Plasma near infrared spectrum mode recognition method for detecting latent hepatitis B
A discriminant model for detecting occult hepatitis B was established using plasma near-infrared spectroscopy pattern recognition, which solved the problems of complexity and cost in detecting occult hepatitis B and achieved rapid and accurate screening.
Patent Information
- Application Number
- CN202610208021.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-12
- Publication Date
- 2026-05-26
AI Technical Summary
Current technologies are insufficient for effectively detecting occult hepatitis B infection (OBI), making HBV DNA testing complex and expensive, and prone to missed diagnoses.
A plasma near-infrared spectroscopy pattern recognition method was adopted. By collecting plasma samples, a spectral dataset was constructed. A discriminant model was established using the k-nearest neighbor algorithm and partial least squares discriminant analysis algorithm. Spectral preprocessing and wavelength selection were performed to achieve rapid screening for occult hepatitis B.
It enables rapid, simple, and reagent-free detection of occult hepatitis B, suitable for large-scale screening, improving the accuracy and efficiency of testing while reducing costs.
Smart Images
Figure CN122084570A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of spectral diagnostic technology for infectious diseases, specifically to a plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B. Background Technology
[0002] Hepatitis B virus infection (HBV) is a serious infectious disease and a global public health problem. Occult HBV infection (OBI) is a specific form of HBV infection, characterized by a negative plasma hepatitis B surface antigen (HBsAg) but a positive HBV DNA test. HBV DNA testing is relatively complex and expensive, and is usually not included in routine health checkups, making it easy to miss OBI diagnoses.
[0003] Near-infrared (NIR) spectroscopy offers advantages such as speed and simplicity, and holds significant potential for application in the medical field. From a molecular perspective, the blood of individuals infected with HBV (OBI) contains HBV DNA (including four open reading frames: S, C, P, and X), causing changes in the content of various specific amino acids. Furthermore, OBI and normal control blood differ in protein metabolism, fatty acid metabolism, and related enzymes. These molecules all contain hydrogen-containing groups (XH), which exhibit absorption characteristics in the near-infrared spectral region, leading to differences in the overall spectral patterns of the two plasma types.
[0004] Therefore, it is feasible to use NIR spectral pattern recognition method to distinguish between OBI and normal control plasma.
[0005] Currently, no studies have been reported involving the discriminant analysis of blood spectra between OBI and normal controls. Therefore, exploring and establishing a rapid blood OBI screening method based on NIR spectroscopy is of great significance for infectious disease prevention and control.
[0006] This invention will introduce a wavelength selection method for detecting occult hepatitis B based on plasma NIR spectral pattern recognition. Summary of the Invention
[0007] The technical problem to be solved by the present invention is to overcome the above-mentioned technical defects and provide a plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B, which can provide a solution for the development of a small dedicated blood analyzer for the detection of occult hepatitis B infection.
[0008] To solve the above-mentioned technical problems, the technical solution provided by the present invention is: a method for plasma near-infrared spectral pattern recognition for the detection of occult hepatitis B, comprising the following steps: S1: Collect spectral data of the test plasma sample in the near-infrared band; S2: Acquire multiple repeats of near-infrared spectral data for each sample and construct a spectral dataset; S3: Divide the samples in the spectral dataset into groups of occult hepatitis B infection group and normal control group, and randomly divide each group of samples into training set, prediction set and independent external test set that do not participate in modeling; S4: Constructing an OBI-normal control spectral discrimination model for plasma samples based on a classifier primitive algorithm; S5: The spectral data is preprocessed using a derivative filtering method with adjustable multiple parameters, and the preprocessing parameters are optimized based on the discriminative performance of the classifier. S6: Construct the optimal wavelength combination; S7: Use the plasma sample spectra from an external independent test set that is not involved in the modeling to test the optimal model.
[0009] Preferably, S1 includes acquiring the raw spectral data of plasma samples from the occult hepatitis B infection group, plasma samples from the normal control group, and plasma samples to be tested in the near-infrared band, wherein the spectral scanning range is 895-1707nm.
[0010] Preferably, in step S2, the spectrum of each sample is measured 5 times, and the 5 spectra and their average spectrum are included in the spectral dataset.
[0011] Preferably, the classifier primitive algorithm in S4 includes the k-nearest neighbor algorithm and the partial least squares discriminant analysis algorithm to establish a plasma OBI-normal control spectral discrimination model.
[0012] Preferably, the derivative filtering method in S5 is Norris derivative filtering; The preprocessing parameters include the smoothing window size and the derivative order, with the classification accuracy of the prediction set as the optimization objective, and the optimal parameter combination is determined through cross-validation.
