A body surface non-invasive tumor detection method, device and medium

Abstract

The application discloses a body surface noninvasive tumor detection method, device and medium, and relates to the technical field of biomedical engineering. The method comprises the following steps: acquiring mixed modal spectra of multiple detection wavebands of a to-be-detected tissue; performing initial fitting on the fluorescence modal spectrum of each mixed modal spectrum to obtain a corresponding initial fluorescence modal fitting spectrum; initializing the weight of all detection wavebands; updating and iterating the weight, the fluorescence modal fitting spectrum, a weighting operator and fitting noise; determining a multi-modal feature vector based on the fluorescence modal fitting spectrum, the weighting operator, the mixed modal spectrum and the fitting noise under each iteration number; inputting the multi-modal feature vector into k first-order prediction models respectively to output initial prediction values of tumor property vectors; splicing the initial prediction values of the tumor property vectors to obtain an indirect feature vector; and inputting the indirect feature vector into a second-order prediction model to obtain final prediction values of the tumor property vectors. The application improves the precision of tumor detection.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of biomedical engineering, in particular to a body surface non-invasive tumor detection method, device and medium. BACKGROUND

[0002] As a major disease that seriously threatens human health, the high incidence and harm of tumors cannot be ignored in today's society. In the comprehensive treatment strategy system of tumors, early diagnosis and screening occupy a core position, which can enable doctors to intervene in time when tumors are still in the early stage and have not occurred extensive metastasis or deterioration, thereby significantly improving the treatment success rate and long-term survival rate of patients. Therefore, early, non-invasive and accurate judgment of tumor properties is of great significance for the treatment of related diseases.

[0003] At present, tumor property detection is mostly based on imaging methods such as nuclear magnetic resonance and ultrasonic examination. Nuclear magnetic resonance can obtain high-resolution and high-identification image data, and has high discrimination for benign and malignant lesions, but the scanning time is long and the operation is complex, which is not suitable for large-scale tumor screening. At the same time, it has high requirements for equipment and environment, and high cost. Ultrasonic examination is low in price and simple in operation, but it is easy to be disturbed by blood clots, ultrasonic pseudo-state shots and the like to cause artifacts, affecting the imaging quality. The above detection methods need to image and detect after the tumor forms a space-occupying effect, and depend on the subjective judgment and diagnosis of imaging doctors, which has low sensitivity for early tumor screening, and patients may have discomfort during the detection process, increasing the physiological and psychological burden of patients. Therefore, it is necessary to study a non-invasive, painless, automatic and early body surface tumor diagnosis method.

[0004] Raman spectroscopy is a molecular fingerprint spectrum, which can study the structure and vibration mode of a substance based on inelastic scattering. This technology is a non-invasive and painless surface detection technology, which reflects the differences at the molecular level by displaying the Raman peaks of specific molecules and groups, and has the ability to distinguish biological molecules that have occurred lesions but have not formed tumor masses, and has the potential to realize early diagnosis of tumor properties. In addition, Raman spectroscopy has no strong absorption effect on water and is suitable for biological sample detection. The entire detection process usually only takes a few seconds to a few minutes and is simple to operate.

[0005] Miniature fiber Raman spectroscopy mainly consists of a miniature spectrometer, a laser, Raman fibers and their probes, a processor, and an algorithm model. Miniature Raman spectrometers are small in size, low in cost, and easy to operate. They can directly detect human tissues, facilitating the development of small, intelligent tumor detection instruments, enabling rapid, safe, non-invasive, painless, and early accurate diagnosis. When acquiring Raman spectra using a miniature Raman spectrometer, excited biological materials in the visible spectral region emit fluorescence. Various endogenous fluorophores from different tissues exhibit complex fluorescence signals, usually dominated by the intrinsic fluorescence of the sample's inherent organic molecules. This fluorescence also contains characteristic information that can be used for tumor diagnosis.

[0006] Currently, for high-precision tumor characterization, some research teams use a combination of multiple types of spectroscopic instruments to obtain multimodal spectra. However, this approach increases hardware costs, the size of the instrument system, and operational complexity, making it unsuitable for clinical use.

[0007] The spectra of tumors on the body surface acquired by fiber optic Raman spectrometers have mixed modes, with severe overlap between Raman and fluorescence modes, which masks their respective spectral characteristics. This limits the use of existing pattern recognition methods and makes it impossible to accurately detect the nature of tumors. Summary of the Invention

[0008] The purpose of this application is to provide a non-invasive method, device, and medium for detecting tumors on the body surface, thereby improving the accuracy of tumor detection.

[0009] To achieve the above objectives, this application provides the following solution:

[0010] In a first aspect, this application provides a non-invasive method for detecting tumors on the body surface, including:

[0011] A fiber optic Raman spectrometer was used to obtain mixed-mode spectra of multiple target wavelength bands of the tissue under test.

[0012] The initial fluorescence mode spectrum of the mixed mode spectrum of each band to be detected is initially fitted to obtain the initial fluorescence mode fitted spectrum of the corresponding band to be detected.

[0013] Initialize the weights of all bands to be inspected;

[0014] Using weighted adaptive mode decomposition, based on the initial weights, mixed mode spectra, and initial fluorescence mode fitting spectra of each detected band, the weights, fluorescence mode fitting spectra, weighting operators, and fitting noise of the corresponding detected bands are updated and iterated to obtain the weights, fluorescence mode fitting spectra, weighting operators, and fitting noise of the detected tissue at each iteration number for each detected band.

[0015] determine the normalized Raman modal spectrum and the normalized fluorescence modal spectrum corresponding to the to-be-detected wave band based on the fluorescence modal fitting spectrum under each iteration number of each to-be-detected wave band, the weighted operator under each iteration number, the mixed modal spectrum of each to-be-detected wave band, and the fitting noise under each iteration number of the to-be-detected tissue;

[0016] determine the Raman feature vector of the to-be-detected tissue based on the normalized Raman modal spectrum of all to-be-detected wave bands;

[0017] determine the fluorescence feature vector of the to-be-detected tissue based on the normalized fluorescence modal spectrum of all to-be-detected wave bands;

[0018] splice and fuse the Raman feature vector and the fluorescence feature vector of the to-be-detected tissue to obtain a multi-modal feature vector of the to-be-detected tissue;

[0019] input the multi-modal feature vector of the to-be-detected tissue into k first-order prediction models respectively, and output initial prediction values of a corresponding tumor property vector; the k first-order prediction models are models obtained by training k initial models through a first-order training sample set, the first-order training sample set includes multi-modal feature vectors of a plurality of training tissues and actual values of corresponding tumor property vectors; the tumor property vector is used to represent a tumor property, and the tumor property is a malignant tumor, a benign tumor, or a normal tissue; the k initial models are selected from 11 initial models including a support vector machine, a fuzzy information entropy weighted nearest neighbor algorithm, a random forest algorithm, an extreme gradient boosting algorithm, a convolutional neural network, a logistic regression algorithm, a naive Bayes classification algorithm, a linear discriminant analysis, a decision tree, a K nearest neighbor algorithm, and an adaptive boosting algorithm, and k≤11;

[0020] splice the initial prediction values of the tumor property vectors output by all first-order prediction models to obtain an indirect feature vector of the to-be-detected tissue;

[0021] input the indirect feature vector of the to-be-detected tissue into a second-order prediction model to obtain a final prediction value of the tumor property vector of the to-be-detected tissue; the second-order prediction model is obtained by training an adaptive boosting algorithm through a second-order training sample set; the second-order training sample set includes indirect feature vectors of a plurality of training tissues and actual values of corresponding tumor property vectors.

