Research method for diagnosing early lung cancer based on terahertz sensor array and artificial intelligence technology

Abstract

The invention provides a method for diagnosing early lung cancer based on a terahertz sensor array and an artificial intelligence technology. 16 terahertz sensors with the same structure and high-quality factors (Q value) are adopted to form a 4 * 4 array, and the 16 lung cancer markers are specifically detected through modification by a molecular imprinting technology. During detection, the modified terahertz sensor array is placed in a closed gas chamber made of polytetrafluoroethylene, nitrogen is firstly introduced to obtain a reference spectrum, then gas exhaled by a subject is introduced to obtain a test spectrum, and the concentration of the marker is reflected through the harmonic peak frequency shift amount. An initial diagnosis model is established in combination with individual information of a subject, and the model is trained and optimized by using an artificial intelligence technology and a large sample. According to the method, the problems of insufficient specificity, low sensitivity, high misdiagnosis rate of a single marker and the like in a traditional method are solved, and non-invasive, rapid and accurate diagnosis of the early lung cancer is realized.

Description

Technical Field

[0001] The present invention relates to the field of terahertz technology and computers, and in particular to a research method for diagnosing early lung cancer based on terahertz sensor array and artificial intelligence technology. Background Art

[0002] Terahertz waves (THz) refer to electromagnetic waves with a frequency of 0.1THz-10THz. Their photon energy is very low and can maintain the original structure and activity of molecules. They are particularly suitable for label-free and non-destructive biological detection and have important application value in the biomedical field.

[0003] According to statistics from the GLOB OCAN International Agency for Research on Cancer, lung cancer accounts for 12.4% of cancer subjects and has a mortality rate of 18.7%, both ranking first, and it is the leading cancer in my country. According to the TNM staging method for lung cancer (see Table 1), the 5-year survival rate of subjects with early lung cancer (IA) is as high as 82% after standardized treatment. Due to the insidious onset of lung cancer and the absence of any typical symptoms in the early stages, it is difficult to detect it in time without targeted physical examinations and screening. At present, about 75% of lung cancer subjects are already in the late stage (III / IV) when diagnosed, and the 5-year survival rate is only 6%-36%. Therefore, conducting research on early diagnosis of lung cancer and achieving early detection and early treatment are of great significance to improving the survival rate of lung cancer subjects. Summary of the invention

[0004] The purpose of the present invention is to provide a research method for diagnosing early lung cancer based on terahertz sensor arrays and artificial intelligence technology, aiming to solve the problems of insufficient specificity, low sensitivity and high misdiagnosis rate of single markers in the prior art for early diagnosis of lung cancer.

[0005] To achieve the above object, the present invention adopts a research method for diagnosing early lung cancer based on terahertz sensor array and artificial intelligence technology, comprising the following steps:

[0006] Design and manufacture high Q value terahertz sensors;

[0007] Modifying molecularly imprinted polymers onto metasurface array sensors;

[0008] Passing nitrogen into the gas chamber in which the modified terahertz array sensor is placed, and measuring the terahertz transmission spectrum of the sensor when it is unloaded as an unloaded spectrum;

[0009] The exhaled gas of the test subject is passed into the air chamber where the modified terahertz array sensor is placed, and the terahertz transmission spectrum when the target marker is present on the sensor surface is measured as the test spectrum;

[0010] Compare the test spectrum with the no-load spectrum to obtain the frequency shift of the sensor resonance peak;

[0011] Based on the frequency shift amount and the individual information of the subject, an initial diagnostic model is established using artificial intelligence technology, and the frequency shift amount and individual information of the training group are used for training;

[0012] Finally, the accuracy of the trained model is verified using the frequency shift amount and individual information of the test group, and the model with an accuracy higher than 95% is used for early lung cancer diagnosis.

[0013] The frequency shift amount of the resonant peak of the transmission spectrum is usually used as experimental data for biosensing in the terahertz field. The frequency shift amount of the resonant peak of the transmission spectrum obtained by comparing the test spectrum and the empty cell spectrum is used as the main input for judgment during the subsequent process of training the artificial intelligence technology model, and is used to give a diagnosis description of the lung cancer stage.

