A Raman spectroscopy data processing method and system for providing auxiliary identification information for psoriatic arthritis.
By combining matrix Raman spectroscopy acquisition with machine learning models, the problem of early identification of psoriatic arthritis has been solved, achieving non-invasive, rapid, and objective auxiliary diagnosis, and improving the accuracy and stability of identification.
Patent Information
- Application Number
- CN202611277501.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies are insufficient for accurate identification of the early stages of psoriatic arthritis through non-invasive methods. In particular, relying on visual observation or invasive pathological examinations makes it impossible to achieve dynamic monitoring at the molecular level. Furthermore, existing Raman spectroscopy studies mainly rely on infrared spectroscopy, which has a weak signal response and is difficult to effectively reflect the conformational changes of disulfide bonds.
Using matrix Raman spectroscopy acquisition technology combined with machine learning models, Raman spectral features are extracted from nail samples through multi-layer acquisition and preprocessing, and input feature vectors are generated. Models such as Elastic Net, linear discriminant analysis, random forest, XGBoost and support vector machine are used for classification, and patient-level auxiliary diagnostic scores are output.
It achieves non-invasive, rapid, and objective auxiliary identification of psoriatic arthritis, improves the representativeness and stability of spectral data, and the model evaluation results are consistent with clinical application needs, taking into account the comprehensive optimality of sensitivity and specificity.
Smart Images

Figure CN123120261A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical artificial intelligence diagnostic technology, and relates to a Raman spectroscopy data processing method and system that provides auxiliary identification information for psoriatic arthritis. Background Technology
[0002] Psoriatic arthritis (PsA) is the leading comorbidity of psoriasis (PsO), with approximately 30% of PsO patients progressing to PsA. PsA not only causes joint pain, swelling, and limited mobility, but as the disease progresses, it can lead to joint deformities and even disability, severely impacting patients' ability to work and their quality of life. The disease progression from PsO to PsA includes five stages: PsO (environmental exposure) – preclinical PsA (abnormal immune activation phase) – subclinical PsA (quiescent inflammatory phase) – PsA with prodromal symptoms (transitional phase) – PsA. Currently, the internationally accepted CASPAR classification criteria are mainly applicable to the confirmed PsA stage, making it difficult to identify and diagnose the early and prodromal stages. However, clinical evidence shows that 86% of PsA patients present with psoriatic lesions as their initial manifestation, and it takes an average of 7-8 years to progress to systemic arthritis. By the time patients meet the CASPAR diagnostic criteria, the optimal intervention window has already passed.
[0003] PsA falls under the categories of "Bai Bi" and "Bi Syndrome," with blood stasis obstructing the meridians as its core pathogenesis. The *Huangdi Neijing* states, "When Bi is in the bones, it causes heaviness; when it is in the vessels, blood congeals and does not flow; when it is in the tendons, they are flexible and cannot extend." Prolonged illness inevitably leads to blood stasis; stagnant blood combines with external pathogens, obstructing the meridians and resulting in poor circulation of Qi and blood. Both traditional Chinese medicine and Western medicine perspectives suggest a close relationship between nail changes and the occurrence and development of PsA. Traditional Chinese medicine theory considers nails as "the surplus of tendons," an important external indicator for observing the circulation of Qi and blood and the state of the meridians. Traditional Chinese medicine's holographic theory further points out that nails, as "miniature holographic elements" of bodily information, have a homologous and isomorphic relationship with the entire body's organs and meridians. 80% of PsA patients also have nail lesions. Modern medical research confirms that nail damage during the progression of PsA is considered an important early signal of inflammation "progressing" from the skin to the joints and a significant risk factor for the transformation of PsO into PsA. However, current clinical nail assessments mostly rely on visual observation or invasive pathological examinations, which are highly subjective and have low patient acceptance, and cannot achieve precise, non-invasive dynamic monitoring at the molecular level.
[0004] Raman spectroscopy is a non-destructive analytical technique that provides the chemical fingerprint of a sample by detecting molecular vibrational spectra. Raman spectroscopy offers advantages such as requiring no complex sample pretreatment, being non-invasive, and rapid. While some studies have explored the application of Raman spectroscopy in the analysis of psoriasis lesions in recent years, investigating its potential in differentiating PsO and PsA from nails, a stable biological material, remains a research gap both domestically and internationally.
