Periodontitis classification system based on saliva SERS

Abstract

The invention discloses a periodontitis classification system based on saliva SERS (Surface Enhanced Raman Scattering), relates to the technical field of oral care, and aims to solve the problem that spectral characteristics are difficult to classify and identify obviously due to the fact that complicated and rich components in saliva often interfere with the accuracy and specificity of SERS diagnosis. In order to solve the problem of low periodontitis classification accuracy due to the fact that saliva SERS spectrum is obtained and periodontitis classification is performed on differential metabolites corresponding to saliva SERS spectrum data, the technical scheme of the application takes silver nanoparticles as an SERS substrate, the SERS spectrum is detected, and the spectrum difference is analyzed. Therefore, the interference of complex and rich components in saliva on SERS diagnosis is reduced. And the periodontitis classification accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oral care, and in particular to a periodontitis classification system based on saliva SERS. Background Art

[0002] Periodontitis is a nonspecific, chronic inflammatory disease initiated by bacterial plaque biofilms and mediated by host immune imbalance. It leads to progressive destruction of tooth-supporting tissues, ultimately causing tooth mobility and loss, severely impacting masticatory function. It affects approximately 20% to 50% of the global population. Currently, clinical diagnosis of periodontitis relies primarily on visual inspection, probing, and radiographic imaging. However, these traditional diagnostic methods are complex and expensive, and they only reflect the cumulative damage caused by periodontitis, failing to effectively monitor the current status and progression of the disease. In recent years, saliva has emerged as a potentially effective biosample for disease diagnosis, monitoring, and prognosis. With the advancement of modern laboratory techniques and chemical instrumentation, the application of saliva in both the laboratory and clinical setting has become increasingly widespread. The use of saliva for the diagnosis and monitoring of oral and systemic diseases has become a research hotspot. Currently, the main techniques for analyzing salivary metabolites include nuclear magnetic resonance spectroscopy, liquid chromatography-mass spectrometry, and capillary electrophoresis. However, these methods often require complex, time-consuming procedures and tedious operations. Therefore, the development of a universal, economical, and rapid saliva testing method is of great significance for monitoring the severity of periodontitis.

[0003] Surface-enhanced Raman spectroscopy (SERS), a fingerprint spectroscopy technique, offers advantages such as high sensitivity, ease of use, and rapid detection. It can be used to detect ultra-low concentrations of molecules in complex biological systems such as cells, tissues, and body fluids. However, the complex and rich composition of saliva often interferes with the accuracy and specificity of SERS diagnostics, making it difficult to clearly identify spectral features. This, in turn, leads to low accuracy in periodontitis classification. Summary of the Invention

[0004] The present invention aims to address the problem that the complex and abundant components in saliva often interfere with the accuracy and specificity of SERS diagnosis, making it difficult to clearly classify and identify spectral features, which in turn leads to low accuracy in periodontitis classification. A periodontitis classification system based on saliva SERS is provided.

[0005] The technical solution adopted by the present invention to solve the above technical problems is:

[0006] A periodontitis classification system based on salivary SERS, the system comprising a spectral data acquisition module, a spectral data processing module and a classification module;

[0007] The spectrum data acquisition module is used to acquire saliva SERS spectrum data;

[0008] The spectral data processing module is used to extract differential metabolites corresponding to saliva SERS spectral data;

[0009] The classification module is used to input differential metabolites into the classification network to obtain classification results;

[0010] The saliva SERS spectrum data is obtained by detecting silver nanoparticles as a SERS substrate.

[0011] Furthermore, the silver nanoparticles are obtained by the following steps:

[0012] First, 2.5 mL of 0.1 M AgNO₃ solution was added to 250 mL of ultrapure water containing 0.6 μM ascorbic acid and 3 μM trisodium citrate. The mixture was stirred at 30°C for 15 minutes until the color of the solution stopped changing. The mixture was then reacted at 100°C for 2 hours. During the reaction, ascorbic acid was used as a reducing agent to reduce the single-crystalline silver core, and citrate ions were used as a stabilizer to protect the reduced single-crystalline silver core.

[0013] When the ascorbic acid is completely consumed, the reducing property of citric acid is used to continue reducing the AgNO3 solution, causing the silver nanoparticles to grow slowly, and the remaining particles in the colloid are eliminated through Ostwald ripening, making the surface of the silver nanoparticles smoother. The silver nanoparticles are centrifuged at 10,000 rpm for 10 minutes, the supernatant is discarded, and 1 mL of ultrapure water is added to further concentrate the silver nanoparticle solution, which is then sealed and stored at 4°C for later use.