[0013] Preferably, constructing the preferred wavelength combination in step S6 includes: S6.1: Using equally spaced combinations, traverse wavelength combinations with different intervals within the 895nm–1707nm band, and use the classifier's discrimination accuracy on the prediction set as the evaluation index to select the initial optimal wavelength subset; S6.2: Based on the obtained optimal wavelength subset, wavelengths are gradually eliminated, individual wavelengths are removed, and the remaining wavelength combination that makes the classifier's discrimination performance optimal is retained each time. This process is repeated until a single wavelength remains, which is then determined as the preferred wavelength combination with the best discrimination performance.
[0014] Preferably, the preferred wavelength combination includes 19 wavelengths based on the k-nearest neighbor algorithm: 1155, 1181, 1225, 1232, 1238, 1258, 1264, 1270, 1277, 1283, 1289, 1296, 1302, 1309, 1315, 1321, 1328, 1334, and 1340 nm.
[0015] Preferably, the preferred wavelength combination includes 36 wavelengths based on the partial least squares discriminant analysis algorithm: 1225, 1340, 1347, 1385, 1404, 1416, 1423, 1429, 1435, 1448, 1454, 1461, 1467, 1473, 1479, 1486, 1492, 1498, 1505, 1511, 1517, 1523, 1530, 1542, 1548, 1555, 1561, 1567, 1573, 1579, 1586, 1592, 1598, 1604, and 1610 nm.
[0016] The advantages of this invention compared to the prior art are: This invention proposes a plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B infection and establishes a plasma OBI-normal control discrimination model. This method requires no reagents, is rapid and simple, and is suitable for screening occult hepatitis B infection in large populations. This invention integrates plasma sample collection and clinical detection of occult hepatitis B, near-infrared spectroscopy measurement, training-prediction-external testing design, classifier primitive algorithm, spectral preprocessing, wavelength selection, and model external testing to achieve integrated innovation of the method; its innovation lies in the selection of wavelength combinations for discriminative analysis of occult hepatitis B infection using plasma NIR spectral pattern recognition. The low-wavelength model proposed in this invention can provide a solution for developing a small, dedicated blood analyzer for detecting occult hepatitis B infection; This invention addresses the risk of missed occult hepatitis B infection in conventional hepatitis B surface antigen testing by proposing a novel rapid detection method for occult hepatitis B based on plasma near-infrared spectroscopy pattern recognition. This method is of great significance for infectious disease control and blood safety. Attached Figure Description
[0017] Figure 1 This is an overall flowchart of the method of the present invention. Detailed Implementation
[0018] The present invention will now be described in further detail with reference to the accompanying drawings.
[0019] Combined with appendix Figure 1 As shown, this invention proposes a plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B. The core of this method is the use of NIR technology to construct a wavelength combination model for discriminative analysis of occult hepatitis B in plasma. This method requires no reagents, is rapid and simple, and is suitable for large-scale, low-cost rapid screening.
[0020] The objective of this invention is achieved through the following technical solution: wavelength selection in a plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B, comprising the following steps: S1. Plasma Sample Collection and Clinical Detection of Occult Hepatitis B: Plasma samples were collected from voluntary blood donors. Hepatitis B surface antigen (HBsAg) was detected twice by two independent operators using an enzyme-linked immunosorbent assay (ELISA) with two different kits and detection systems. HBsAg-negative samples were retested using a high-sensitivity electrochemiluminescence immunoassay (ECLIA) method for confirmation; HBV DNA was detected in HBsAg-negative samples using a cobas 6800 nucleic acid detection system. HBV DNA positive and negative results in HBsAg-negative samples were identified as OBI and normal controls, respectively.
[0021] S2. Near-infrared spectroscopy measurement: A near-infrared spectrometer (spectral range 895-1707nm) was used. The transmission measurement accessory for liquid samples was a 1mm quartz cuvette. The spectra of plasma samples were measured, with each sample having its spectrum measured 5 times. The 5 spectra and their average spectrum were included in the dataset for training, prediction, and external verification of the spectra. S3, Training-Prediction-External Test Design: OBI (positive) and normal control (negative) samples are divided into four groups according to category and gender. Each group is randomly divided into training set, prediction set and external test set. The training set and prediction set are used for modeling and parameter optimization, while the independent external test set, which does not participate in modeling, is used to evaluate the model. S4. Classifier primitive algorithm: k-nearest neighbor (kNN) and partial least squares discriminant analysis (PLS-DA) models are used as classifier primitive algorithms to establish a plasma OBI-normal control spectral discrimination model; S5. Spectral preprocessing: Norris derivative filtering (NDF) with multiple modes was used for spectral preprocessing; based on the modeling effects of kNN and PLS-DA models, the preprocessing parameters were optimized. S6. Wavelength Selection: Based on spectral preprocessing, the equal-interval combination (EC) method based on variable initial wavelength, number of wavelengths and number of wavelength intervals was used for wavelength selection in the first stage; based on the wavelength selection in the first stage, the wavelength elimination (WSP) method was used for wavelength selection in the second stage; both stages optimized wavelength combinations based on the modeling effects of the kNN and PLS-DA models. S7. External Model Validation: The optimal model is validated using the spectra of plasma samples from an external, independent validation set that are not involved in the modeling process, thus completing the method evaluation.