[0022] Optionally, based on the initial weight of each to-be-detected waveband, the mixed modal spectrum and the initial to-be-detected fluorescence modal fitted spectrum, the weight and the fluorescence modal fitted spectrum of the corresponding to-be-detected waveband, the weighting operator and the fitting noise are updated and iterated to obtain the weight and the fluorescence modal fitted spectrum of the corresponding to-be-detected waveband at each iteration number, the weighting operator at each iteration number and the fitting noise of the to-be-detected tissue at each iteration number, comprising:

[0023] any iteration number is determined as a current iteration number, any to-be-detected waveband is determined as a current to-be-detected waveband, and each current to-be-detected waveband at the current iteration number is updated; the update process of the current to-be-detected waveband at the current iteration number comprises:

[0024] the fluorescence modal fitted spectrum of all to-be-detected wavebands at the last iteration number, the weight of the current to-be-detected waveband at the last iteration number and the fitting noise of the to-be-detected tissue at the last iteration number are obtained;

[0025] the weight of the current to-be-detected waveband at the current iteration number is calculated according to the fluorescence modal fitted spectrum of all to-be-detected wavebands at the last iteration number, the weight of the current to-be-detected waveband at the last iteration number, the fitting noise of the to-be-detected tissue at the last iteration number and the mixed modal spectrum of the current to-be-detected waveband;

[0026] the fluorescence modal fitted spectrum of the current to-be-detected waveband at the current iteration number is calculated according to the mixed modal spectrum of the current to-be-detected waveband and the weight of all to-be-detected wavebands at the current iteration number;

[0027] the fitting noise of the to-be-detected tissue at the current iteration number is calculated according to the mixed modal spectrum of the current to-be-detected waveband and the fluorescence modal fitted spectrum of all to-be-detected wavebands at the current iteration number;

[0028] the weighting operator at the current iteration number is calculated according to the mixed modal spectrum of all to-be-detected wavebands and the fluorescence modal fitted spectrum of all to-be-detected wavebands at the current iteration number;

[0029] whether a stop condition is met is judged; the stop condition is that a preset training number is reached or the number of to-be-detected wavebands with non-zero weight at the current iteration number is less than a preset value;

[0030] if yes, the iteration is stopped to obtain the weight and the fluorescence modal fitted spectrum of all to-be-detected wavebands at each iteration number, the weighting operator at each iteration number and the fitting noise of the to-be-detected tissue at each iteration number;

[0031] if no, the current iteration number is updated to a next iteration number, and the process returns to “any to-be-detected waveband is determined as a current to-be-detected waveband, and each current to-be-detected waveband at the current iteration number is updated”.

[0032] Optionally, the weight of the current iteration of the current to-be-detected wave band is calculated according to the fluorescence modal fitting spectrum of all to-be-detected wave bands in the last iteration, the weight of the current to-be-detected wave band in the last iteration, the fitting noise of the to-be-detected tissue in the last iteration and the mixed modal spectrum of the current to-be-detected wave band, comprising:

[0033] The sum of the fitting noise of the to-be-detected tissue in the last iteration and the mixed modal spectrum of the current to-be-detected wave band is determined as a weight update threshold of the current to-be-detected wave band;

[0034] It is judged whether the mixed modal spectrum of the current to-be-detected wave band is greater than the weight update threshold of the current to-be-detected wave band, to obtain a first judgment result;

[0035] If the first judgment result is yes, the weight of the current to-be-detected wave band in the current iteration is set to zero;

[0036] If the first judgment result is no, it is judged whether the mixed modal spectrum of the current to-be-detected wave band is less than the fluorescence modal fitting spectrum of all to-be-detected wave bands in the last iteration, to obtain a second judgment result;

[0037] If the second judgment result is yes, the weight of the current to-be-detected wave band in the current iteration is calculated according to the fluorescence modal fitting spectrum of all to-be-detected wave bands in the last iteration, the weight of the current to-be-detected wave band in the last iteration and the mixed modal spectrum of the current to-be-detected wave band by using a first weight update formula; the first weight update formula is:

[0038]

[0039] Wherein, is the weight of the i-th wave band in the k-th iteration, k≥1; is the weight of the i-th wave band in the (k-1)-th iteration; L k-1 is the fitting error rate of the wave band included in the weight iteration in the (k-1)-th iteration, M is the total number of wave bands of the mixed modal spectrum, is the fluorescence modal fitting error of the i-th wave band in the (k-1)-th iteration, O i is the mixed modal spectrum of the i-th wave band, F i k-1 is the fluorescence modal fitting spectrum of the i-th wave band in the (k-1)-th iteration;

[0040] If the second determination result is no, a second weight updating formula is used to calculate the weight of the current iteration of the current to-be-detected wave band according to the fluorescence modal fitting spectrum of all to-be-detected wave bands in the last iteration, the weight of the last iteration of the current to-be-detected wave band and the mixed modal spectrum of the current to-be-detected wave band; the second weight updating formula is:

[0041]

[0042] Optionally, based on the fluorescence modal fitting spectrum of each to-be-detected wave band in each iteration, the weighting operator in each iteration, the mixed modal spectrum of each to-be-detected wave band and the fitting noise of the to-be-detected tissue in each iteration, the normalized Raman modal spectrum and the normalized fluorescence modal spectrum of the corresponding to-be-detected wave band are determined, comprising:

[0043] determining any to-be-detected wave band as the current to-be-detected wave band;

[0044] calculating the target fluorescence modal fitting spectrum of the current to-be-detected wave band according to the fluorescence modal fitting spectrum of the current to-be-detected wave band in each iteration and the weighting operator in each iteration;

[0045] calculating the target fitting noise of the to-be-detected tissue according to the fitting noise of the to-be-detected tissue in each iteration and the weighting operator in each iteration;

[0046] determining the target Raman modal fitting spectrum of the current to-be-detected wave band according to the mixed modal spectrum of the current to-be-detected wave band and the target fluorescence modal fitting spectrum and the target fitting noise of the to-be-detected tissue;

[0047] normalizing the target fluorescence modal fitting spectrum of the current to-be-detected wave band to obtain the normalized fluorescence modal spectrum of the current to-be-detected wave band;

[0048] normalizing the target Raman modal fitting spectrum of the current to-be-detected wave band to obtain the normalized Raman modal spectrum of the current to-be-detected wave band.