[0014] Furthermore, the specific steps of obtaining saliva SERS spectrum data are:

[0015] Step 1: Remove any remaining food debris from the mouth. The subject gently touches the roots of the upper or lower teeth with the tip of the tongue to stimulate saliva secretion in a natural, non-stimulating manner, and spits the saliva into a sterile saliva collection tube.

[0016] Step 2: The collected samples were immediately frozen in a -80°C refrigerator;

[0017] Step 3: Take the saliva sample out of the -80°C freezer and thaw it at room temperature;

[0018] Step 4: Centrifuge the thawed saliva sample at 14,000 rpm for 10 minutes, take the supernatant and mix it with silver nanoparticles in a 1:1 ratio. After thorough mixing, let it stand at room temperature until it is naturally dry, and then collect the SERS spectrum.

[0019] Furthermore, in step 4, 10 SERS spectra are collected randomly from the positions of the "coffee ring" formed after the silver nanoparticles and saliva are mixed.

[0020] Furthermore, the differential metabolites include glycogen, ascorbic acid, uric acid, hypoxanthine, glutathione, D-mannose and phenylalanine.

[0021] Furthermore, after the spectral data acquisition module acquires the saliva SERS spectral data, it pre-processes the saliva SERS spectral data through BWSpec and normalizes the pre-processed spectral data.

[0022] Furthermore, the spectral data processing module extracts differential metabolites corresponding to the saliva SERS spectral data through a PLS-DA model.

[0023] Furthermore, the differential metabolites had a variable importance projection value of ≥1 for the first principal component in the PLS-DA model and a P value <0.05 in the variance analysis.

[0024] Furthermore, the classification network is a CNN model.

[0025] Furthermore, the classification network is a CNN-Transformer hybrid model, which includes a CNN model and a Transformer encoder. The CNN model is used for local feature extraction, and the Transformer encoder extracts sequence features based on the local features extracted by the CNN model.

[0026] The CNN model includes three convolutional layers and two pooling layers. The first convolutional layer contains 32 convolution kernels with a convolution kernel size of 3×1, a stride of 1, and a ReLU activation function. The first pooling layer uses maximum pooling with a pooling window size of 2×1 and a stride of 2. The second convolutional layer contains 64 convolution kernels with a convolution kernel size of 3×1, a stride of 1, and a ReLU activation function. The second pooling layer uses maximum pooling with a pooling window size of 2×1 and a stride of 2. The third convolutional layer contains 128 convolution kernels with a convolution kernel size of 3×1, a stride of 1, and a ReLU activation function.

[0027] The Transformer encoder uses sine-cosine functions to generate positional encodings, the dimension of which is consistent with the CNN output feature dimension. Eight attention heads are set, the dimension of each attention head is 64, and the total hidden dimension is 512.

[0028] The beneficial effects of the present invention are:

[0029] This application obtains saliva SERS spectra and classifies periodontitis based on differential metabolites corresponding to these saliva SERS spectral data. Using silver nanoparticles as a SERS substrate, this technology detects SERS spectra and analyzes spectral differences. This reduces the interference of the complex and rich components of saliva on SERS diagnosis, thereby improving the accuracy of periodontitis classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Schematic diagram of monitoring the severity of periodontitis using saliva SERS spectroscopy;

[0031] Figure 2 Schematic diagram of X-ray films of the control group and groups with different severity of periodontitis;

[0032] Figure 3 is a characterization diagram of silver nanoparticles;

[0033] Figure 4 Schematic diagram of saliva SERS technology analysis of different severity levels of periodontitis;

[0034] Figure 5 Schematic diagram of PLS-DA analysis of differences between different severity levels of periodontitis;

[0035] Figure 6 This is a schematic diagram of the CNN model classification results for this application. DETAILED DESCRIPTION

[0036] It should be noted that, unless there is any conflict, the various embodiments disclosed in this application can be combined with each other.

[0037] Specific embodiment 1: A periodontitis classification system based on salivary SERS described in this embodiment is characterized in that the system includes a spectral data acquisition module, a spectral data processing module and a classification module;

[0038] The spectrum data acquisition module is used to acquire saliva SERS spectrum data;

[0039] The spectral data processing module is used to extract differential metabolites corresponding to saliva SERS spectral data;

[0040] The classification module is used to input differential metabolites into the classification network to obtain classification results;

[0041] The saliva SERS spectrum data is obtained by using silver nanoparticles as a SERS substrate and performing detection.

