System and method for prospectively predicting properties of thyroid nodules

Abstract

The invention provides a system and a method for prospectively predicting the property of a thyroid nodule. The method comprises the following steps: an acquisition module is used for acquiring a tissue biopsy sample; the quantitative polymerase chain reaction detection module is used for detecting BRAF V600E information; the MRM MS detection module is used for detecting quantitative information of the peptide fragment; and the evaluation module is used for evaluating the property of the thyroid nodule by using a model for predicting the property of the thyroid nodule by taking the quantitative information of the final characteristic peptide fragment, the BRAF V600E information and the gender and age of the target subject as characteristic information. BRAF V600E mutation and the content of various polypeptides in a thyroid puncture sample are detected, the optimal feature combination is selected by using various machine learning means in combination with clinical information of gender, age and nodule size under ultrasound, a prediction model is established, the properties of thyroid nodules are evaluated and classified, clinicians are assisted in discriminating the thyroid nodules, and the accuracy of thyroid nodule identification is improved. Therefore, the problem of inaccurate diagnosis in clinic is relieved.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, specifically to a system and method for prospectively predicting the nature of thyroid nodules. Background Technology

[0002] Thyroid nodules are a very common endocrine system disorder. With advancements in ultrasound technology and the widespread availability of health checkups, an increasing number of thyroid nodules are being detected, leading to a rapid rise in their incidence rate year by year. In a survey of healthy individuals undergoing health checkups in China, the detection rate of thyroid nodules was 36% of the total population examined. Nevertheless, the vast majority of nodules are benign and usually do not require intervention.

[0003] In clinical practice, thyroid nodules are comprehensively evaluated preoperatively using hematological, ultrasound, and fine-needle aspiration (FNA) methods. Ultrasound-guided fine-needle aspiration (FNA) is considered the most reliable method for preoperative assessment. However, approximately 10-30% of thyroid nodules cannot be accurately assessed through cytopathology and are defined as indeterminate thyroid nodules (ITNs), requiring postoperative histopathological diagnosis. Furthermore, the cytopathological assessment of the nature of thyroid nodules relies on subjective scoring based on the pathologist's microscopic morphological observation; the accuracy and reliability of the results are highly dependent on the pathologist's experience, and the accuracy of assessments varies between hospitals of different levels.

[0004] Because preoperative assessment is highly subjective, physicians with varying experience levels often have vastly different clinical management approaches for ITN. Most conservative physicians would recommend surgical removal; however, after thyroidectomy, lifelong thyroid hormone replacement therapy is required, impacting the patient's quality of life. Therefore, there is an urgent need to establish an accurate, objective, non-invasive, and efficient assessment method to aid in predicting the nature of thyroid nodules and thus guide clinical diagnosis and treatment.

[0005] The foregoing background information is intended to help those skilled in the art understand prior art that is similar to the present invention, and to facilitate the understanding of the inventive concept and technical solution of the present invention. It should be clearly stated that, in the absence of clear evidence that the above content was disclosed before the filing date of this patent application, the foregoing background information should not be used to evaluate the novelty of the technical solution of this application. Summary of the Invention

[0006] Technical issues To address the aforementioned issues, the present invention aims to provide a system and method for prospectively predicting the nature of thyroid nodules. This system detects BRAF V600E mutations and the content of various peptides in thyroid biopsy samples, combines clinical information such as gender, age, and nodule size under ultrasound, utilizes multiple machine learning techniques to select the optimal feature combination, establishes a predictive model, and evaluates and classifies the nature of thyroid nodules. This assists clinicians in identifying thyroid nodules, thereby alleviating the problem of inaccurate diagnosis in clinical practice.

[0007] Technical solution That is, the present invention includes the following solutions.

[0008] Option 1: A method for using AI algorithms to screen detection targets for assessing the nature of thyroid nodules, the method comprising: Obtain the training dataset, each training dataset containing the subject's peptide quantification information, BRAF V600E information, and whether the thyroid nodule is malignant; Based on the training set, the first processor uses the peptide quantitative information of different peptide combinations as test feature information to construct and train corresponding test prediction models for the property assessment of thyroid nodules based on the first AI algorithm, compares the performance of each trained test prediction model, and selects the first group of candidate prediction models that meet the performance requirements. Based on the training set, the first processor uses the peptide quantitative information of different peptide combinations as test feature information to construct and train corresponding test prediction models based on at least two derivative algorithms of the second AI algorithm, compares the performance of each trained test prediction model, and selects a second group of candidate prediction models that meet the performance requirements. The first processor selects the quantitative information of the 15 most frequently occurring peptides from the test feature information used in the first group of candidate prediction models and the second group of candidate prediction models, which are used as detection targets for the nature assessment of thyroid nodules.