[0014] Among them, the method for obtaining the test spectrum and the empty cell spectrum includes the following steps:

[0015] Use 16 sensors to form a 4×4 array, and through molecular imprinting technology, each unit sensor is specific only to one lung cancer biomarker;

[0016] Vacuum the gas chamber, and sequentially test the empty cell spectra of each sensor in the vacuum environment by a two-dimensional scanning method;

[0017] Introduce the enriched gas sample into the gas chamber, close the intake valve after reaching the set air pressure, and sequentially test the terahertz transmission spectra of each sensor by a two-dimensional scanning method to obtain the test spectra of each sensor after ventilation;

[0018] Compare the test spectrum with the empty cell spectrum to obtain the frequency shift amount of the resonant peak of the sample. The method of averaging multiple measurements can reduce the measurement error.

[0019] Among them, the design and processing of high-Q terahertz sensors includes the following steps:

[0020] Use the electromagnetic full-wave analysis software CST Microwave Studio based on the finite-difference time-domain technique to establish an initial structural model of the terahertz sensor;

[0021] Use the frequency shift solver to calculate the transmission coefficient S21 when the terahertz wave is incident perpendicularly to the surface;

[0022] By the method of controlling variables, change one structural parameter each time to obtain the S-parameter curves corresponding to the changes of each structural parameter;

[0023] Compare with the required Q value of the technical index. If it meets the requirements, process the metasurface structure. If it does not meet the conditions, continue to adjust the structural parameters.

[0024] Clean and polish the surface of the silicon wafer, spin-coat a layer of polymer on the silicon wafer surface and perform photolithography;

[0025] Use a mask plate containing a sensor structure for ultraviolet exposure, put the exposed sample into a developer for development, cleaning, and then post-baking;

[0026] Evaporate gold on the surface of the photoresist to peel the polymer from the surface of the silicon wafer, obtaining a sensor with a metal resonator structure.

[0027] Among them, before obtaining the experimental spectrum, it is necessary to perform molecularly imprinted polymer modification on the metasurface array sensor, including the following steps:

[0028] Use a certain lung cancer biomarker to be detected in exhaled gas as the template molecule, and select a functional monomer; the template molecule and the functional monomer form a host-guest complex through self-assembly;

[0029] Select a crosslinking agent / initiator, form a copolymer, break the binding bond, remove the template molecule, and form a molecularly imprinted polymer with good specificity;

[0030] Prepare the molecularly imprinted polymer into a solution with a certain concentration, and use the spin-coating method to evenly coat the molecularly imprinted polymer solution on the surface of the sensor;

[0031] Dry the sensor, peel the polymer from the surface of the silicon wafer, and obtain a sensor with a modified surface.

[0032] In the process of establishing and training a model by combining the frequency shift amount with the individual information of the subject through artificial intelligence technology, including the following steps:

[0033] Encode 8 lung cancer stages and set the stage threshold between stages;

[0034] Adopt a multi-model voting strategy to train 3 deep learning algorithms, and use the frequency shift amount feature values of 16 sensors as the main feature as input 1 for each algorithm;

[0035] Collect the gender, age, past medical history, and living habits of the subject as auxiliary feature as input 2;

[0036] Adopt a multi-modal fusion network to process the main feature and the auxiliary feature respectively;

[0037] 70% of the sample data is training data, and 30% of the sample data is test data;

[0038] When the similarity rate of the output result on the test set is greater than 95%, the deep learning result is considered credible;

[0039] Otherwise, change the network parameters, adjust the stage threshold until the requirement that the similarity rate is greater than 95% is met.