[0042] In order to solve the problems existing in the prior art, this application proposes a technical solution for monitoring the severity of periodontitis (specifically, Figure 1First, using silver nanoparticles as the SERS substrate, the saliva SERS spectra of the control group (periodontal health group) and periodontitis groups of varying severity (mild, moderate, and severe periodontitis) were detected. The spectral differences between the control group and the periodontitis groups of varying severity were analyzed. Subsequently, the PLS-DA model was used to analyze these differential spectral data and screen out metabolites associated with the disease state. Finally, the CNN-Transformer hybrid model was used to analyze the SERS spectral data and establish a classification and identification model for the control group and the periodontitis groups of varying severity. This method shows great potential in monitoring the severity of periodontitis.

[0043] Silver nitrate (AgNO3), trisodium citrate (C6H5Na3O7), and ascorbic acid (AA) were purchased from Sinopharm Chemical Reagent Co., Ltd. Saliva collection tubes were purchased from Suqian Yiyile Trading Co., Ltd. The ultrapure water (18.0 MΩ·cm) used in the experiment was prepared by an Arium 611UV ultrapure water system (Sartorius, Germany). Raman spectra were collected using a BWS415-785H Raman spectrometer (B&W Tek, USA) with a central wavelength of 785 nm and a spectral resolution of less than 3 cm. -1 The laser power was 10 mW, and the integration time was 10 to 30 s. UV-visible absorption spectra were measured using a TU1901 UV-visible spectrophotometer (Beijing Spectroscopy). Scanning electron microscopy (SEM) of the nanoparticles was performed using an SU8010 field emission scanning electron microscope (Hitachi, Japan) with an accelerating voltage range of 0.1 to 30 kV.

[0044] The clinical saliva samples for this application were collected from the Affiliated Hospital of Inner Mongolia University for Nationalities, and the sampling method has been approved by the Medical Ethics Committee (No.: NM-LL-2024-03-05-05). A total of 58 participants were collected, including 16 in the control group (periodontal health group) and 14 in each group with different degrees of periodontitis severity. All participants signed informed consent. The inclusion criteria are as follows: 1) Control group: no probing bleeding, attachment loss, alveolar bone resorption, loose teeth or root bifurcation lesions; 2) The grading criteria for the periodontitis group are detailed in Tables 1 and Figure 2 Exclusion criteria included: 1) systemic disease; 2) use of antibiotics or steroids within the past 3 months; 3) pregnancy or taking sex hormones or contraceptives; 4) history of smoking; 5) occlusal trauma due to severe malocclusion or deformity; and 6) no formal orthodontic treatment or untreated periodontitis. The clinical characteristics of the control group and the groups with different periodontitis severities are detailed in Table 2.

[0045] Table 1. Diagnostic criteria for periodontitis grading

[0046]

[0047] a. Exclude gingival hyperplasia and recession; b. Smoking patients may not bleed on probing

[0048] Table 2 Clinical characteristics of the control group and groups with different severity of periodontitis

[0049]

[0050] The preparation of silver nanoparticles primarily follows the method proposed by Yaqiong Qin et al. Using AA as a reducing agent, a large number of single-crystalline silver nuclei are rapidly reduced from an AgNO₃ solution. During the reaction, citrate ions act as stabilizers, protecting the newly formed silver nuclei and preventing excessive growth, aggregation, and precipitation. Once the AA is consumed, the AgNO₃ solution is further reduced using the reducing properties of citric acid, resulting in slow growth of the silver nanoparticles. Smaller particles in the colloid are eliminated through Ostwald ripening, resulting in a smoother surface. To enhance the effect, the silver nanoparticles are centrifuged at 10,000 rpm for 10 minutes. The supernatant is discarded, and the silver nanoparticle solution is further concentrated and stored at 4°C until further use.

[0051] The collected saliva samples were centrifuged at 14,000 rpm for 10 minutes. The supernatant was then mixed with silver nanoparticles in a 1:1 ratio, thoroughly mixed, and allowed to air dry at room temperature to collect their SERS spectra. To obtain ideal saliva SERS spectral data, 10 SERS spectra were randomly collected from each sample at the location of the "coffee ring" formed by the mixing of silver nanoparticles and saliva. This was repeated at least three times for each sample. Finally, the SERS spectrum of each saliva sample was used for subsequent data analysis.

