies, which is used to capture the frequency difference between cancer tissue and normal tissue. The time domain features and frequency domain features are fused to capture the interactive characteristics of the time series and frequency domain. The specific formula is expressed as follows: in, is the input signal in the time domain, is the frequency domain of the input signal, is the Fourier transform.

10. The cancer detection method based on multi-spectral electrical impedance according to claim 9, characterized in that: The step S42 specifically includes: S421: By deeply fusing the extracted time domain and frequency domain features to form a comprehensive feature representation containing time series and frequency domain information, a comprehensive feature representation is obtained by splicing. The specific formula is as follows: in, Represents the comprehensive characteristics of splicing time domain and frequency domain, is the input signal in the time domain, is the frequency domain of the input signal; S422: Then the comprehensive features The input is processed into a fully connected layer to further extract deeper information from the fused features and output a linear transformation formula as follows: in, is the output of the fully connected layer, , They are the weights and biases of the convolutional layer respectively; S423: The output of the fully connected layer is converted into a category probability through an activation function, and finally the detection result of the tissue is output. The specific formula is as follows: in, It is samples belong to the category The predicted probability, It is samples belong to the category The original score, Yes Category The bias term, is an exponential function used to amplify and normalize the original score. is to index and sum the scores of all categories.

Images