[0040] The present invention discloses a research method for diagnosing early lung cancer based on a terahertz sensor array and artificial intelligence technology. The method diagnoses early lung cancer by constructing a supersurface array sensor combined with artificial intelligence technology. First, sixteen terahertz supersurface sensors with the same structure and high Q value are designed and processed, and the sixteen supersurface sensor array units are modified by molecular imprinting technology; then, nitrogen is introduced into an air chamber in which the modified terahertz array sensor is arranged, and the terahertz transmission spectrum of the sensor when it is unloaded is measured as an unloaded spectrum; then, the exhaled gas of the subject is introduced into the air chamber in which the modified terahertz array sensor is arranged, and the terahertz transmission spectrum when there is a target marker on the surface of the sensor is measured as a test spectrum; the test spectrum is compared with the unloaded spectrum to obtain the frequency shift of the resonance peak, and an initial diagnosis model is established by artificial intelligence technology based on the frequency shift and the individual information of the subject, and the frequency shift and individual information of the training group are used for training, and finally, the accuracy of the trained model is verified by using the frequency shift and individual information of the test group, and the model with an accuracy higher than 95% is used for early diagnosis of lung cancer. This method achieves high-specificity and high-sensitivity diagnosis of early lung cancer, solving the technical problem in existing technologies that a single marker cannot provide a definitive diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 It is a flow chart of a research method for diagnosing early lung cancer based on terahertz sensor array and artificial intelligence technology of the present invention;

[0043] Figure 2 It is a schematic diagram of the steps of modifying the super surface sensor using molecular imprinting technology in a specific embodiment of the present invention;

[0044] Figure 3 is a schematic diagram of a gas detection spectrum system in a specific embodiment of the present invention;

[0045] Figure 4 is a response spectrum diagram of a single sensor to gases of different concentrations in a specific embodiment of the present invention;

[0046] Figure 5 It is a schematic diagram of the steps of a method for training the frequency shift amount in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0047] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0048] In the present application document, related terms may also have corresponding names, such as the projection peak frequency shift Δfi, where i is the frequency shift of the nth unit structure and Δf is the average value.

[0049] See also Figure 1 The present invention provides a research method for diagnosing early lung cancer based on a terahertz sensor array and artificial intelligence technology, comprising the following steps:

[0050] S11: Design and manufacture high Q value terahertz sensors;

[0051] S12: Modification of molecular imprinting polymers onto metasurface array sensors;

[0052] S13: introducing nitrogen into the gas chamber in which the modified terahertz array sensor is placed, and measuring the terahertz transmission spectrum of the sensor when it is unloaded as an unloaded spectrum;

[0053] S14: passing the exhaled gas of the subject into the air chamber of the modified terahertz array sensor to obtain a test spectrum of the terahertz transmission spectrum;

[0054] S15: Compare the test spectrum with the no-load spectrum to obtain the frequency shift of the sensor resonance peak;

[0055] S16: Based on the frequency shift amount and the individual information of the subjects, an initial diagnostic model is established using artificial intelligence technology, and the frequency shift amount and individual information of the training group are used for training;

[0056] S17: Use the frequency shift and individual information of the test group to verify the accuracy of the trained model, and use the model with an accuracy higher than 95% for early diagnosis of lung cancer.

[0057] The design of a high-Q terahertz sensor comprises the following steps:

[0058] S21: Use CSTMicrowave Studio, an electromagnetic field full-wave analysis software based on time-domain finite integration technology, to establish the initial structural model of the terahertz sensor;

[0059] S22: Use the frequency shift solver to calculate the transmission coefficient when the terahertz wave is incident perpendicular to the surface;

[0060] S23: By using the control variable method, one structural parameter is changed each time to obtain the corresponding S parameter curve when each structural parameter changes;

[0061] S24: Compare with the Q value required by the technical indicators. If the requirements are met, the structural design is completed. If the conditions are not satisfied, continue to adjust the structural parameters.

[0062] Among them, in obtaining the terahertz transmission spectrum of the metasurface array sensor, it is necessary to build the Guo Jian metasurface array sensor and the experimental system. The building process includes the following steps:

[0063] S101: Use 16 sensors to form a 4×4 array, and make each unit sensor specific to only one lung cancer biomarker through molecular imprinting technology;

[0064] S102: Evacuate the gas chamber, and sequentially test the no-load spectra of each sensor in the vacuum environment by two-dimensional scanning method;

[0065] S103: Introduce the enriched gas sample into the gas chamber, close the intake valve after reaching the set air pressure, and sequentially test the terahertz transmission spectra of each sensor by two-dimensional scanning method to obtain the test spectra of each sensor after ventilation;

[0066] S104: Compare the test spectrum with the no-load spectrum to obtain the resonance peak frequency shift amount of the sample. The method of averaging multiple measurements can reduce the measurement error.