[0052] First, the SERS spectral data were preprocessed using the BWSpec software provided with the Raman spectrometer. In order to eliminate the influence of experimental errors such as changes in laser focusing conditions on the spectrum, all preprocessed spectral data were subjected to minimum-maximum normalization. Subsequently, a differential metabolite model was established using PLS-DA. Metabolites related to disease status were screened out based on the variable importance projection value (VIP) ≥ 1 of the first principal component of PLS-DA and the P value < 0.05 in the analysis of variance (ANOVA). Finally, a CNN-Transformer hybrid model was established, combining the local feature extraction capability of CNN with the global association modeling advantages of Transformer to improve the accuracy and robustness of periodontitis severity classification.

[0053] Statistical analysis was performed using GraphPad Prism 8.0 and SPSS software. All quantitative data were expressed as x ± s, and multiple group comparisons were performed using analysis of variance. Enumeration data were expressed as percentages or constituent ratios, and differences between groups were tested using the chi-square test. P < 0.05 was considered statistically significant.

[0054] The present application firstly characterized the prepared silver nanoparticles by scanning electron microscopy (SEM) and UV-visible absorption spectroscopy. Figure 3 A) shows that the silver nanoparticles are quasi-spherical and uniform in size. UV absorption spectrum ( Figure 3 B) shows a significant absorption peak at 425nm, indicating that silver nanoparticles have good plasmon resonance properties. To further test its repeatability, 35 SERS spectra were collected under the same conditions using malachite green (MG) as the probe molecule ( Figure 3 C). The results show that the MG molecule has a peak at 1171 cm -1 The relative standard deviation (RSD) of the SERS signal intensity at 100 nm was as low as 5.17%, indicating that the silver nanoparticles have good repeatability and uniformity ( Figure 3 D).

[0055] Figure 3 A SEM image, B UV-visible absorption spectrum, C 10 -6 mol / L MG 35 times SERS spectrum, D probe MG in the relative standard deviation, E 10 -3 Conventional Raman spectra of MG solution of 10 mol / L -8 mol / L MG solution SERS spectrum, F saliva SERS spectrum

[0056] also, Figure 3 E shows the SERS performance of silver nanoparticles with an analytical enhancement factor (AEF) of 4.0×10 5 , showing a significant enhancement effect. Therefore, this application selected silver nanoparticles as the SERS substrate, collected and analyzed the SERS spectrum of saliva ( Figure 3 F) The SERS spectrum of saliva is primarily determined by the different vibrational modes of biomolecules. Based on previous research, the positions and species attributions of the main Raman peaks are summarized in Table 3 to better analyze the SERS properties of biomolecules in saliva.

[0057] Table 3 Saliva characteristic peaks attribution

[0058] <![CDATA[Raman shift (cm -1 )]]> Characteristic peak attribution 493 Glycogen 530 Cytosine,taurine 588 Ascorbic acid 631 Uric acid 725 Hypoxanthine 810 Glutathione 884 Glutathione 955 Creatine,Tryptophan,Lactic acid 1002 Phenylalanine,lactose,glucosamine 1042 Uric acid 1132 D-mannose 1202 1-tryptophan, phenylalanine 1245 Tryptophan, Thiocyanate, Uric acid 1270 Unsaturated fatty acids,phospholipid,amide-III,collagen 1325 Collagen, nuclear acid bases 1371 Inosine,uric acid 1444 Phenylalanine 1571 Phenylalanine, riboflavin, guanine, adenine 1620 Amide-I 1656 Chitin,phospholipid,amide-I,histidine

[0059] Although traditional diagnostic methods for periodontitis are widely used in clinical practice, these methods only reflect past damage to periodontal tissues and cannot accurately assess the severity of the current disease, the degree of tissue damage, and the response to treatment. Therefore, this application aims to explore the differences in SERS spectra between the control group and groups with different degrees of periodontitis severity. 480, 420, 420, and 420 salivary SERS spectra were collected for the control group and groups with different degrees of periodontitis severity ( Figure 4AD). The normalized average SERS spectra of the four groups of saliva samples were analyzed, and the shaded area represents the standard deviation (SD) of the average SERS spectra of each group. The smaller the shaded area, the better the reproducibility of the spectra between the groups ( Figure 4 E). The SERS spectral intensity differences between the control group and the mild periodontitis group, the moderate periodontitis group, and the severe periodontitis group, as well as between the mild and moderate, mild and severe, and moderate and severe periodontitis groups, were further analyzed. The results showed that the characteristic peaks at 493, 588, 631, 725, 810, 884, 955, 1042, 1132, 1325, 1444, and 1620 cm-1 were similar between the periodontitis groups of different severity and the control group, but the intensities were different ( Figure 4 F, G). These differences suggest that certain biomolecules in saliva may be altered, suggesting that potential biomarkers for monitoring the severity of periodontitis may exist in saliva samples.