[0049] Optionally, based on the normalized Raman modal spectrum of all to-be-detected wave bands, the Raman feature vector of the to-be-detected tissue is determined, comprising:

[0050] based on the normalized Raman modal spectrum of all to-be-detected wave bands, the comprehensive Raman modal spectrum of the to-be-detected tissue is determined;

[0051] using a peak searching algorithm to search for all characteristic peaks of the comprehensive Raman modal spectrum of the to-be-detected tissue to obtain a plurality of Raman characteristic peaks;

[0052] determining the peak position information and the half-peak width information of each Raman characteristic peak;

[0053] According to the peak position information and the half-peak width information of each Raman characteristic peak, the area of the corresponding Raman characteristic peak is calculated;

[0054] Based on the area of each Raman characteristic peak, the Raman characteristic vector of the tissue to be detected is determined.

[0055] Optionally, based on the normalized fluorescence modal spectrum of all the to-be-detected wave bands, the fluorescence characteristic vector of the tissue to be detected is determined, comprising:

[0056] Based on the normalized fluorescence modal spectrum of all the to-be-detected wave bands, the comprehensive fluorescence modal spectrum of the tissue to be detected is determined;

[0057] Based on the wave band corresponding to each Raman characteristic peak in the comprehensive fluorescence modal spectrum, a plurality of fluorescence characteristic peaks are determined;

[0058] The peak position information and the half-peak width information of each fluorescence characteristic peak are determined;

[0059] According to the peak position information and the half-peak width information of each fluorescence characteristic peak, the area of the corresponding fluorescence characteristic peak is calculated;

[0060] Based on the area of each fluorescence characteristic peak, the fluorescence characteristic vector of the tissue to be detected is determined.

[0061] Optionally, the training process of the k first-order prediction models comprises:

[0062] The actual values of the mixed modal spectrum of a plurality of wave bands and the tumor property vector of a plurality of training tissues are obtained;

[0063] For any training tissue:

[0064] The initial fluorescence modal fitting spectrum of the corresponding wave band is obtained by performing fluorescence modal spectrum initial fitting on the mixed modal spectrum of each wave band;

[0065] The weights of all wave bands are initialized;

[0066] Using weight adaptive modal decomposition, based on the initial weight, the mixed modal spectrum and the initial to-be-detected fluorescence modal fitting spectrum of each wave band, the weight and the fluorescence modal fitting spectrum of the corresponding wave band, the weighting operator and the fitting noise are updated and iterated to obtain the weight and the fluorescence modal fitting spectrum of the corresponding wave band at each iteration number, the weighting operator at each iteration number and the fitting noise of the training tissue at each iteration number;

[0067] Based on the fluorescence modal fitting spectrum at each iteration number of each wave band, the weighting operator at each iteration number, the mixed modal spectrum of each wave band and the fitting noise of the training tissue at each iteration number, the normalized Raman modal spectrum and the normalized fluorescence modal spectrum of the corresponding wave band are determined;

[0068] determine a Raman feature vector of the training tissue based on the normalized Raman modal spectrum of all wavebands;

[0069] determine a fluorescence feature vector of the training tissue based on the normalized fluorescence modal spectrum of all wavebands;

[0070] splice and fuse the Raman feature vector and the fluorescence feature vector of the training tissue to obtain a multi-modal feature vector of the training tissue;

[0071] train the k initial models respectively with the multi-modal feature vector of each training tissue as input and the actual value of the tumor property vector corresponding to the training tissue as output, to obtain k first-order prediction models.

[0072] Optionally, the training process of the second-order prediction model comprises:

[0073] for any training tissue:

[0074] input the multi-modal feature vector of the training tissue into the k first-order prediction models respectively to obtain initial prediction values of the tumor property vector corresponding to the training tissue;

[0075] splice the initial prediction values of the tumor property vector of the training tissue output by all the first-order prediction models to obtain an indirect feature vector of the training tissue;

[0076] train the adaptive enhancement algorithm with the indirect feature vector of each training tissue as input and the actual value of the tumor property vector corresponding to the training tissue as output, to obtain the second-order prediction model.

[0077] In a second aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the tumor detection method of any one of the above aspects.

[0078] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the tumor detection method of any one of the above aspects.

[0079] According to the embodiments of the present application, the following technical effects are achieved:

[0080] 1. The application discloses a body surface non-invasive tumor detection method, device and medium, which measures the Raman spectrum of the body surface skin, and the measurement range includes skin cancer such as melanoma, oral cancer, lymph nodes close to the body surface, breast tumors, thyroid tumors, parotid gland tumors, cervical cancer and other tumors located on the body surface or close to the body surface. The method has the advantages of early detection, non-invasiveness, painlessness and automatic diagnosis, and is suitable for large-scale physical examination screening and outpatient diagnosis.

[0081] 2. Since the spectrum collected by the optical fiber Raman spectrometer contains serious mixed information of Raman mode and intrinsic fluorescence mode, the two modes are mixed with each other, and the spectral characteristic information is covered. The application proposes a mode decomposition method, which decomposes the Raman mode spectrum and the fluorescence mode spectrum from the mixed mode spectrum through mathematical algorithm, maximizes the richness of spectral information, and improves the tumor recognition accuracy without increasing the hardware cost compared with the existing high-precision tumor detection method using multiple spectral instruments to collect multi-modal spectrum.

[0082] 3. The application proposes a weight adaptive mode decomposition method, which comprehensively takes the mode fitting degree and the fitting smoothness as the objective function. In the iteration process, the key waveband is focused, the fitting error rate operator is proposed, the iteration waveband weight is calculated based on the spatial position relationship between the original spectrum and the fitted spectrum, the mode fitting degree of different wavebands of the spectrum is adjusted, the fine degree of spectral mode decomposition is improved, and the degree of waveband weight change is gradually enlarged with the iteration rounds to improve the iteration fitting speed, minimize the mode mixing degree, and finally weight the fitting results of each round based on the fitting error rate, reduce the interference of random noise while ensuring the mode fitting fidelity.

[0083] 4. The application uses the peak area ratio information as the representation of spectral characteristic information to enhance the spectral characteristic difference. The features are extracted on the Raman mode and the fluorescence mode respectively and fused, without additional hardware cost, to obtain a digital multi-modal spectral feature vector, widen the source of biological spectral information, improve the utilization rate of spectral data, fully exert the synergy or complementarity between different modal spectra, and be conducive to obtaining more accurate tumor property judgment results.