[0022] In specific implementation of the present invention, This invention uses plasma near-infrared (NIR) spectral pattern recognition for the identification of occult hepatitis B infection as an example. It systematically introduces an innovative implementation method integrating plasma sample collection and clinical detection of occult hepatitis B, near-infrared spectral measurement, training-prediction-external testing design, classifier primitive algorithms, spectral preprocessing, wavelength selection, and model external testing. k-nearest neighbor (kNN) and partial least squares-discriminant analysis (PLS-DA) are used as primitive algorithms for spectral discriminant analysis; Norris derivative filtering (NDF) is used for spectral preprocessing; and equal interval combination (EC) combined with wavelength elimination (WSP) is used for wavelength selection. The implementation method and effects of wavelength selection in the plasma near-infrared spectral pattern recognition method for occult hepatitis B detection are described in detail, but the implementation of this invention is not limited to this.
[0023] The specific implementation steps are as follows: S1. Plasma sample collection and clinical detection of occult hepatitis B. With the approval of the ethics committee and the informed consent of the volunteers, plasma samples were collected from 433 voluntary blood donors aged 18-60 years. Hepatitis B surface antigen (HBsAg) was detected twice by two independent operators using an enzyme-linked immunosorbent assay (ELISA) with two different kits and detection systems. HBsAg-negative samples were retested using a high-sensitivity electrochemiluminescence immunoassay (ECLIA) method; HBV DNA was detected in HBsAg-negative samples using a cobas 6800 nucleic acid detection system. In HBsAg-negative samples, those testing positive and negative for HBV DNA were identified as OBI and normal controls, respectively. Through these rigorous clinical tests, among the 433 plasma samples, 203 were OBI samples (165 males and 38 females), and 230 were normal control samples (178 males and 52 females).
[0024] S2, Near-infrared spectroscopy measurement The NIRQuest+ 1.7 spectrometer (Ocean Optics, USA) was used, with an InGaAs photodiode array detector (Hamamatsu). The spectral scanning range was 895-1707 nm (wavelength interval 1.63 nm, wavelength number 512). A liquid transmission measurement accessory (1 mm quartz cuvette) was used. The integration time for the spectral measurement was 2 ms, and the signal averaging was performed 10 times. The spectra were measured at room temperature of 25 ± 1 °C and relative humidity of 46 ± 1%. Five spectra were measured for each sample, and the average spectrum of the five spectra was included in the training, prediction, and external verification datasets for the spectra.
[0025] S3, Training-Prediction-External Testing Design A rigorous training-prediction-external validation experimental design was employed. Samples were divided into training, prediction, and external validation sets. The training and prediction phases were used for model building and optimization; in the external validation phase, independent samples not involved in modeling were used for model validation. OBI (positive) and normal control (negative) samples were divided into four groups according to category and gender. Each group was randomly assigned to the training set (OBI 85; healthy 85), prediction set (OBI 60; healthy 60), and external validation set (OBI 58; healthy 85). The training and prediction sets were used for modeling and parameter optimization, while the independent external validation set was used to evaluate the model.
[0026] Table 1. Sample categories and number of samples for training, prediction, and external test sets. The evaluation metrics for the model are as follows: In the modeling phase, for the training set and the prediction set, the identification accuracy of negative samples, positive samples, and total samples is calculated respectively. , , , , , ; %), the calculation formula is as follows: , , (1) , , (2) in, , , and These represent the number of negative and positive spectra detected clinically in the training and prediction sets, respectively. , , and These represent the number of correctly identified negative and positive spectra in the training and prediction sets, respectively. The accuracy is based on the total number of samples identified in the prediction set. The optimal model is selected based on the maximum value.