[0005] Publication number CN121483562A discloses a method for early identification of PsA based on the infrared spectrum and scanning electron microscope images of nails. However, it uses infrared spectroscopy instead of Raman spectroscopy. The structural integrity of nail keratin mainly depends on disulfide bonds. Raman spectroscopy has high sensitivity to disulfide bonds, while infrared spectroscopy has a weak signal response in the detection of disulfide bonds and is difficult to effectively reflect the conformational changes of disulfide bonds.
[0006] Therefore, there is an urgent need for a Raman spectroscopy data processing method and system that can provide auxiliary identification information for psoriatic arthritis. This method would acquire comprehensive and stable Raman spectral information through matrix spectral acquisition and combine it with a machine learning model to achieve non-invasive auxiliary identification of PsA. Summary of the Invention
[0007] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a Raman spectroscopy data processing method and system that provides auxiliary identification information for psoriatic arthritis.
[0008] The objective of this invention can be achieved through the following technical solutions: A Raman spectroscopy data processing method that provides auxiliary identification information for psoriatic arthritis includes the following steps: Matrix Raman spectroscopy was performed on the nail sample to be tested to obtain Raman spectral data of the nail sample. The nail sample to be tested was the affected nail or suspected affected nail of a psoriasis patient. The nail sample was an excised nail sample. The Raman spectral data are preprocessed to obtain preprocessed Raman spectral data; Raman spectral features are extracted from the preprocessed Raman spectral data to generate an input feature vector; The input feature vector of the nail sample to be tested is input into a trained machine learning classification model to obtain the classification probability of a single spectrum; The classification probabilities of the individual spectra corresponding to multiple spectra of the same patient are aggregated to obtain a patient-level auxiliary diagnostic score; Based on the comparison results between the auxiliary diagnostic score and the preset discrimination threshold, the auxiliary identification information of psoriatic arthritis corresponding to the nail sample to be tested is output. The machine learning classification model is trained using a labeled training sample set, which includes Raman spectral data of nails from patients with clinically diagnosed psoriatic arthritis and Raman spectral data of nails from patients with clinically diagnosed non-psoriatic arthritis. The Raman spectral data in the training sample set is preprocessed and Raman spectral features are extracted to generate an input feature vector. The input feature vector of the training sample set is then associated with the corresponding clinical diagnostic labels to generate the labeled training sample set.
[0009] Furthermore, the training sample set is divided into patient units to ensure that all Raman spectra of the same patient are used only for training or validation of the machine learning classification model.
[0010] Furthermore, the clinical diagnostic label includes a positive category and a negative category, wherein the positive category is for patients with clinically diagnosed psoriatic arthritis and the negative category is for patients with clinically diagnosed non-arthritic psoriasis.
[0011] Furthermore, the matrix Raman spectroscopy acquisition includes: acquiring nail samples using a Raman spectrometer, with the acquisition range including the fingerprint area and the high wavenumber area; acquiring at the acquisition positions along the length and width directions of the nail coronal surface to form an acquisition matrix, with the acquisition area spanning the dorsal, intermediate, and ventral layers of the nail plate to obtain multiple Raman spectra.
[0012] Furthermore, the preprocessing includes one or more of the following: cosmic ray removal, wavenumber range truncation, wavenumber axis correction, interpolation alignment, baseline correction, smoothing, normalization, and outlier spectrum removal.
[0013] Furthermore, the Raman spectral features include the normalized spectral intensity corresponding to each Raman shift point, the intensity ratio between different Raman peaks, principal component features, and discriminant Raman shift features.
[0014] Furthermore, the trained machine learning classification model includes one of the following: Elastic Net logistic regression model, linear discriminant analysis model, random forest model, XGBoost model, support vector machine model, and neural network model.
[0015] Furthermore, the aggregation adopts the 90th percentile aggregation method.
[0016] Furthermore, the preset discrimination threshold is determined by the ROC curve of the training sample set, and one of the following criteria is used as the threshold selection criteria: maximizing the Youden index, prioritizing sensitivity, prioritizing specificity, maximizing accuracy, or a clinically preset threshold.