[0084] 5. The application adopts a double-stage pattern recognition correction fusion algorithm system, constructs a double-layer progressive pattern recognition model of "first accurate prediction-second intelligent correction", and comprehensively uses the core advantages of various pattern recognition algorithms, so that each sample can be predicted twice, the second correction mechanism is started for the errors of the first prediction, thereby weakening the prediction errors caused by the limitations of a single model, and improving the accuracy and robustness of the body surface tumor property judgment. BRIEF DESCRIPTION OF DRAWINGS

[0085] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0086] Figure 1 The flowchart of the body surface non-invasive tumor detection method provided by an embodiment of the present application is shown in the figure.

[0087] Figure 2 The flowchart of the body surface tumor property determination scheme based on modal decomposition and digital multi-modal fusion is shown in the figure.

[0088] Figure 3 The device architecture diagram of the tumor property detection system provided by the present application is shown in the figure.

[0089] Figure 4 The structural diagram of a computer device provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0090] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0091] The purpose of the present application is to provide a body surface non-invasive tumor detection method, device and medium, aiming to improve the accuracy of tumor detection.

[0092] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the present application will be further described in detail with reference to the drawings and specific embodiments.

[0093] In an exemplary embodiment, as shown in Figure 1 and Figure 2 The body surface non-invasive tumor detection method in the present embodiment comprises:

[0094] Step 01: Using a fiber Raman spectrometer to obtain a mixed modal spectrum of a plurality of detection wavebands of the tissue to be detected.

[0095] In fact, the mixed modal spectrum includes Raman modal spectrum, fluorescence modal spectrum and noise caused by environmental, instrumental and other factors, that is:

[0096] O i =R i +F i +Ni .

[0097] wherein, O i is the mixed modal spectrum of the i-th waveband; R i is the Raman modal spectrum of the i-th waveband; F i is the fluorescence modal spectrum of the i-th waveband; N i is the noise of the i-th waveband.

[0098] Specifically, the mixed modal spectrum is acquired by using a tumor property detection system as shown in Figure 3 The tumor property detection system comprises a control storage unit, a spectrum acquisition unit and a spectrum analysis and processing unit.

[0099] The control storage unit comprises a processor, a memory, a serial communication module, a control program, a key control input and a display device. The control program instructs the processor to execute the spectrum acquisition command, and the spectrum acquisition unit is called through the serial communication to acquire the mixed modal spectrum. The acquired spectrum can be saved in the memory, and the spectrum after acquisition is displayed through the display device. The key control input part can adjust the input parameters. The spectrum analysis and processing unit contains a digital multi-modal decomposition algorithm and a tissue property detection algorithm, and the analysis result is given according to the algorithm. The algorithm is saved in the memory, and the control program calls the algorithm to execute the judgment spectrum, and the display device in the control storage unit displays the judgment result. The spectrum acquisition unit can be a miniature spectrometer or other instruments capable of spectrum acquisition. In the embodiment, the laser adopts a 785nm excitation light source, and the Y-type optical fiber comprises an optical fiber and a probe. The spectrometer is connected with the laser through the Y-type optical fiber. The excitation light emitted by the laser is reflected by the sample surface and received by the spectrometer, so as to acquire the mixed modal spectrum.

[0100] Step 02: The initial fluorescence modal spectrum fitting of the mixed modal spectrum of each waveband to be detected is performed to obtain the initial fluorescence modal fitting spectrum corresponding to the waveband to be detected.

[0101] Specifically, the initial fluorescence modal spectrum fitting of the mixed modal spectrum of each waveband to be detected is performed by using a polynomial fitting to obtain the initial fluorescence modal fitting spectrum corresponding to the waveband to be detected.

[0102] Step 03: The weights of all wavebands to be detected are initialized.

[0103] Specifically, the formula for initializing the weights of all wavebands to be detected is as follows:

[0104]

[0105] wherein, is the initial weight of the i-th waveband; and M is the total number of wavebands of the mixed modal spectrum.

[0106] Step 04: using the weight adaptive modal decomposition, based on the initial weight of each to-be-detected waveband, the mixed modal spectrum and the initial to-be-detected fluorescence modal fitting spectrum, updating and iterating the weight and the fluorescence modal fitting spectrum of the corresponding to-be-detected waveband, the weighting operator and the fitting noise to obtain the weight and the fluorescence modal fitting spectrum of the corresponding to-be-detected waveband at each iteration number, the weighting operator at each iteration number and the fitting noise of the to-be-detected tissue at each iteration number.

[0107] As an optional implementation, step 04 comprises:

[0108] determining any iteration number as a current iteration number, determining any to-be-detected waveband as a current to-be-detected waveband, and updating each current to-be-detected waveband at the current iteration number; the updating process of the current to-be-detected waveband at the current iteration number comprises:

[0109] Step 041: obtaining the fluorescence modal fitting spectrum of all to-be-detected wavebands at the previous iteration number, the weight of the current to-be-detected waveband at the previous iteration number and the fitting noise of the to-be-detected tissue at the previous iteration number.

[0110] Step 042: calculating the weight of the current to-be-detected waveband at the current iteration number according to the fluorescence modal fitting spectrum of all to-be-detected wavebands at the previous iteration number, the weight of the current to-be-detected waveband at the previous iteration number, the fitting noise of the to-be-detected tissue at the previous iteration number and the mixed modal spectrum of the current to-be-detected waveband.

[0111] As an optional implementation, step 042 comprises:

[0112] Step 0421: determining the sum of the fitting noise of the to-be-detected tissue at the previous iteration number and the mixed modal spectrum of the current to-be-detected waveband as the weight update threshold of the current to-be-detected waveband.

[0113] Step 0422: judging whether the mixed modal spectrum of the current to-be-detected waveband is greater than the weight update threshold of the current to-be-detected waveband to obtain a first judgment result.

[0114] Step 0423: if the first judgment result is yes, setting the weight of the current to-be-detected waveband at the current iteration number to zero.

[0115] Step 0424: if the first judgment result is no, judging whether the mixed modal spectrum of the current to-be-detected waveband is less than the fluorescence modal fitting spectrum of the current to-be-detected waveband at the previous iteration number to obtain a second judgment result.

[0116] Step 0425: if the second determination result is yes, then the weight of the current iteration of the current to-be-detected wave band is calculated according to the fluorescence modal fitting spectrum of all to-be-detected wave bands in the last iteration, the weight of the current to-be-detected wave band in the last iteration and the mixed modal spectrum of the current to-be-detected wave band by using a first weight updating formula; the first weight updating formula is:

[0117]

[0118] wherein, is the weight of the i-th wave band in the k-th iteration, k≥1; is the weight of the i-th wave band in the (k-1)-th iteration; L k-1 is the fitting error rate of the wave band included in the weight iteration in the (k-1)-th iteration, M is the total number of wave bands of the mixed modal spectrum, is the fluorescence modal fitting error of the i-th wave band in the (k-1)-th iteration, O i is the mixed modal spectrum of the i-th wave band, F i k-1 is the fluorescence modal fitting spectrum of the i-th wave band in the (k-1)-th iteration.