[0027] In the external testing set, the number of positive and negative sample spectra after clinical testing are respectively denoted as... , The number of correctly identified negative and positive spectra is recorded as follows: , The number of false negatives and false positives in spectral identification were respectively , The sensitivity (SEN), specificity (SPE), false negative rate (FNR), and false positive rate (FPR) of the external test set are calculated using the following formulas: , , , (3) S4, Classifier primitive algorithm, see above; S5, Spectral Preprocessing Multi-mode Norris derivative filtering (NDF) was used for spectral preprocessing. The derivative order (…) was employed. d ), smoothing points ( s ) and difference interval ( g Using the search parameters, all NDF preprocessing modes are traversed, and the corresponding preprocessed spectra are used for modeling. OBI-normal control spectral discrimination models based on kNN and PLS-DA are established respectively. The overall recognition accuracy of the prediction set is used as the basis for the modeling. The maximum value determines the optimal NDF parameters. In the current embodiment, the NDF parameters are set as follows: derivative order. d =0, 1, 2; smoothing points s =1, 3, ..., 21; number of difference intervals g= 1, 2, ..., 20.
[0028] S6, Wavelength Selection Based on spectral preprocessing, the equal-interval combination (EC) method with forward optimization was used for wavelength selection in the first stage; an initial wavelength ( I ), number of wavelengths ( N ) and wavelength intervals ( G The search parameters are used to iterate through all equally spaced wavelength combinations within the wavelength range for modeling. OBI-normal control spectral discrimination models based on kNN and PLS-DA are established respectively. The optimal wavelength combination is determined based on the maximum total recognition accuracy of the prediction set. In this embodiment, the wavelength search range is 895-1707 nm, the total number of wavelengths is 512, and the parameters are set as follows: , , .
[0029] Building upon the wavelength selection in the first stage, a backward-optimized wavelength elimination (WSP) method was used for the second stage. From the existing wavelength combinations, one wavelength was eliminated at a time based on an ascending wavelength order. The remaining wavelength combinations were used for modeling, establishing OBI-normal control spectral discrimination models based on kNN and PLS-DA, respectively. The overall recognition accuracy of the prediction set was then used. The maximum value is determined to be the wavelength to be eliminated that gives the best modeling effect to the remaining wavelength combination. Based on this, the same method is used to eliminate one more continuous wavelength, and so on, until only one wavelength remains. Then, the corresponding global optimal model is selected from the best model of each round of elimination.
[0030] Establish two methods with forward and backward optimization functions: the equal-interval combination-wavelength elimination-k nearest neighbor (EC-WSP-kNN) method and the equal-interval combination-wavelength elimination-partial least squares-discriminant analysis (EC-WSP-PLS-DA) method.
[0031] S7, Plasma OBI-Normal Control Spectral Discriminant Model Using the spectral preprocessing method proposed in S5, kNN and PLS-DA models based on NDF spectra were established, and their optimal NDF parameters were respectively... d =2, s =3, g =1 and d =2, s =5, g =1, modeling parameters and discrimination results are shown in Table 2.
[0032] Using the wavelength selection method proposed in S6, EC-WSP-kNN and EC-WSP-PLS-DA models based on NDF preprocessed spectra were established. The optimal wavelength combination for EC-WSP-kNN is: 1155, 1181, 1225, 1232, 1238, 1258, 1264, 1270, 1277, 1283, 1289, 1296, 1302, 1309, 1315, 1321, 1328, 1334, 1340 (nm). N =19); The optimal wavelength combination for EC-WSP-PLS-DA is: 1225, 1340, 1347, 1385, 1404, 1416, 1423, 1429, 1435, 1448, 1454, 1461, 1467, 1473, 1479, 1486, 1492, 1498, 1505, 1511, 1517, 1523, 1530, 1542, 1548, 1555, 1561, 1567, 1573, 1579, 1586, 1592, 1598, 1604, 1610 (nm). N =36). See Table 3 for modeling parameters and discrimination results.
[0033] Table 2 Modeling parameters and discrimination performance of the optimal kNN and PLS-DA models based on NDF preprocessing Method Lv <![CDATA[RAR P ]]> kNN 512 2 3 1 1 - 99.2% PLS-DA 512 2 5 1 - 6 100% Table 3 Modeling parameters and discrimination performance of the optimal EC-WSP-kNN and optimal EC-WSP-PLS-DA models based on NDF preprocessing Method Lv <![CDATA[RAR P ]]> EC-WSP-kNN 19 2 3 1 1 - 100% EC-WSP-PLS-DA 36 2 5 1 - 8 100% S8. External Model Testing Plasma samples from an external, independent test set not involved in modeling were used to test the EC-WSP-kNN and EC-WSP-PLS-DA models based on NDF preprocessed spectra (see Table 4). The optimal EC-WSP-PLS-DA model achieved 100% SEN and SPE, while the optimal EC-WSP-kNN model achieved 98.3% and 98.8% SEN and SPE, respectively. The plasma NIR spectral pattern recognition method demonstrated high accuracy in identifying occult hepatitis B infection.