[0017] Furthermore, a system for processing Raman spectroscopy data to provide auxiliary identification information for psoriatic arthritis includes: The Raman spectroscopy acquisition module is used to perform matrix Raman spectroscopy acquisition on the nail sample to be tested, and to obtain the Raman spectral data of the nail sample to be tested; The preprocessing module is used to preprocess the Raman spectral data to obtain preprocessed Raman spectral data; The feature extraction module is used to extract Raman spectral features from the preprocessed Raman spectral data and generate an input feature vector; The classification prediction module stores a trained machine learning classification model, which is used to input the input feature vector into the classification model to obtain an auxiliary diagnostic score; The result output module is used to output auxiliary identification information of psoriatic arthritis corresponding to the nail sample to be tested based on the comparison result of the auxiliary diagnostic score and the preset discrimination threshold.
[0018] Compared with the prior art, the present invention has the following advantages: (1) This invention establishes a complete data processing flow from Raman spectroscopy acquisition, preprocessing, feature extraction to machine learning classification of nail samples, realizes the output of auxiliary identification information for psoriatic arthritis, solves the technical problem of how to extract effective information from Raman spectroscopy data and transform it into auxiliary identification information, and realizes non-invasive, rapid and objective auxiliary identification of psoriatic arthritis.
[0019] (2) This invention performs matrix Raman spectroscopy on the dorsal, intermediate, and ventral layers of the nail plate. Existing technologies mostly use single-point or limited-point acquisition, which makes it difficult to fully reflect the differences in molecular structure among different layers of the nail. This invention achieves comprehensive characterization of the chemical information of the entire nail layer through matrix acquisition, improving the representativeness and stability of the spectral data.
[0020] (3) In the model training and validation process, the present invention divides the data by patient, ensuring that all Raman spectra of the same patient are included in only one of the training set or the validation set, thus avoiding the inflated model performance caused by data leakage. In the model output stage, the classification probabilities of multiple spectra of the same patient are aggregated into a patient-level score by the 90th percentile, so that the model evaluation results are consistent with the "patient-level diagnosis" requirements in clinical applications, taking into account the comprehensive optimization of sensitivity and specificity. Attached Figure Description
[0021] Figure 1 A flowchart of a Raman spectroscopy data processing method for providing auxiliary identification information for psoriatic arthritis in this invention; Figure 2 This is a schematic diagram illustrating the principle of the present invention; Figure 3 A schematic diagram of the Raman spectroscopy data processing system that provides auxiliary identification information for psoriatic arthritis for this invention; Figure 4 The average Raman spectra and spectral differences of the fingernails of patients in the PsO and PsA groups; Figure 5To compare the performance of five machine learning models—Elastic Net, LDA, Random Forest, XGBoost, and SVM—on the training set and the external validation set, Figure A shows the ROC of the training set, Figure B shows the ROC of the external validation set, Figures C and D show the calibration curves of the training and validation sets, and Figures E and F show the comprehensive metrics of the training and validation sets, including AUC, Accuracy, Sensitivity, Specificity, PPV, NPV, and F1. Figure 6 The following are the SHAP interpretation results for the XGBoost model. A shows the importance of each Raman shift to the model's discrimination. B further shows the relationship between the value of each feature and the model's discrimination direction. C shows the local interpretation result of a sample with the highest PsA probability, which is determined by the model to be p(PsA) = 0.947. D shows the relationship between the key wavenumber intensity change and the SHAP value. Detailed Implementation
[0022] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.
[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.
[0024] In this application, the technical features described in an open-ended manner include both closed technical solutions consisting of the listed features and open technical solutions that include the listed features.
[0025] Example 1 This embodiment provides a Raman spectroscopy data processing method to provide auxiliary identification information for psoriatic arthritis, used to assist in identifying whether a psoriasis patient has psoriatic arthritis. The overall process is as follows: Figure 1 and Figure 2 As shown.
[0026] S1: Matrix Raman spectroscopy is performed on the nail sample to be tested to obtain Raman spectral data. The nail sample is the affected or suspected affected nail of a psoriasis patient, and the nail sample is an excised nail sample. In this embodiment, the Raman spectroscopy acquisition range includes 400-1800 cm⁻¹. -1 Fingerprint area and / or 2800-3200 cm -1In the high wavenumber region, 15 points were collected along the length of the nail coronal plane and 12 points along the width, forming a 15×12 acquisition matrix. The acquisition area covered the dorsal, intermediate, and ventral layers of the nail plate, and a total of 180 Raman spectra were obtained.