[0119] Step 0426: if the second determination result is no, then the weight of the current iteration of the current to-be-detected wave band is calculated according to the fluorescence modal fitting spectrum of all to-be-detected wave bands in the last iteration, the weight of the current to-be-detected wave band in the last iteration and the mixed modal spectrum of the current to-be-detected wave band by using a second weight updating formula; the second weight updating formula is:

[0120]

[0121] Step 043: the fluorescence modal fitting spectrum of the current to-be-detected wave band in the current iteration is calculated according to the mixed modal spectrum of the current to-be-detected wave band and the weight of all to-be-detected wave bands in the current iteration.

[0122] Specifically, the calculation formula of the fluorescence modal fitting spectrum is:

[0123] F i k =(A k +βΔ T Δ) -1 A k O i .

[0124] wherein, F i k is the fluorescence modal fitting spectrum of the i-th wave band in the k-th iteration; A kThe matrix composed of the weights under the kth iteration for all wavebands, The weight under the kth iteration for the first waveband, The weight under the kth iteration for the second waveband, The weight under the kth iteration for the Mth waveband; β is a smoothing penalty coefficient, which controls the distribution of each term in the objective function together with the weights, reduces the noise in the fitted mode, and smoothes the fitted spectrum while minimizing the degree of modal aliasing, where β is artificially set to 0.1; Δ is a first-order difference operator, The smoothing degree of the fluorescence mode fitted spectrum can be calculated; T is a transpose.

[0125] Further, the derivation process of the calculation formula of the fluorescence mode fitted spectrum includes:

[0126] In the iterative updating process, the weighted fitting error square with a smoothing penalty term is taken as the objective function:

[0127]

[0128] To solve the objective function, F i k The objective function can be rewritten in the following form:

[0129]

[0130] Wherein, The reference value of the objective function under the kth iteration for the i-th waveband.

[0131] The partial derivative of the reference value of the objective function is obtained

[0132]

[0133] Setting the partial derivative to zero can obtain the calculation formula of the fluorescence mode fitted spectrum.

[0134] Step 044: According to the mixed mode spectrum of the current waveband to be detected and the fluorescence mode fitted spectrum of all wavebands to be detected under the current iteration number, the fitting noise of the current iteration number of the detected tissue is calculated.

[0135] Specifically, the calculation formula of the fitting noise is:

[0136]

[0137] Wherein, ANL k The fitting noise under the kth iteration; The fluorescence mode fitting error under the kth iteration for the i-th waveband; This represents the average value of the fluorescence mode fitting error under the k-th iteration.

[0138] Step 045: Calculate the weighted operator for the current iteration number based on the mixed mode spectra of all bands to be tested and the fluorescence mode fitting spectra of all bands to be tested at the current iteration number.

[0139] Specifically, the calculation formula for the weighted operator is as follows:

[0140]

[0141] Where, ω k This is the weighted operator in the k-th iteration.

[0142] Step 046: Determine whether the stopping condition is met; the stopping condition is that the preset number of training iterations has been reached or the number of test bands with non-zero weights under the current iteration number is less than the preset value.

[0143] Specifically, the preset values ​​include, but are not limited to: 0.001M.

[0144] Step 047: If yes, stop the iteration and obtain the weights and fluorescence mode fitting spectra of all bands to be tested at each iteration number, the weighted operators at each iteration number, and the fitting noise of the tissue to be tested at each iteration number.

[0145] Step 048: If not, update the current iteration number to the next iteration number and return "Determine any band to be tested as the current band to be tested, and update each current band to be tested under the current iteration number".

[0146] Step 05: Based on the fluorescence mode fitting spectra of each test band at each iteration number, the weighting operator at each iteration number, the mixed mode spectra of each test band, and the fitting noise of the tissue to be detected at each iteration number, determine the normalized Raman mode spectrum and normalized fluorescence mode spectrum of the corresponding test band.

[0147] As an optional implementation, step 05 includes:

[0148] Step 051: Select any band to be tested as the current band to be tested.

[0149] Step 052: Calculate the target fluorescence mode fitting spectrum of the current band under test based on the fluorescence mode fitting spectrum of each iteration number and the weighting operator of each iteration number.

[0150] Specifically, the formula for calculating the target fluorescence modal fitting spectrum is as follows:

[0151]

[0152] Among them, Fi e t represents the fitted spectrum of the target fluorescence mode in the i-th band; t is the total number of iterations.

[0153] Step 053: Calculate the target fitting noise of the tissue under test based on the fitting noise at each iteration number and the weighted operator at each iteration number.

[0154] Specifically, the formula for calculating the target fitting noise is:

[0155]

[0156] Among them, ANL e Fit noise to the target of the organization.

[0157] Step 054: Determine the target Raman mode fitting spectrum of the current test band based on the mixed mode spectrum of the current test band, the target fluorescence mode fitting spectrum, and the target fitting noise of the tissue to be detected.

[0158] Specifically, the mixed-mode spectrum of the current test band is used as the processing basis, and the sum of the target fluorescence mode fitting spectrum and the target fitting noise of the tissue to be detected is used as a reference. Differential operation is performed to separate the fluorescence mode spectrum and noise from the mixed-mode spectrum, thereby obtaining the target Raman mode fitting spectrum of the current test band dominated by Raman information and with significantly reduced noise interference.

[0159] Step 055: Normalize the fitted spectrum of the target fluorescence mode in the current test band to obtain the normalized fluorescence mode spectrum of the current test band.

[0160] Specifically, the range of the target fluorescence mode fitting spectrum data for the current test band is limited to [0,1] to obtain the normalized fluorescence mode spectrum of the current test band.

[0161] Step 056: Normalize the fitted spectrum of the target Raman mode in the current test band to obtain the normalized Raman mode spectrum of the current test band.

[0162] Specifically, the range of the target Raman mode fitting spectrum data for the current test band is limited to [0,1] to obtain the normalized Raman mode spectrum for the current test band.

[0163] Step 06: Determine the Raman feature vector of the tissue to be detected based on the normalized Raman mode spectra of all the bands to be detected.

[0164] As an optional implementation, step 06 includes:

[0165] Step 061: Determine the comprehensive Raman mode spectrum of the tissue to be tested based on the normalized Raman mode spectra of all the bands to be tested.

[0166] Step 062: Using the peak-finding algorithm, find all characteristic peaks of the comprehensive Raman mode spectrum of the tissue to be detected, and obtain multiple Raman characteristic peaks.

[0167] Step 063: Determine the peak position and full width at half maximum (FWHM) information of each Raman characteristic peak.

[0168] Specifically, the peak position information of the d Raman characteristic peaks is represented as follows: The half-peak width information is represented as The peak position information of a Raman characteristic peak is the band corresponding to the Raman characteristic peak, and the half-width information of a Raman characteristic peak is half the peak width corresponding to the Raman characteristic peak.

[0169] Step 064: Calculate the area of ​​the corresponding Raman characteristic peak based on the peak position information and half-peak width information of each Raman characteristic peak.