[0060] Figure 4 In the middle, the AD control group, mild periodontitis, moderate periodontitis, and severe periodontitis groups were detected at 480, 420, 420, and 420 times respectively. E normalized SERS spectrum, F and G difference graphs

[0061] Although the results of the saliva SERS peak intensity analysis showed that there were differences in the biomolecular components in the saliva of different groups, some specific SERS peaks still overlapped, which limited the extraction of more potential diagnostic information from the saliva SERS spectrum. Therefore, this application aims to further explore the biomarkers and mechanisms that may be used in saliva samples to monitor the severity of periodontitis. The PLS-DA supervised statistical analysis model was used to comprehensively evaluate the inter-group variability and the variability of related variables between samples. The P value was analyzed by ANOVA to compare the significance of the mean difference between two or more groups. The results showed that at 493, 588, 631, 652, 725, 810, 884, 955, 1042, 1132, 1325, 1444, and 1620 cm -1 There are significant differences ( Figure 5 A). At the same time, calculate its VIP value ( Figure 5 B) Evaluate the importance of each variable. Using the screening criteria of VIP value ≥ 1 and P < 0.05, the 493, 588, 631, 725, 810, 884, 1132, and 1444 cm -1 Seven metabolites were detected in the saliva samples, including glycogen, ascorbic acid, uric acid, hypoxanthine, glutathione, D-mannose and phenylalanine. These differences suggest that certain biomolecules in saliva may have changed, indicating that there may be potential biomarkers in saliva samples for monitoring the severity of periodontitis. The radar chart shows ( Figure 5C) As the severity of periodontitis increases, the energy required for the reproduction of the main pathogens increases, and the 493cm -1 The glycogen content at 588cm -1 The vitamin C content gradually decreases with the severity of the disease. This decrease may lead to the inability of osteoblasts to form osteoid, thereby causing changes in the alveolar bone and other bone tissues, and aggravating the progression of periodontitis. -1 Uric acid at 810 and 884 cm is the main antioxidant in saliva, accounting for about 70% of the total antioxidant capacity. Increased uric acid consumption or decreased production may be the reason for the lower uric acid level in the periodontitis group. As the severity of periodontitis increases, the uric acid levels at 810 and 884 cm in saliva increase. -1 The glutathione level in the periodontal tissue was significantly reduced, which may weaken its antioxidant capacity and immune response, further aggravating the damage of periodontal tissue. -1 D-mannose is a natural and safe small molecule that has good anti-inflammatory and bone protection effects. It can inhibit the expression of pro-inflammatory factors and increase the level of anti-inflammatory factors by regulating the immune response of neutrophils, thereby inhibiting the inflammatory response. -1 Hypoxanthine is a purine metabolite, commonly produced during cellular metabolism and inflammatory responses. Studies have shown that oxidative stress levels are positively correlated with purine degradation metabolism, and increased purine degradation generally indicates an enhanced inflammatory response. Oxidative stress levels were elevated in the periodontitis group, and the levels of purine degradation metabolites also increased significantly, creating favorable conditions for bacterial proliferation. -1 Phenylalanine, a key product of amino acid metabolism in the body, can be converted into the alkaline compound phenylacetic acid through redox reactions. Phenylalanine metabolism can help limit the formation of dental caries by regulating the pH of oral biofilms. Periodontitis-associated bacteria require phenylalanine for growth and produce phenylacetic acid, which is closely associated with the clinical symptoms of periodontal disease. These results indicate that with increasing periodontitis severity, saliva levels of glycogen, ascorbic acid, uric acid, glutathione, and D-mannose decrease, while hypoxanthine and phenylalanine levels increase, suggesting the presence of potential saliva biomarkers for monitoring periodontitis severity. Therefore, studying periodontitis-related biomarkers and their relationship with disease severity is important for the early prevention and monitoring of periodontitis.