[0009] Preferably, the polypeptide quantification information is detected through the following process: Fine-needle aspiration (FNA) tissue biopsy samples were collected from the subjects' thyroid glands. The tissue biopsy samples were pretreated using pressure cycling technology; The quantitative information of each peptide was determined by liquid chromatography and multiple reaction ion monitoring-mass spectrometry (MRMMS) combined with the AQUA standard peptide quantification process for pre-processed tissue biopsy samples.

[0010] Preferably, the 15 peptides are as follows: ELSDFISYLQR (SEQ ID NO.1), FSMVVQDGIVK (SEQ ID NO.2), GESGPSGPAGPTGAR (SEQ ID NO.3), GETGPSGPVGPAGAVGPR (SEQ ID NO.4), GLGTDEDTLIEILASR (SEQ ID NO.5), GQTLVVQFTVK (SEQ ID NO.6), IYLYTLNDNAR (SEQ ID NO.7), LFDQAFGLPR (SEQ ID NO.8), LGPLVEQGR (SEQ ID NO.9), LMEDLDR (SEQ ID NO.10), LQDAEIAR (SEQ ID NO.11), MQQQLDEYQELLDIK (SEQ ID NO.12), SFVLNLGK (SEQ ID NO.13), SGVYQHVTGEMMGGHAIR (SEQ ID NO.14), TVAAPSVFIFPPSDEQLK (SEQ ID NO.15).

[0011] Preferably, the first AI algorithm includes the LASSO algorithm.

[0012] Preferably, the second AI algorithm includes the XGBoost algorithm.

[0013] Preferably, the derivative algorithms of the second AI algorithm include: a recursive feature elimination algorithm (RFE) based on XGBoost and its model importance, a recursive feature elimination algorithm (RFE) based on XGBoost and its Shapley value, and a Boruta algorithm based on XGBoost and its Shapley value.

[0014] Preferably, the performance of each test prediction model includes accuracy and AUC.

[0015] Preferably, after the model selects 15 peptides, based on the consideration of operational complexity and industrialization cost, at least 3, preferably 3-10, and more preferably 3-6 peptides are selected as the final feature peptides. The final features, together with BRAF V600E information, serve as the microbiological feature information of the actual prediction model.

[0016] Preferably, the method further includes: Based on the training set, the second processor uses the quantitative information of the final feature peptide, along with BRAF V600E information, the subject's gender and age as feature information, to construct and train an actual prediction model based on the second AI algorithm. The second processor uses the validation set to perform cross-validation on the trained actual prediction model to determine the final actual prediction model.

[0017] More preferably, the construction and training of the actual prediction model based on the second AI algorithm specifically includes: using a grid search strategy to fine-tune five hyperparameters, including the learning rate (learning_rate), the random sampling ratio (subsample) for each tree, the proportion of columns randomly sampled for each tree (colsample_bytree), the proportion of columns sampled for each node split of each tree (colsample_bylevel), and the balancing parameter (scale_pos_weight).

[0018] Preferably, the first processor and the second processor are integrated into a single processor, or they are separate, independent processors.

[0019] Option 2, a method for constructing a model to predict the nature of thyroid nodules, includes: Obtain training datasets and internal validation sets. Each training or validation dataset contains quantitative information on peptides, BRAF V600E information, and whether the thyroid nodules are malignant. Based on the training set, the second processor uses the quantitative information of the 15 peptides obtained from Scheme 1, along with BRAFV600E information, the subject's gender and age as feature information, to construct and train an actual prediction model based on the second AI algorithm. The second processor uses the internal validation set to perform cross-validation on the trained actual prediction model to determine the final actual prediction model.

[0020] Preferably, the second AI algorithm includes the XGBoost algorithm.

[0021] Preferably, the construction and training of the actual prediction model based on the second AI algorithm specifically includes: using a grid search strategy to fine-tune five hyperparameters, including the learning rate (learning_rate), the random sampling ratio (subsample) for each tree, the proportion of columns randomly sampled for each tree (colsample_bytree), the proportion of columns sampled for each node split of each tree (colsample_bylevel), and the balancing parameter (scale_pos_weight).