[0170] Specifically, the formula for calculating the area of ​​the Raman characteristic peak is:

[0171]

[0172] in, Let d be the area of ​​the j-th Raman characteristic peak, where j = 1, 2, ..., d; This provides information on the peak position of the j-th Raman characteristic peak. This represents the full width at half maximum (FWHM) information for the j-th Raman characteristic peak. To synthesize Raman mode spectra.

[0173] Step 065: Determine the Raman feature vector of the tissue to be detected based on the area of ​​each Raman feature peak.

[0174] Specifically, Raman eigenvectors are generated by the pairwise ratios of the areas of each Raman characteristic peak.

[0175]

[0176] Step 07: Determine the fluorescence feature vector of the tissue to be tested based on the normalized fluorescence modal spectra of all the bands to be tested.

[0177] As an optional implementation, step 07 includes:

[0178] Step 071: Determine the comprehensive fluorescence modal spectrum of the tissue to be tested based on the normalized fluorescence modal spectra of all the bands to be tested.

[0179] Step 072: Determine multiple fluorescence characteristic peaks in the comprehensive fluorescence modal spectrum based on the bands corresponding to each Raman characteristic peak.

[0180] Step 073: Determine the peak position and full width at half maximum (FWHM) information of each fluorescence characteristic peak.

[0181] Step 074: Calculate the area of ​​the corresponding fluorescence characteristic peak based on the peak position and half-width information of each fluorescence characteristic peak.

[0182] Specifically, the formula for calculating the area of ​​the fluorescence characteristic peak is:

[0183]

[0184] in, Let be the area of ​​the j-th fluorescence characteristic peak, where j = 1, 2, ..., d; This represents the peak position information of the j-th fluorescence characteristic peak; This represents the full width at half maximum (FWHM) information of the j-th fluorescence characteristic peak. This is a comprehensive fluorescence modal spectrum.

[0185] Step 075: Determine the fluorescence feature vector of the tissue to be detected based on the area of ​​each fluorescence characteristic peak.

[0186] Specifically, fluorescence feature vectors are generated by the pairwise ratios of the areas of each fluorescence characteristic peak.

[0187]

[0188] Step 08: The Raman feature vector and fluorescence feature vector of the tissue to be detected are spliced ​​and fused to obtain the multimodal feature vector of the tissue to be detected.

[0189] Step 09: Input the multimodal feature vector of the tissue to be detected into k first-order prediction models respectively, and output the initial predicted value of the corresponding tumor property vector.

[0190] Among them, the k first-order prediction models are models obtained by training the k initial models through the first-order training sample set. The first-order training sample set includes the actual values ​​of multimodal feature vectors of multiple training tissues and the corresponding tumor property vectors. The tumor property vector is used to characterize the tumor property, which is: malignant tumor, benign tumor, or normal tissue. The k initial models are selected from 11 initial models, including support vector machine, fuzzy information entropy weighted nearest neighbor algorithm, random forest algorithm, extreme gradient boosting algorithm, convolutional neural network, logistic regression algorithm, Naive Bayes classification algorithm, linear discriminant analysis, decision tree, K nearest neighbor algorithm and adaptive enhancement algorithm, where k≤11.

[0191] As an optional implementation, the training process for k first-order prediction models includes:

[0192] Step 091: Obtain the actual values ​​of the mixed modal spectra and tumor property vectors of multiple bands from multiple training tissues.

[0193] Step 092: For any training tissue:

[0194] Step 0921: Perform initial fluorescence mode spectral fitting on the mixed mode spectra of each band to obtain the initial fluorescence mode fitting spectra of the corresponding bands.

[0195] Step 0922: Initialize the weights of all bands.

[0196] Step 0923: Using weighted adaptive mode decomposition, based on the initial weights, mixed mode spectra, and initial fluorescence mode fitting spectra of each band, update and iterate the weights, fluorescence mode fitting spectra, weighting operators, and fitting noise of the corresponding bands to obtain the weights, fluorescence mode fitting spectra, weighting operators, and fitting noise of the training tissue at each iteration number for the corresponding bands.

[0197] Step 0924: Based on the fluorescence mode fitting spectra of each band at each iteration number, the weighting operators at each iteration number, the mixed mode spectra of each band, and the fitting noise of the training tissue at each iteration number, determine the normalized Raman mode spectrum and normalized fluorescence mode spectrum of the corresponding band.

[0198] Step 0925: Determine the Raman eigenvectors of the training tissue based on the normalized Raman modal spectra of all bands.

[0199] Step 0926: Determine the fluorescence feature vector of the training tissue based on the normalized fluorescence modal spectra of all bands.

[0200] Step 0927: Concatenate and fuse the Raman feature vector and fluorescence feature vector of the training tissue to obtain the multimodal feature vector of the training tissue.

[0201] Step 093: Using the multimodal feature vectors of each training tissue as input and the actual values ​​of the tumor property vectors of the corresponding training tissues as output, train k initial models respectively to obtain k first-order prediction models.

[0202] Step 10: Concatenate the initial predicted values ​​of the tumor property vectors output by all first-order prediction models to obtain the indirect feature vector of the tissue to be detected.

[0203] Step 11: Input the indirect feature vector of the tissue to be detected into the second-order prediction model to obtain the final predicted value of the tumor property vector of the tissue to be detected.

[0204] The second-order prediction model is obtained by training the adaptive enhancement algorithm using the second-order training sample set. The second-order training sample set includes the actual values ​​of indirect feature vectors of multiple training tissues and the corresponding tumor property vectors.

[0205] As an optional implementation, the training process of the second-order prediction model includes:

[0206] For any training tissue:

[0207] Step 111: Input the multimodal feature vectors of the training tissue into k first-order prediction models respectively to obtain the initial prediction values ​​of the corresponding tumor property vectors.

[0208] Step 112: Concatenate the initial predicted values ​​of the tumor property vectors of the training tissues output by all first-order prediction models to obtain the indirect feature vectors of the training tissues.

[0209] Step 113: Using the indirect feature vectors of each training tissue as input and the actual values ​​of the tumor property vectors of the corresponding training tissues as output, train the adaptive enhancement algorithm to obtain the second-order prediction model. Specific Implementation

[0211] The method of this application was tested using a test set. Four sets of comparative experiments were conducted separately using Support Vector Machine, Fuzzy Information Entropy Weighted Nearest Neighbor Algorithm, Random Forest Algorithm, and Extreme Gradient Boosting Algorithm. A mixed-modality spectral dataset without mode decomposition was used for both training and testing. The training set included 127 malignant breast tumor samples, 100 benign breast tumor samples, and 110 normal breast tissue samples. The test set included 44 malignant breast tumor samples, 30 benign breast tumor samples, and 38 normal breast tissue samples. The spectral band acquired by the fiber optic Raman spectrometer was 0-3000 cm⁻¹. -1 The prediction results on the test set are shown in Table 1. The comparison shows that the method proposed in this application outperforms other methods in terms of recognition accuracy.