[0027] S2: Preprocess the Raman spectral data to obtain preprocessed Raman spectral data. Specifically, the original Raman spectra are subjected to cosmic ray removal, wavenumber truncation, wavenumber alignment, baseline correction, smoothing, normalization, and quality control in sequence to obtain standardized Raman spectra. The average Raman spectra and difference spectra of the two groups are shown below. Figure 4 As shown. In this embodiment, the normalization process uses area normalization. Through the above preprocessing, the impact of spike noise, fluorescence background, instrument drift, overall signal intensity differences, and abnormal spectra on model training and discrimination results can be reduced.
[0028] S3: Extract Raman spectral features from the preprocessed Raman spectral data to generate an input feature vector. The preprocessed Raman spectrum is used to extract Raman spectral features of the fingernail. The extracted fingernail Raman spectral features may include the normalized spectral intensity corresponding to each Raman shift point, the intensity ratio between different Raman peaks, principal component features, and discriminant Raman shift features automatically selected by a machine learning model. A spectral feature matrix is then constructed, and the flattened spectral feature matrix is used as the input feature vector. In this embodiment, the preset wavenumber range is 400-1800 cm⁻¹. -1 and 2800-3200 cm -1 .
[0029] S4: Input the feature vector of the nail sample to be tested into the trained machine learning classification model to obtain the classification probability of a single spectrum. Specifically, the trained machine learning classification model is selected from any of the following models: Elastic Net logistic regression model, Linear Discriminant Analysis (LDA) model, Random Forest (RF) model, XGBoost model, and Support Vector Machine (SVM) model. Each Raman spectrum of the nail sample to be tested is preprocessed and feature extracted, and then input into the model. The model outputs the corresponding classification probability value, which represents the predicted probability that the spectrum belongs to the psoriatic arthritis type (PsA) category. If multiple Raman spectra are collected from the nail sample to be tested, each spectrum is input into the model separately to obtain multiple classification probability values, which are used for subsequent patient-level aggregation.
[0030] S5: Aggregate the classification probabilities of individual spectra corresponding to multiple spectra of the same patient to obtain a patient-level auxiliary diagnostic score. In this embodiment, the 90th percentile aggregation method is used to calculate the patient-level auxiliary diagnostic score.
[0031] S6: Based on the comparison between the auxiliary diagnostic score and the preset discrimination threshold, output the auxiliary identification information for psoriatic arthritis corresponding to the nail sample to be tested. The patient-level auxiliary diagnostic score is compared with the discrimination threshold determined by the training sample set. If the score is greater than or equal to the threshold, an auxiliary identification result favoring PsA is output; if the score is less than the threshold, an auxiliary identification result favoring PsO is output. The preset discrimination threshold is determined by the ROC curve of the training sample set, using one of the following criteria: maximizing the Youden exponent, prioritizing sensitivity, prioritizing specificity, maximizing accuracy, or a clinically preset threshold. In this embodiment, the Youden exponent maximization principle is adopted, that is, the probability threshold that optimizes both sensitivity and specificity is selected as the discrimination threshold. The specific process is as follows: Step 1: The model outputs the PsA classification probability for each Raman spectrum in the training queue; Step 2: The probabilities of multiple spectra for the same patient are aggregated, and the 90th percentile is taken; Step 3: The patient-level auxiliary diagnostic score for each training patient is obtained; Step 4: ROC curves are plotted based on the real clinical diagnostic labels PsO / PsA of the training patients; Step 5: All possible thresholds are iterated, and sensitivity, specificity, and Youden index are calculated; Step 6: The threshold with the largest Youden index is selected as the preset discrimination threshold; Step 7: After this threshold is fixed, it is directly used for the external validation queue and future test patients. In this embodiment, the thresholds for the six algorithms Elastic Net, LDA, Random Forest, XGBoost, SVM, and 1D-CNN are 0.59, 0.42, 0.61, 0.72, 0.68, and 0.58, respectively.