[0034] Table 4. Results of independent external test sets for the optimal EC-WSP-kNN and EC-WSP-PLS-DA models. Method SEN SPE FNR FPR EC-WSP-kNN 98.3% 98.8% 1.7% 1.2% EC-WSP-PLS-DA 100% 100% 0% 0% Experimental results show that the near-infrared spectroscopy pattern recognition method can accurately discriminate between plasma OBI and normal controls. The proposed low-wavelength model can provide a solution for developing a small, dedicated blood analyzer for detecting occult hepatitis B infection. This method is reagent-free, rapid, and simple, and suitable for screening large populations for occult hepatitis B infection. The detection and modeling methods proposed in this embodiment are also expected to be applied to the discriminant analysis of other blood samples.
[0035] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.
Claims
1. A method for plasma near-infrared spectral pattern recognition in the detection of occult hepatitis B, characterized in that: Includes the following steps: S1: Collect spectral data of the test plasma sample in the near-infrared band; S2: Acquire multiple repeats of near-infrared spectral data for each sample and construct a spectral dataset; S3: Divide the samples in the spectral dataset into groups of occult hepatitis B infection group and normal control group, and randomly divide each group of samples into training set, prediction set and independent external test set that do not participate in modeling; S4: Constructing an OBI-normal control spectral discrimination model for plasma samples based on a classifier primitive algorithm; S5: The spectral data is preprocessed using a derivative filtering method with adjustable multiple parameters, and the preprocessing parameters are optimized based on the discriminative performance of the classifier. S6: Construct the optimal wavelength combination; S7: Use the plasma sample spectra from an external independent test set that is not involved in the modeling to test the optimal model.
2. The plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B according to claim 1, characterized in that: S1 includes acquiring the raw spectral data of plasma samples from the occult hepatitis B infection group, plasma samples from the normal control group, and plasma samples to be tested in the near-infrared band, wherein the spectral scanning range is 895-1707nm.
3. The plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B according to claim 1, characterized in that: The S2 measures the spectrum of each sample five times, and the five spectra and their average spectrum are included in the spectral dataset.
4. The plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B according to claim 1, characterized in that: The classifier primitive algorithm in S4 includes the k-nearest neighbor algorithm and the partial least squares discriminant analysis algorithm to establish a plasma OBI-normal control spectral discrimination model.
5. The plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B according to claim 4, characterized in that: The derivative filtering method in S5 is Norris derivative filtering; The preprocessing parameters include the smoothing window size and the derivative order, with the classification accuracy of the prediction set as the optimization objective, and the optimal parameter combination is determined through cross-validation.
6. The plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B according to claim 5, characterized in that: The construction of the preferred wavelength combination in S6 includes: S6.1: Using equally spaced combinations, traverse wavelength combinations with different intervals within the 895nm–1707nm band, and use the classifier's discrimination accuracy on the prediction set as the evaluation index to select the initial optimal wavelength subset; S6.2: Based on the obtained optimal wavelength subset, wavelengths are gradually eliminated. Each time, the remaining wavelength combination that makes the classifier's discrimination performance optimal is retained. This process is repeated until a single wavelength remains, which is then determined as the preferred wavelength combination with the best discrimination performance.
7. The plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B according to claim 6, characterized in that: The preferred wavelength combination includes 19 wavelengths based on the k-nearest neighbor algorithm: 1155, 1181, 1225, 1232, 1238, 1258, 1264, 1270, 1277, 1283, 1289, 1296, 1302, 1309, 1315, 1321, 1328, 1334, and 1340 nm.
8. The plasma near-infrared spectral pattern recognition method for detecting occult hepatitis B according to claim 6, characterized in that: The preferred wavelength combination includes 36 wavelengths based on the partial least squares discriminant analysis algorithm: 1225, 1340, 1347, 1385, 1404, 1416, 1423, 1429, 1435, 1448, 1454, 1461, 1467, 1473, 1479, 1486, 1492, 1498, 1505, 1511, 1517, 1523, 1530, 1542, 1548, 1555, 1561, 1567, 1573, 1579, 1586, 1592, 1598, 1604, and 1610 nm.