[0067] Among them, the process of fabricating a high-Q terahertz sensor includes the following steps:

[0068] S31: Polish and clean the silicon wafer surface, spin-coat a layer of polymer on the silicon wafer surface and perform photolithography;

[0069] S32: Perform ultraviolet exposure using a mask plate containing the sensor structure, put the exposed sample into the developer for development, cleaning, and then post-baking;

[0070] S33: Evaporate gold on the photoresist surface and peel off the polymer from the silicon wafer surface to obtain a sensor with a metal resonator structure.

[0071] Among them, the modification process of the metasurface array sensor modified with molecularly imprinted polymer includes the following steps:

[0072] S41: Use a certain lung cancer biomarker to be detected in exhaled breath as the template molecule and select a functional monomer;

[0073] S42: The template molecule and the functional monomer form a host-guest complex through self-assembly;

[0074] S43: Select a crosslinking agent / initiator, form a copolymer, break the binding bond, remove the template molecule, and form a molecularly imprinted polymer with good specificity;

[0075] S44: Prepare the molecularly imprinted polymer into a solution with a certain concentration, and evenly coat the molecularly imprinted polymer solution on the surface of the sensor by spin coating;

[0076] S45: Dry the sensor, peel the polymer from the surface of the silicon wafer to obtain a sensor with a modified surface.

[0077] Among them, in the process of establishing and training a model by combining the frequency shift amount with the individual information of the subject through artificial intelligence technology, the following steps are included:

[0078] S51: Encode 8 lung cancer stages and set the stage thresholds between the stages;

[0079] S52: Adopt the strategy of multi-model voting, train 3 deep learning algorithms, and use the frequency shift amount eigenvalues of 16 sensors as the main features as input 1 for each algorithm;

[0080] S53: Collect the gender, age, past medical history, living habits, etc. of the subject as auxiliary features as input 2;

[0081] S54: Adopt a multi-modal fusion network to process the main features and auxiliary features respectively;

[0082] S55: 70% of the sample data is training data, and 30% of the sample data is test data;

[0083] S56: When the similarity rate of the output results on the test set is greater than 95%, the deep learning results are considered reliable;

[0084] S57: Otherwise, change the network parameters and adjust the stage thresholds until the requirement that the similarity rate is greater than 95% is met.

[0085] The research method for diagnosing early lung cancer based on a terahertz sensor array and artificial intelligence technology designs and fabricates an array composed of sixteen high-Q terahertz metasurface sensors, and uses molecular imprinting technology to functionalize each unit so that it specifically captures sixteen characteristic volatile organic compounds in the exhaled gas of early lung cancer; by comparing the nitrogen empty spectrum with the test spectrum of the gas to be measured, extract the frequency shift amount of each sensor resonance peak to form a sixteen-dimensional feature vector, and input it into the diagnostic model trained with samples and individual information of the subject! Conduct multi-parameter joint analysis, and finally realize non-invasive, highly sensitive and highly specific early diagnosis of lung cancer, effectively solve the limitations of single biomarker detection, and significantly improve the screening accuracy and reliability.

[0086] The present invention also proposes a specific embodiment:

[0087] A research method for diagnosing early lung cancer based on a terahertz sensor array and artificial intelligence technology in this embodiment is as follows:

[0088] The molecularly imprinted polymers modified on the surface of each sensor are differentially designed according to the molecular characteristics of the target lung cancer markers to ensure specific response to the target volatile organic compounds.

[0089] (1) A high-Q resonance structure (such as split-ring resonator, photonic crystal or metamaterial structure) is designed. The molecularly imprinted polymers modified on the surface of each sensor are differentially designed according to the molecular characteristics of the target lung cancer markers to ensure that the array cover can specifically absorb the target volatile organic compounds and cause an obvious frequency shift of the resonance peak.