[0032] During the training of the machine learning classification model, the training data came from patients with PsO and PsA who visited the dermatology outpatient clinic and ward of a certain hospital, and were grouped according to the following criteria: (1) PsO group: those who meet the clinical or pathological diagnostic criteria for psoriasis and currently have no clear musculoskeletal symptoms, no inflammatory signs such as arthritis, enthesitis, or digititis, and whose musculoskeletal ultrasound or MRI does not show any inflammatory abnormalities of PsA.
[0033] (2) PsA group: those diagnosed with PsA by dermatologists based on a comprehensive assessment of medical history, clinical manifestations, physical examination, laboratory tests and imaging results, and the CASPAR classification criteria are used as the supporting classification basis.
[0034] Inclusion criteria: (1) Previous or first diagnosis of psoriasis; (2) Age 18-80 years old, gender not limited; (3) Those who sign the informed consent form.
[0035] Exclusion criteria (1) Individuals with severe mental illness or cognitive impairment; (2) Individuals with severe systemic diseases; (3) History of malignant tumors, primary or secondary immunodeficiency; (4) Those with severe nail trauma, nail fungal infection, or obvious nail plate damage that affects Raman spectroscopy detection; (5) Those who have recently received local treatment that may significantly affect the nail structure.
[0036] (6) Individuals whom the researchers deem unsuitable for enrollment.
[0037] In this embodiment, two groups of psoriasis patients were included as modeling samples. The first group consisted of patients with non-arthritic psoriasis (PsO), as the negative category; the second group consisted of patients with psoriatic arthritis (PsA), as the positive category. In this embodiment, 200 PsO patients and 100 PsA patients were collected. Ex vivo nail samples were obtained from all patients, including affected nails and suspected affected nails, and corresponding clinical diagnostic labels (independent diagnosis by two dermatologists), demographic and disease data, including age, sex, psoriasis course, and classification, were recorded. Raman spectral data of the nail samples were acquired using a Witec confocal micro-Raman spectrometer (532 nm laser). Matrix spectral acquisition was performed on each nail sample, according to 15... A Raman spectroscopy scan was performed using a 12-point setup to ensure that the Raman spectral data for each sample covered the ventral, intermediate, and posterior layers. The spectral acquisition range included 400-3200 cm⁻¹. -1 Ultimately, each patient yielded 180 Raman spectra of their fingernails. The raw Raman spectra were then subjected to cosmic ray removal, wavenumber cutoff, wavenumber alignment, baseline correction, smoothing, normalization, and quality control to obtain standardized Raman spectra. The average Raman spectra and difference spectra of the two groups are shown below. Figure 4As shown. During model training, the modeling samples are divided into training and validation sets based on patients, ensuring that all Raman spectra of the same patient enter only one set, avoiding information leakage caused by different Raman spectra of the same patient entering both the training and validation sets simultaneously. The training set is used for model training, and the validation set is used for internal validation and model selection, employing repeated five-fold cross-validation for internal validation. Model performance is evaluated using metrics such as area under the ROC curve, accuracy, sensitivity, specificity, and decision curve analysis. The model outputs the classification probability of each Raman spectrum belonging to the PsA category. In this embodiment, five models—Elastic Net, LDA, Random Forest, XGBoost, and SVM—are used for training. For the linear discriminant analysis model, principal component analysis is first performed on the spectral feature matrix of the training sample set, and the principal components are selected for LDA model training. The other four models directly use the original spectral feature matrix as input.
[0038] After model training, to evaluate the PsO / PsA-assisted differential diagnosis capability based on fingernail Raman spectroscopy, this study constructed and compared five machine learning models: Elastic Net, LDA, Random Forest, XGBoost, and SVM. Figure 5 In the training set, all five machine learning models (A, C, E) demonstrated some PsO / PsA discriminative ability. RandomForest showed the highest training set fitting performance, with an AUC close to 1.000 and near-optimal overall performance, suggesting that this model can effectively capture the discriminative spectral features in the training data. However, the excessively high training set performance of RandomForest suggests a potential overfitting tendency. ElasticNet, SVM, XGBoost, and LDA also showed superior discriminative ability compared to random classification in the training set. The training set calibration curves showed some deviation between the predicted probabilities of different models and the actual observed proportions, indicating that the model probability output still needs further calibration. Internal validation results are shown in Table 1. All machine learning models possessed some classification ability. RandomForest and XGBoost performed best, with AUCs of 0.792 and 0.781, respectively. The SVM model had an internal validation AUC of 0.723, an accuracy of 64.6%, a sensitivity of 73.6%, and a specificity of 59.5%.