[0062] Figure 5 A: Comparison of SERS intensity at P < 0.05 (ANOVA, *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001), BPLS-DAVIP value, C: Radar plot of average SERS intensity.

[0063] SERS spectroscopy can identify a variety of important biomolecules in saliva samples, but relying solely on SERS characteristic peaks to monitor the severity of periodontitis still lacks specificity. Therefore, in order to more effectively monitor the severity of periodontitis in clinical applications, this application proposes a CNN-Transformer hybrid model. The architecture of the model is as follows: The specific parameters of the CNN feature extraction module are set as follows: the first convolution layer contains 32 convolution kernels, the convolution kernel size is 3×1, the stride is 1, and the ReLU activation function is used; the first pooling layer uses maximum pooling, the pooling window size is 2×1, and the stride is 2; the second convolution layer contains 64 convolution kernels, the convolution kernel size is 3×1, the stride is 1, and the ReLU activation function is used; the second pooling layer uses maximum pooling, the pooling window size is 2×1, and the stride is 2; the third convolution layer contains 128 convolution kernels, the convolution kernel size is 3×1, the stride is 1, and the ReLU activation function is used. Based on the features extracted by CNN, the Transformer encoder is introduced to further process the sequence features: in order to capture the position information of the sequence, the sine-cosine function is used to generate the position code, and the dimension is consistent with the CNN output feature dimension. Set 8 attention heads, the dimension of each attention head is 64, and the total hidden dimension is 512. Feedforward neural network: contains two linear layers, using the GELU activation function in the middle, the first layer dimension is 512, and the second layer dimension is 512. Residual connection and layer normalization: residual connection and layer normalization are applied before and after the self-attention and feedforward network modules. Number of encoder layers: set 4 Transformer encoding layers to stack. From the curve of the change in accuracy and loss value during the model training process, it can be seen that the accuracy of the training set and the verification set continues to improve with the increase in the number of iterations, and tends to stabilize after reaching a certain number of iterations ( Figure 6 B). On the contrary, the loss value continues to decrease as the number of iterations increases, and the loss values ​​of the training set and the validation set also tend to stabilize after reaching a certain number of iterations ( Figure 6 C). To evaluate the classification performance of the CNN-Transformer hybrid model for the control group, mild, moderate, and severe periodontitis groups, a confusion matrix was used as a quantitative visualization method to describe the relationship between the true distribution of sample data and the predicted results in the form of a matrix. The results showed that the prediction accuracy of the control group, mild, moderate, and severe periodontitis groups was 100%, 98%, 95%, and 100%, respectively, and the average recognition accuracy reached 98% ( Figure 6 D), indicating that the model has good classification performance. Therefore, the salivary SERS technology combined with the CNN-Transformer hybrid model has important potential application value in monitoring the severity of periodontitis.

[0064] Figure 6ACNN-Transformer hybrid model architecture, B accuracy of the control group, mild, moderate, and severe periodontitis groups during iterative training, C loss value, and D confusion matrix (expressed in percentage (%))

[0065] Monitoring the severity of periodontitis is of great significance for the control and treatment of the disease. To this end, this application uses silver nanoparticles as the SERS substrate to detect the SERS spectra of the control group and groups with different degrees of periodontitis severity, and analyzes the spectral differences between them. The results showed that the characteristic peaks of saliva metabolites in the periodontitis groups with different degrees of severity were similar to those in the control group, but there were differences in intensity. Then, the PLS-DA differential metabolite model was used to screen out differential metabolites related to the disease state from the saliva SERS spectra of the control group and groups with different degrees of periodontitis based on VIP≥1 and P<0.05, including glycogen, ascorbic acid, uric acid, hypoxanthine, glutathione, D-mannose and phenylalanine. Finally, the CNN-Transformer hybrid model was used for classification, and the results showed that the average classification accuracy of the control group and groups with different degrees of periodontitis severity reached 98%.

[0066] It should be noted that the specific embodiments are merely explanations and illustrations of the technical solutions of the present invention and cannot be used to limit the scope of protection. Any minor changes made based on the claims and description of the present invention shall still fall within the scope of protection of the present invention.

Claims

1. A periodontitis classification system based on salivary SERS, characterized by The system includes a spectral data acquisition module, a spectral data processing module and a classification module; The spectrum data acquisition module is used to acquire saliva SERS spectrum data; The spectral data processing module is used to extract differential metabolites corresponding to saliva SERS spectral data; The classification module is used to input differential metabolites into the classification network to obtain classification results; The saliva SERS spectrum data is obtained by detecting silver nanoparticles as a SERS substrate.