[0022] BRAF V600E is the most common mutation in malignant thyroid nodules, accounting for over 60% of all malignant nodules. This gene mutation exhibits excellent specificity for papillary thyroid carcinoma, exceeding 95%, but its sensitivity is relatively low at 87%. However, overall, for indeterminate thyroid nodules, such as but not limited to Bethesda nodules in categories III to V, the sensitivity of BRAF V600E mutation analysis is extremely low, only about 40%, which is insufficient for industrial applications. Furthermore, the role of the BRAF V600E wild-type gene is extremely weak compared to other follicular thyroid tumors. Therefore, in addition to BRAF V600E mutations, other molecular levels are needed for auxiliary analysis. Thus, using gene mutations in preoperative samples to determine the benign or malignant nature of thyroid nodules is not entirely accurate, as not all tumors possess mutation sites; only 60-70% of thyroid nodules have mutation sites. Proteins are the executors and manifestations of biological activities, directly constituting biological phenotypes. Therefore, detecting proteins or protein products is particularly valuable, especially for thyroid nodules with wild-type gene phenotypes. The inventors unexpectedly discovered that quantitative detection of peptides from thyroid biopsy samples, combined with basic clinical information and BRAF V600E gene mutation information, and comprehensive scoring using a machine learning model, outperforms currently available clinical information in assessing thyroid nodules. This scoring demonstrates excellent accuracy, sensitivity, and specificity, enabling precise prediction of the nature of clinical thyroid nodules and assisting clinicians in disease assessment.

[0023] Option 3 is a model for predicting the nature of thyroid nodules obtained by the construction method described in Option 2.

[0024] Option 4, a system for evaluating the nature of thyroid nodules using AI algorithms, includes: The acquisition module is configured to acquire fine-needle aspiration (FNA) tissue biopsy samples of the thyroid gland of the target subject. A quantitative polymerase chain reaction (PCR) detection module configured to detect BRAF V600E information in tissue biopsy samples; The MRM MS detection module is configured to detect quantitative information of at least 3 of the following 15 peptides as final characteristic peptides in tissue biopsy samples: ELSDFISYLQR, FSMVVQDGIVK, GESGPSGPAGPTGAR, GETGPSGPVGPAGAVGPR, GLGTDEDTLIEILASR, GQTLVVQFTVK, IYLYTLNDNAR, LFDQAFGLPR, LGPLVEQGR, LMEDLDR, LQDAEIAR, MQQQLDEYQELLDIK, SFVLNLGK, SGVYQHVTGEMMGGHAIR, and TVAAPSVFIFPPSDEQLK. The evaluation module includes a third processor configured to evaluate the nature of the thyroid nodules using a model that predicts the nature of the thyroid nodules, taking quantitative information of the final characteristic peptide along with BRAF V600E information, the gender and age of the target subject as characteristic information.

[0025] Preferably, the assessment of the nature of the thyroid nodule is intended to characterize the malignancy risk of the thyroid nodule.

[0026] Preferably, the evaluation module is configured as follows: Obtain personally identifiable information of the target subjects, including their gender and age; The device is communicatively connected to the quantitative polymerase chain reaction (PCR) detection module and the MRM MS detection module to obtain BRAF V600E information from each subject in the PCR detection module and quantitative information of the final characteristic peptides from each subject in the MRM MS detection module. From the BRAF V600E information and quantitative information of the final characteristic peptide obtained from each subject, BRAF V600E information and quantitative information of the final characteristic peptide that match the personal identification information of the target subject are identified in order to assess the nature of thyroid nodules.

[0027] Preferably, the model for predicting the nature of thyroid nodules is obtained through the following process: The third processor, based on the training set, where each piece of training data contains quantitative information on the subject's peptides, BRAF V600E information, and whether the thyroid nodule is malignant, uses the absolute quantitative information of the final characteristic peptide, along with BRAF V600E information, the subject's gender and age as feature information, to build and train the actual prediction model based on the second AI algorithm. The third processor uses an internal validation set to cross-validate the trained actual prediction model to determine the final actual prediction model.

[0028] Preferably, the second processor and the third processor are integrated into a single processor; or The third processor is independent of the second processor and is configured to load the final actual prediction model when the nature of thyroid nodules is to be assessed.

[0029] Option 5, the use of the system described in Option 4 above in the preparation of a device for assessing the properties of thyroid nodules in subjects based on targeted detection peptides and AI.

[0030] Option 6: Use of reagents for detecting targeted detection peptides in the preparation of evaluation products, wherein the evaluation products are used to assess the properties of thyroid nodules in subjects based on targeted detection peptides and AI, wherein the targeted detection peptides are selected from at least 3 of the following 15 peptides: ELSDFISYLQR, FSMVVQDGIVK, GESGPSGPAGPTGAR, GETGPSGPVGPAGAVGPR, GLGDDEDTLIEILASR, GQTLVVQFTVK, IYLYTLNDNAR, LFDQAFGLPR, LGPLVEQGR, LMEDLDR, LQDAEIAR, MQQQLDEYQELLDIK, SFVLNLGK, SGVYQHVTGEMMGGHAIR, and TVAAPSVFIFPPSDEQLK.