[0212] Table 1. Comparison of recognition results of several pattern recognition algorithms with the method of this invention.

[0213]

[0214] In one exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer program to implement a non-invasive tumor detection method for the body surface.

[0215] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements a non-invasive tumor detection method for the body surface.

[0216] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements a non-invasive tumor detection method on the body surface.

[0217] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media to run. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a non-invasive tumor detection method for the body surface.

[0218] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0219] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0220] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0221] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0222] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0223] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for non-invasive tumor detection on the body surface, characterized in that, The body surface non-invasive tumor detection method comprises: Obtaining a plurality of mixed mode spectra of a plurality of detection wavebands of the tissue to be detected by using a fiber Raman spectrometer; Performing initial fitting of the fluorescence mode spectrum on the mixed mode spectrum of each detection waveband to obtain an initial fluorescence mode fitting spectrum corresponding to the detection waveband; Initializing the weight of all detection wavebands; Using weight adaptive mode decomposition, updating and iterating the weight and the fluorescence mode fitting spectrum corresponding to the detection waveband, the weighting operator and the fitting noise based on the initial weight of each detection waveband, the mixed mode spectrum and the initial fluorescence mode fitting spectrum of the detection waveband to obtain the weight and the fluorescence mode fitting spectrum of each iteration number, the weighting operator of each iteration number and the fitting noise of each iteration number of the tissue to be detected; Based on the fluorescence mode fitting spectrum of each iteration number of each detection waveband, the weighting operator of each iteration number, the mixed mode spectrum of each detection waveband and the fitting noise of each iteration number of the tissue to be detected, the normalized Raman mode spectrum and the normalized fluorescence mode spectrum corresponding to the detection waveband are determined; Based on the normalized Raman mode spectrum of all detection wavebands, the Raman feature vector of the tissue to be detected is determined; Based on the normalized fluorescence mode spectrum of all detection wavebands, the fluorescence feature vector of the tissue to be detected is determined; The Raman feature vector and the fluorescence feature vector of the tissue to be detected are spliced and fused to obtain a multi-modal feature vector of the tissue to be detected; The multi-modal feature vector of the tissue to be detected is input into k first-order prediction models respectively, and the initial prediction value of the corresponding tumor property vector is output; the k first-order prediction models are models obtained by training k initial models through a first-order training sample set, the first-order training sample set comprises multi-modal feature vectors of a plurality of training tissues and actual values of corresponding tumor property vectors; the tumor property vector is used to represent tumor properties, and the tumor properties are malignant tumors, benign tumors or normal tissues; the k initial models are selected from 11 initial models including support vector machines, fuzzy information entropy weighted nearest neighbor algorithms, random forest algorithms, extreme gradient boosting algorithms, convolutional neural networks, logistic regression algorithms, naive Bayes classification algorithms, linear discriminant analysis, decision trees, K-nearest neighbor algorithms and adaptive enhancement algorithms, and k≤11; The initial prediction values of the tumor property vectors output by all first-order prediction models are spliced to obtain an indirect feature vector of the tissue to be detected; The indirect feature vector of the tissue to be detected is input into a second-order prediction model to obtain the final prediction value of the tumor property vector of the tissue to be detected; the second-order prediction model is obtained by training an adaptive enhancement algorithm through a second-order training sample set; the second-order training sample set comprises indirect feature vectors of a plurality of training tissues and actual values of corresponding tumor property vectors.

2. The method of claim 1, wherein, The weight adaptive modal decomposition is utilized to update and iterate the weight and the fluorescence modal fitting spectrum of the corresponding to-be-detected wave band, the weighting operator and the fitting noise based on the initial weight of each to-be-detected wave band, the mixed modal spectrum and the initial to-be-detected fluorescence modal fitting spectrum, so as to obtain the weight and the fluorescence modal fitting spectrum of each iteration number of each to-be-detected wave band, the weighting operator of each iteration number and the fitting noise of each iteration number of the to-be-detected tissue, comprising: determining any iteration number as a current iteration number and determining any to-be-detected wave band as a current to-be-detected wave band, and updating each current to-be-detected wave band at the current iteration number; the updating process of the current to-be-detected wave band at the current iteration number comprises: obtaining the fluorescence modal fitting spectrum of all to-be-detected wave bands at the last iteration number, the weight of the current to-be-detected wave band at the last iteration number and the fitting noise of the to-be-detected tissue at the last iteration number; calculating the weight of the current to-be-detected wave band at the current iteration number according to the fluorescence modal fitting spectrum of all to-be-detected wave bands at the last iteration number, the weight of the current to-be-detected wave band at the last iteration number, the fitting noise of the to-be-detected tissue at the last iteration number and the mixed modal spectrum of the current to-be-detected wave band; calculating the fluorescence modal fitting spectrum of the current to-be-detected wave band at the current iteration number according to the mixed modal spectrum of the current to-be-detected wave band and the weight of all to-be-detected wave bands at the current iteration number; calculating the fitting noise of the to-be-detected tissue at the current iteration number according to the mixed modal spectrum of the current to-be-detected wave band and the fluorescence modal fitting spectrum of all to-be-detected wave bands at the current iteration number; calculating the weighting operator at the current iteration number according to the mixed modal spectrum of all to-be-detected wave bands and the fluorescence modal fitting spectrum of all to-be-detected wave bands at the current iteration number; determining whether a stop condition is met; the stop condition is that a preset training number is reached or the number of to-be-detected wave bands with non-zero weight at the current iteration number is less than a preset value; if yes, stopping iteration to obtain the weight and the fluorescence modal fitting spectrum of all to-be-detected wave bands at each iteration number, the weighting operator at each iteration number and the fitting noise of the to-be-detected tissue at each iteration number; if no, updating the current iteration number to a next iteration number and returning to "determining any to-be-detected wave band as a current to-be-detected wave band, and updating each current to-be-detected wave band at the current iteration number".