2. A periodontitis classification system based on salivary SERS according to claim 1, characterized in that The silver nanoparticles are obtained by the following steps: First, 2.5 mL of 0.1 M AgNO₃ solution was added to 250 mL of ultrapure water containing 0.6 μM ascorbic acid and 3 μM trisodium citrate. The solution was stirred at 30°C for 15 minutes until the color of the solution no longer changed. The solution was then reacted at 100°C for 2 hours. During the reaction, ascorbic acid was used as a reducing agent to reduce the single crystal silver core, and citrate ions were used as a stabilizer to protect the reduced single crystal silver core. When the ascorbic acid is completely consumed, the reducing property of citric acid is used to continue reducing the AgNO3 solution, causing the silver nanoparticles to grow slowly, and the remaining particles in the colloid are eliminated through Ostwald ripening, making the surface of the silver nanoparticles smoother. The silver nanoparticles are centrifuged at 10,000 rpm for 10 minutes, the supernatant is discarded, and 1 mL of ultrapure water is added to further concentrate the silver nanoparticle solution, which is then sealed and stored at 4°C for later use.

3. A periodontitis classification system based on salivary SERS according to claim 2, characterized in that The specific steps of obtaining saliva SERS spectrum data are: Step 1: Remove any remaining food debris from the mouth. The subject gently touches the roots of the upper or lower teeth with the tip of the tongue to stimulate saliva secretion in a natural, non-stimulating manner, and spits the saliva into a sterile saliva collection tube. Step 2: The collected samples were immediately frozen in a -80°C refrigerator; Step 3: Take the saliva sample out of the -80°C freezer and thaw it at room temperature; Step 4: Centrifuge the thawed saliva sample at 14,000 rpm for 10 minutes, take the supernatant and mix it with silver nanoparticles in a 1:1 ratio. After thorough mixing, let it stand at room temperature until it is naturally dry, and then collect the SERS spectrum.

4. A periodontitis classification system based on salivary SERS according to claim 3, characterized in that In step 4, 10 SERS spectra are collected randomly from the positions of the "coffee ring" formed after the silver nanoparticles and saliva are mixed.

5. The periodontitis classification system based on salivary SERS according to claim 1, characterized in that The differential metabolites included glycogen, ascorbic acid, uric acid, hypoxanthine, glutathione, D-mannose, and phenylalanine.

6. A periodontitis classification system based on salivary SERS according to claim 1, characterized in that After the spectral data acquisition module acquires the saliva SERS spectral data, it pre-processes the saliva SERS spectral data through BWSpec and performs normalization processing on the pre-processed spectral data.

7. A periodontitis classification system based on salivary SERS according to claim 1, characterized in that The spectral data processing module extracts differential metabolites corresponding to the saliva SERS spectral data through the PLS-DA model.

8. A periodontitis classification system based on salivary SERS according to claim 7, characterized in that The differential metabolites were evaluated by the variable importance projection value of the first principal component in the PLS-DA model ≥ 1. P Value < 0.

05.

9. The periodontitis classification system based on salivary SERS according to claim 1, characterized in that The classification network is a CNN model.

10. The periodontitis classification system based on salivary SERS according to claim 1, characterized in that The classification network is a CNN-Transformer hybrid model, which includes a CNN model and a Transformer encoder. The CNN model is used for local feature extraction, and the Transformer encoder extracts sequence features based on the local features extracted by the CNN model. The CNN model includes three convolutional layers and two pooling layers. The first convolutional layer contains 32 convolution kernels with a convolution kernel size of 3×1, a stride of 1, and a ReLU activation function. The first pooling layer uses maximum pooling with a pooling window size of 2×1 and a stride of 2. The second convolutional layer contains 64 convolution kernels with a convolution kernel size of 3×1, a stride of 1, and a ReLU activation function. The second pooling layer uses maximum pooling with a pooling window size of 2×1 and a stride of 2. The third convolutional layer contains 128 convolution kernels with a convolution kernel size of 3×1, a stride of 1, and a ReLU activation function. The Transformer encoder uses sine-cosine functions to generate positional encodings, the dimension of which is consistent with the CNN output feature dimension. Eight attention heads are set, the dimension of each attention head is 64, and the total hidden dimension is 512.

Images