[0039] Table 1 Performance evaluation of different machine learning models in the internal validation queue Further SHAP analysis or model coefficient analysis is used to identify Raman shifts that contribute significantly to the model's discrimination. For example... Figure 6 The results show that the key features of the model are concentrated at 1009 cm. -1648 cm -1 637 cm -1 1598-1646 cm -1 1474-1478 cm -1 821-848 cm -1 The wavelength range indicates that the differences between PsO and PsA mainly involve changes in keratin structure, sulfur / disulfide bonds, aromatic amino acids, and protein secondary structure.
[0040] In the external validation phase, a total of 90 patients were included, including 60 with PsO and 30 with PsA. Inclusion and exclusion criteria were consistent with those in the modeling phase. Diagnosis was made independently by two dermatologists based on patient history, physical examination, laboratory tests, and imaging data. Based on ex vivo nail samples and Raman spectra, after preprocessing, model classification, and patient-level aggregation, an auxiliary discrimination result indicating whether the patient currently leans towards PsO or PsA was obtained. The final clinical diagnosis was used as a reference standard to calculate AUC, accuracy, sensitivity, and specificity. Figure 5 As shown in B, D, F and Table 2, in the external validation cohort, the SVM model exhibited good diagnostic performance, with an area under the receiver operating characteristic (AUC) of 0.876, accuracy of 81.1%, sensitivity of 86.7%, specificity of 78.3%, positive predictive value (PPV) of 66.7%, negative predictive value (NPV) of 92.2%, and F1 score of 75.4%. The ROC curves of all models were significantly higher than the random classification line, demonstrating good generalization and discrimination ability. SVM, Random Forest, Elastic Net, and XGBoost performed similarly in the external validation set, with AUCs of approximately 0.86-0.88, while LDA had a slightly lower AUC but still exceeded 0.80. Overall performance metrics show that different models have their own advantages in terms of sensitivity, specificity, PPV, NPV, and F1 score. SVM and Elastic Net are more inclined to improve PsA recognition ability, while Random Forest and XGBoost have certain advantages in specificity. The calibration curves on the external validation set show some fluctuations, suggesting that the models are suitable as auxiliary diagnostic tools for PsO / PsA.
[0041] Table 2 Performance evaluation of different machine learning models in the external validation queue To further evaluate the consistency between the method of this invention and clinical diagnostic results, the model-assisted discrimination results were compared with the diagnostic results made by dermatologists based on clinical data. As shown in Table 3, the overall consistency rate between the SVM model prediction results and the clinician's diagnostic results reached 81.1%, and the Cohen's kappa consistency coefficient was 0.605, indicating that the method of this invention has good consistency with the clinician's diagnosis and can be used as a PsA auxiliary identification tool.
[0042] Table 3 shows the consistency between the model and the doctor's diagnosis. If the above methods are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0043] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0044] Example 2 See Figure 3 This embodiment provides a Raman spectroscopy data processing system for assisting in the identification of psoriatic arthritis, including: a Raman spectroscopy acquisition module, a preprocessing module, a feature extraction module, a classification prediction module, and a result output module. The Raman spectroscopy acquisition module is used to acquire Raman spectral data of the nail sample to be tested. The Raman spectral data is acquired along a 15×12 matrix along the coronal plane of the nail, spanning the dorsal / intermediate / ventral layers, totaling 180 spectra. The preprocessing module is used to preprocess the Raman spectral data to obtain preprocessed Raman spectral data. The feature extraction module is used to extract 400-1800 cm⁻¹ features from the preprocessed Raman spectral data. -1and 2800-3200 cm -1 The system uses standardized spectral intensities within a given range to form a spectral feature matrix, and further selects the top 30 principal components as input feature vectors through principal component analysis. A classification prediction module stores a machine learning classification model trained on a PsA / PsO training sample set, used to input the input feature vectors into the model to obtain an auxiliary diagnostic score. A result output module outputs auxiliary identification information for psoriatic arthritis based on a comparison between the auxiliary diagnostic score and a preset discrimination threshold. These modules are connected sequentially to constitute the system described in this embodiment.