[0090] (2) Please refer to Figure 2 The corresponding metasurface structures are modified with different molecularly imprinted polymers. The specific steps are as follows: For the characteristic volatile organic compounds in the exhaled gas in the early stage of lung cancer, sixteen highly correlated biomarkers are respectively selected as template molecules, and functional monomers are selected; the template molecules and functional monomers form host-guest complexes through self-assembly; crosslinking agents / initiators are selected to form copolymers; the binding bonds are broken, and the template molecules are removed to form molecularly imprinted polymers with good specificity.

[0091] (3) The molecularly imprinted polymers are formulated into a solution with a certain concentration; the molecularly imprinted polymer solution is uniformly coated on the surface of the corresponding sensor by spin coating; the sensor is dried; the sensor with a modified surface is obtained. The 16 metasurface sensors are placed on a two-dimensional moving platform as shown in Figure 3 to form a metasurface array.

[0092] (5) Please refer to Figure 3 , open the valves b and c on the left side of the figure, introduce high-purity nitrogen into the closed gas chamber of the sensor array, control the valves of the air extractor and the waste treatment bottle to make the reading of the barometer one atmospheric pressure. After the gas chamber environment is stable, use the terahertz time-domain spectroscopy system (THz-TDS) to scan each sensor unit and record the no-load resonance peak frequency of its transmission spectrum (denoted as f oi , i = 1, 2,..., 16);

[0093] (6) Please refer to Figure 3 , collect the exhaled gas sample of the subject to be tested. After pre-treatment of dehumidification and filtration, open the valves a and c on the left side of the figure and introduce it into the gas chamber. Under the same conditions, obtain the test transmission spectra of each sensor unit and extract the test resonance peak frequency (denoted as f ti );

[0094] (7) For an example of the response spectrum of a single sensor to gases with different concentrations, please refer to Figure 4 . For each sensor unit, calculate the resonance peak frequency shift amount Δf i = f ti - f oi, a sixteen - dimensional frequency shift feature vector [Δf1, Δf2,..., Δf 16 is obtained. The magnitude of the frequency shift is positively correlated with the adsorption amount of the target molecule. The resolution ability for complex VOCs mixed gas can be significantly improved through the joint detection of multiple parameters of the array.

[0095] (8) More than 100 subjects with lung cancer at different stages (IA, IB, IIA, IIB, IIIA, IIIB, IV) diagnosed by existing clinical diagnostic methods and healthy people are respectively selected for sampling. The total number of samples is not less than 800, and the samples are corresponding to the diagnostic grades. 70% of them are used as the training group, and the remaining 30% are used as the test group.

[0096] (9) Please refer to Figure 5 , input the obtained sixteen - dimensional frequency shift feature vector of the training group and the individual information of the subjects into the model through artificial intelligence technology for training. Input the test group data into the model. According to the comparison between the diagnostic function value M and the corresponding threshold Nn for different stages, perform lung cancer staging. If the accuracy rate is lower than 95%, modify each threshold Nn and the model parameters until the accuracy rate requirement is met, and then output the diagnostic model.

[0097] The above - disclosed is only a preferred embodiment of the present invention. Of course, it cannot be used to limit the scope of the rights of the present invention. Those of ordinary skill in the art can understand all or part of the processes of implementing the above - mentioned embodiment, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.

Claims

1. A method for diagnosing early-stage lung cancer based on a terahertz sensor array and artificial intelligence technology, characterized in that, The following steps are involved: Design and process high Q value terahertz sensors; Modifying molecularly imprinted polymers onto metasurface array sensors; Passing nitrogen into the gas chamber in which the modified terahertz array sensor is placed, and measuring the terahertz transmission spectrum of the sensor when it is unloaded as an unloaded spectrum; The exhaled gas of the test subject is passed into the air chamber where the modified terahertz array sensor is placed, and the terahertz transmission spectrum when the target marker is present on the sensor surface is measured as the test spectrum; Compare the test spectrum with the no-load spectrum to obtain the frequency shift of the sensor resonance peak; Based on the frequency shift and the individual information of the subjects, an initial diagnostic model is established using artificial intelligence technology, and the frequency shift and individual information of the training group are used for training; Finally, the frequency shift and individual information of the test group were used to verify the accuracy of the trained model, and the model with an accuracy higher than 95% was used for early diagnosis of lung cancer.