3. The method of claim 2, wherein the method is a method of detecting a tumor on a body surface without incision. calculating the weight of the current to-be-detected wave band at the current iteration number according to the fluorescence modal fitting spectrum of all to-be-detected wave bands at the last iteration number, the weight of the current to-be-detected wave band at the last iteration number, the fitting noise of the to-be-detected tissue at the last iteration number and the mixed modal spectrum of the current to-be-detected wave band, comprising: determining the sum of the fitting noise of the to-be-detected tissue at the last iteration number and the mixed modal spectrum of the current to-be-detected wave band as a weight update threshold of the current to-be-detected wave band; determining whether the mixed modal spectrum of the current to-be-detected wave band is greater than the weight update threshold of the current to-be-detected wave band to obtain a first determination result; if the first determination result is yes, setting the weight of the current to-be-detected wave band at the current iteration number to zero; If the first determination result is no, it is determined whether the mixed modal spectrum of the current to-be-detected waveband is less than the fluorescent modal fitting spectrum of the previous iteration of the current to-be-detected waveband, to obtain a second determination result; If the second determination result is yes, a first weight updating formula is used to calculate the weight of the current iteration of the current to-be-detected waveband according to the fluorescent modal fitting spectrum of the previous iteration of all to-be-detected wavebands, the weight of the previous iteration of the current to-be-detected waveband and the mixed modal spectrum of the current to-be-detected waveband; the first weight updating formula is: wherein, Wk,i is the weight at the kth iteration for the ith band, k > 1 ; Wk-1,i is the weight at the (k-1)th iteration for the ith band; L k-1 Ek-1 is the fitting error rate of the inclusion weight iteration band at the (k-1)th iteration, M is the total number of bands of the mixed modality spectrum, Ek-1,i is the fluorescence modality fitting error at the (k-1)th iteration for the ith band, O i is the mixed modality spectrum for the ith band, is the fluorescence modality fitting spectrum at the (k-1)th iteration for the ith band; If the second determination result is no, a second weight updating formula is used to calculate the weight of the current iteration of the current to-be-detected waveband according to the fluorescent modal fitting spectrum of the previous iteration of all to-be-detected wavebands, the weight of the previous iteration of the current to-be-detected waveband and the mixed modal spectrum of the current to-be-detected waveband; the second weight updating formula is:

4. The method of claim 3, wherein the method is a method of detecting a tumor on a body surface without incision. Based on the fluorescent modal fitting spectrum of each iteration of each to-be-detected waveband, the weighted operator of each iteration, the mixed modal spectrum of each to-be-detected waveband and the fitting noise of each iteration of the to-be-detected tissue, the normalized Raman modal spectrum and the normalized fluorescent modal spectrum of the corresponding to-be-detected waveband are determined, including: Any to-be-detected waveband is determined as the current to-be-detected waveband; The target fluorescent modal fitting spectrum of the current to-be-detected waveband is calculated according to the fluorescent modal fitting spectrum of each iteration and the weighted operator of each iteration of the current to-be-detected waveband; The target fitting noise of the to-be-detected tissue is calculated according to the fitting noise of each iteration and the weighted operator of each iteration of the to-be-detected tissue; The target Raman modal fitting spectrum of the current to-be-detected waveband is determined according to the mixed modal spectrum of the current to-be-detected waveband, the target fluorescent modal fitting spectrum and the target fitting noise of the to-be-detected tissue; The target fluorescent modal fitting spectrum of the current to-be-detected waveband is normalized to obtain the normalized fluorescent modal spectrum of the current to-be-detected waveband; The target Raman modal fitting spectrum of the current to-be-detected waveband is normalized to obtain the normalized Raman modal spectrum of the current to-be-detected waveband.

5. The method of claim 1, wherein the method is a non-invasive tumor detection method. Based on the normalized Raman modal spectrum of all to-be-detected wavebands, the Raman feature vector of the to-be-detected tissue is determined, including: Based on the normalized Raman modal spectrum of all to-be-detected wavebands, the comprehensive Raman modal spectrum of the to-be-detected tissue is determined; Using a peak searching algorithm, all characteristic peaks of the comprehensive Raman modal spectrum of the to-be-detected tissue are searched to obtain a plurality of Raman characteristic peaks; The peak position information and the half-peak width information of each Raman characteristic peak are determined; The area of the corresponding Raman characteristic peak is calculated according to the peak position information and the half-peak width information of each Raman characteristic peak; Based on the area of each Raman characteristic peak, the Raman feature vector of the to-be-detected tissue is determined.

6. The non-invasive tumor detection method for the body surface according to claim 5, characterized in that, Based on the normalized fluorescent modal spectrum of all to-be-detected wavebands, the fluorescent feature vector of the to-be-detected tissue is determined, including: Based on the normalized fluorescent modal spectrum of all to-be-detected wavebands, the comprehensive fluorescent modal spectrum of the to-be-detected tissue is determined; Based on the waveband corresponding to each Raman characteristic peak in the comprehensive fluorescent modal spectrum, a plurality of fluorescent characteristic peaks are determined; The peak position information and the half-peak width information of each fluorescent characteristic peak are determined; According to the peak position information and the half-peak width information of each fluorescence characteristic peak, the area of the corresponding fluorescence characteristic peak is calculated; Based on the area of each fluorescence characteristic peak, the fluorescence characteristic vector of the tissue to be detected is determined.

7. The method of claim 1, wherein the method is a non-invasive tumor detection method. The training process of the k first-order prediction models includes: Obtaining the actual values of the multi-band mixed modal spectrum and the tumor property vector of the plurality of training tissues; For any training tissue: Performing initial fluorescence modal spectrum fitting on the mixed modal spectrum of each band to obtain the initial fluorescence modal fitting spectrum corresponding to the band; Initializing the weights of all bands; Using weight adaptive modal decomposition, updating and iterating the weight and the fluorescence modal fitting spectrum corresponding to the band, the weighting operator and the fitting noise based on the initial weight of each band, the mixed modal spectrum and the initial to-be-detected fluorescence modal fitting spectrum, to obtain the weight and the fluorescence modal fitting spectrum of each iteration number, the weighting operator of each iteration number and the fitting noise of each iteration number of the training tissue; Based on the fluorescence modal fitting spectrum of each iteration number of each band, the weighting operator of each iteration number, the mixed modal spectrum of each band and the fitting noise of each iteration number of the training tissue, the normalized Raman modal spectrum and the normalized fluorescence modal spectrum corresponding to the band are determined; Based on the normalized Raman modal spectrum of all bands, the Raman characteristic vector of the training tissue is determined; Based on the normalized fluorescence modal spectrum of all bands, the fluorescence characteristic vector of the training tissue is determined; The Raman characteristic vector and the fluorescence characteristic vector of the training tissue are spliced and fused to obtain the multi-modal characteristic vector of the training tissue; The k initial models are trained respectively with the multi-modal characteristic vector of each training tissue as input and the actual value of the tumor property vector of the corresponding training tissue as output, to obtain the k first-order prediction models.

8. The method of claim 7, wherein the method is a method of detecting a tumor on a body surface without incision. The training process of the second-order prediction model includes: For any training tissue: Input the multi-modal characteristic vector of the training tissue into the k first-order prediction models respectively to obtain the initial prediction value of the tumor property vector corresponding to the training tissue; Splice the initial prediction values of the tumor property vectors of the training tissues output by all first-order prediction models to obtain the indirect characteristic vector of the training tissue; Train the adaptive enhancement algorithm with the indirect characteristic vector of each training tissue as input and the actual value of the tumor property vector of the corresponding training tissue as output to obtain the second-order prediction model.

9. A computer apparatus comprising: The memory, the processor and the computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the body surface non-invasive tumor detection method of any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the body surface non-invasive tumor detection method of any one of claims 1-7.

Images