[0045] The rest is the same as in Example 1.
[0046] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.
Claims
1. A Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis, characterized in that, Includes the following steps: Matrix Raman spectroscopy was performed on the nail sample to be tested to obtain Raman spectral data of the nail sample. The nail sample to be tested was the affected nail or suspected affected nail of a psoriasis patient. The nail sample was an excised nail sample. The Raman spectral data are preprocessed to obtain preprocessed Raman spectral data; Raman spectral features are extracted from the preprocessed Raman spectral data to generate an input feature vector; The input feature vector of the nail sample to be tested is input into a trained machine learning classification model to obtain the classification probability of a single spectrum; The classification probabilities of the individual spectra corresponding to multiple spectra of the same patient are aggregated to obtain a patient-level auxiliary diagnostic score; Based on the comparison results between the auxiliary diagnostic score and the preset discrimination threshold, the auxiliary identification information of psoriatic arthritis corresponding to the nail sample to be tested is output. The machine learning classification model is trained using a labeled training sample set, which includes Raman spectral data of nails from patients with clinically diagnosed psoriatic arthritis and Raman spectral data of nails from patients with clinically diagnosed non-psoriatic arthritis. The Raman spectral data in the training sample set is preprocessed and Raman spectral features are extracted to generate an input feature vector. The input feature vector of the training sample set is then associated with the corresponding clinical diagnostic labels to generate the labeled training sample set.
2. The Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis according to claim 1, characterized in that, The training sample set is divided by patient to ensure that all Raman spectra of the same patient are used only for training or validation of the machine learning classification model.
3. The Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis according to claim 1, characterized in that, The clinical diagnostic label includes a positive category and a negative category. The positive category is for patients with clinically diagnosed psoriatic arthritis, and the negative category is for patients with clinically diagnosed non-arthritic psoriasis.
4. The Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis according to claim 1, characterized in that, The matrix Raman spectroscopy acquisition includes: acquiring nail samples using a Raman spectrometer, with the acquisition range including the fingerprint area and the high wavenumber area; acquiring samples along the length and width directions of the nail coronal plane to form an acquisition matrix, with the acquisition area spanning the dorsal, intermediate, and ventral layers of the nail plate, thereby obtaining multiple Raman spectra.
5. The Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis according to claim 1, characterized in that, The preprocessing includes one or more of the following: cosmic ray removal, wavenumber range truncation, wavenumber axis correction, interpolation alignment, baseline correction, smoothing, normalization, and outlier spectrum removal.
6. The Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis according to claim 1, characterized in that, The Raman spectral features include the normalized spectral intensity corresponding to each Raman shift point, the intensity ratio between different Raman peaks, principal component features, and discriminant Raman shift features.
7. The Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis according to claim 1, characterized in that, The trained machine learning classification model includes one of the following: Elastic Net logistic regression model, linear discriminant analysis model, random forest model, XGBoost model, support vector machine model, and neural network model.
8. The Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis according to claim 1, characterized in that, The aggregation method used is the 90th percentile aggregation method.
9. The Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis according to claim 1, characterized in that, The preset discrimination threshold is determined by the ROC curve of the training sample set, and one of the following criteria is used as the threshold selection criteria: maximizing the Youden index, prioritizing sensitivity, prioritizing specificity, maximizing accuracy, or a clinically preset threshold.
10. A system for implementing the Raman spectral data processing method for providing auxiliary identification information for psoriatic arthritis as described in any one of claims 1-9, characterized in that, include: The Raman spectroscopy acquisition module is used to perform matrix Raman spectroscopy acquisition on the nail sample to be tested, and to obtain the Raman spectral data of the nail sample to be tested; The preprocessing module is used to preprocess the Raman spectral data to obtain preprocessed Raman spectral data; The feature extraction module is used to extract Raman spectral features from the preprocessed Raman spectral data and generate an input feature vector; The classification prediction module stores a trained machine learning classification model, which is used to input the input feature vector into the classification model to obtain an auxiliary diagnostic score; The result output module is used to output auxiliary identification information of psoriatic arthritis corresponding to the nail sample to be tested based on the comparison result of the auxiliary diagnostic score and the preset discrimination threshold.
Citation Information
Patent Citations
Method and system for constructing intelligent typing diagnosis model of psoriasis
CN121483562A