2. The method for diagnosing early-stage lung cancer based on a terahertz sensor array and artificial intelligence technology according to claim 1, wherein Obtaining the resonance peak frequency shift comprises the following steps: Sixteen sensors were used to form a 4×4 array, and molecular imprinting technology was used to make each modified sensor unit specific to only one lung cancer marker; The air chamber is evacuated, and the no-load spectrum of each sensor in the vacuum environment is tested in turn by a two-dimensional scanning method; The enriched gas sample is introduced into the gas chamber, and the air inlet valve is closed after reaching the set air pressure. The terahertz transmission spectrum of each sensor is tested in turn by a two-dimensional scanning method to obtain the test spectrum of each sensor after ventilation; The test spectrum is compared with the no-load spectrum to obtain the resonance peak frequency shift of the sample. The method of averaging multiple measurements can reduce the measurement error.

3. The research method for diagnosing early-stage lung cancer based on a terahertz sensor array and artificial intelligence technology according to claim 1, characterized in that, The design of a high-Q terahertz sensor comprises the following steps: The initial structural model of the terahertz sensor was established using the electromagnetic field full-wave analysis software CSTMicr·iwaVe Studio based on the time-domain finite integration technology; The transmission coefficient S21 of the terahertz wave incident perpendicular to the surface is calculated using the frequency shift solver; By using the control variable method, one structural parameter is changed each time to obtain the corresponding S parameter curve when each structural parameter changes; Compare the Q value with the technical indicator requirement. If it meets the requirements, complete the structural design. If it does not meet the conditions, continue to adjust the structural parameters.

4. The research method for diagnosing early-stage lung cancer based on a terahertz sensor array and artificial intelligence technology according to claim 1, wherein The method for processing a high-Q terahertz sensor comprises the following steps: The surface of the polished silicon wafer is cleaned, a polymer layer is spin-coated on the surface of the silicon wafer and photolithography is performed; A mask plate including a sensor structure is used for UV exposure, and the exposed sample is placed in a developer for development, cleaning, and post-baking; Gold is evaporated on the surface of the photoresist to peel off the polymer from the surface of the silicon wafer, thereby obtaining a sensor with a metal resonator structure.

5. The research method for diagnosing early-stage lung cancer based on a terahertz sensor array and artificial intelligence technology according to claim 1, wherein The molecular imprinting polymer modified super surface array sensor comprises the following steps: Using a certain lung cancer marker to be detected in exhaled breath as a template molecule, a functional monomer is selected; The template molecule and the functional monomer form a host-guest complex through self-assembly; Select cross-linking agent / initiator to form copolymer, break the bond, remove the template molecule, and form molecular imprinting polymer with good specificity; The molecular imprinting polymer is prepared into a solution of a certain concentration, and the molecular imprinting polymer solution is evenly coated on the surface of the sensor by a spin coating method; Sensor drying treatment, stripping the polymer from the silicon wafer surface to obtain a surface-modified sensor.

6. The research method for diagnosing early-stage lung cancer based on a terahertz sensor array and artificial intelligence technology according to claim 1, wherein In the process of establishing and training a model by combining the frequency shift amount with the individual information of the subject through artificial intelligence technology, the following steps are included: Encoding 8 lung cancer stages and setting the stage thresholds between the stages; Adopting a multi-model voting strategy, training 3 deep learning algorithms, and using the frequency shift amount eigenvalues of 16 sensors as the main feature as input 1 for each algorithm; Collecting the personal information of the subject (such as gender, age, past medical history, and living habits, etc.) as the auxiliary feature as input 2; Adopting a multi-modal fusion network to process the main feature and the auxiliary feature respectively; 70% of the sample data is training data, and 30% of the sample data is test data; When the similarity rate of the output results on the test set is greater than 95%, the deep learning results are considered credible; Otherwise, change the network parameters and adjust the stage threshold until the requirement that the similarity rate is greater than 95% is met.

Images