[0031] Preferably, the property assessment specifically includes: Provide the subject with a fine-needle aspiration (FNA) tissue biopsy sample of the thyroid gland; The tissue biopsy samples were pretreated using pressure cycling technology; Quantitative information on targeted peptides in tissue biopsy samples after pretreatment; The quantitative information of the detected targeted peptides is input into the aforementioned model for predicting the nature of thyroid nodules; and Output the assessment results, which reflect the malignancy risk of the thyroid nodules.

[0032] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined to obtain specific implementation methods.

[0033] Beneficial effects According to the present invention, polypeptides can be quantitatively detected from trace amounts of thyroid biopsy samples, and basic clinical information and BRAF V600E gene mutation information can be integrated. By inputting the information obtained above into the custom thyroid nodule assessment model of the present invention, a comprehensive score is performed. This score is superior to the assessment of thyroid nodules using currently available clinical information (such as BRAF V600E, single prediction). Finally, the clinical nature of thyroid nodules is scored to assist clinicians in assessing the disease.

[0034] The present invention adopts the above-mentioned technical solution to achieve the above objectives, which makes up for the shortcomings of the prior art, is reasonably designed, and is easy to operate. Attached Figure Description

[0035] To make the above and / or other objects, features, advantages and examples of the present invention more apparent and understandable, the accompanying drawings used in the specific embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0036] Figure 1 A flowchart illustrating the workflow of a system that uses AI algorithms to assess the nature of thyroid nodules; Figure 2 Represents a polypeptide spectrum; Figure 3 A graph showing the ROC curve of the validation set for the entire queue; Figure 4 This represents the ROC curve of the entire queue's test set. Detailed Implementation

[0037] Those skilled in the art can refer to the content of this document and appropriately replace and / or modify the process parameters to achieve the desired results. However, it should be particularly noted that all similar replacements and / or modifications are obvious to those skilled in the art and are considered to be included in this invention. The products and preparation methods described in this invention have been described through preferred examples, and those skilled in the art can obviously modify or appropriately change and combine the products and preparation methods described herein without departing from the content, spirit, and scope of this invention to realize and apply the technology of this invention.

[0038] Unless otherwise defined, the technical and scientific terms used herein have the same meanings as commonly understood by one of ordinary skill in the art to which this invention pertains. This invention uses the methods and materials described herein; however, other suitable methods and materials known in the art may also be used. The materials, methods, and examples described herein are illustrative only and are not intended to be limiting. All publications, patent applications, patent cases, provisional applications, database entries, and other references mentioned herein are incorporated herein by reference in their entirety. In case of conflict, the definitions included in this specification shall prevail.

[0039] Unless otherwise specified, the materials, methods, and examples described herein are exemplary and not limiting. While similar or equivalent methods and materials can be used to implement or test the invention, suitable methods and materials are described herein.

[0040] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Furthermore, in the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other.

[0041] It should be understood that none of the technical solutions claimed in this invention relate to the diagnosis and treatment of diseases.

[0042] To facilitate understanding of the embodiments of the present invention, the abbreviations and key terms that may be involved in the embodiments of the present invention will first be explained or defined. For undefined abbreviations or key terms, they are all conventionally understood by those skilled in the art.

[0043] Furthermore, unless otherwise specified, the experimental methods used in the embodiments are conventional methods; the materials and reagents used are commercially available unless otherwise specified. Reagents or instruments whose manufacturers are not specified are all commercially available products. All disclosures and other references mentioned herein are incorporated herein by reference in their entirety.

[0044] For the LASSO algorithm in this invention, see Tibshirani, R. (1996) Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society Series B: Statistical Methodology 58, 267-288. For the XGBoost algorithm, see Chen, T., and Guestrin, C. (2016) XGBoost: A Scalable Tree Boosting System. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 785–794, Association for Computing Machinery, San Francisco, California, USA. For RFE, see Guyon, I., Weston, J., Barnhill, S., and Vapnik, V. (2002) Gene selection for cancer classification using support vector machines. Machine learning 46, 389-422. For Boruta, see Kursa, M. B., and Rudnicki, W. R. (2010) Feature selection with the Boruta package. Journal of statistical software 36, 1-13. For SHAP, see Lundberg, S. M., and Lee, S.-I. (2017) A unified approach to interpreting model predictions. Advances in neural information processing systems 30.

[0045] It should be understood that the term "at least 3" in this invention includes solutions with at least 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, and all 15. It should also be understood that, provided they do not conflict with each other, violate natural laws, or violate relevant laws and regulations, the features described in this application can be combined to form new technical solutions, and these new technical solutions are naturally described in this application because the relevant technical features are already described therein.

[0046] The present invention will now be described in detail.

[0047] Example 1: A method for using AI algorithms to screen for detection targets for the nature assessment of thyroid nodules is provided, as follows.

[0048] 1. Sample inclusion Samples collected from the prospective clinical trial in China (registration number: ChiCTR2000034454) between September 2019 and September 2022 were included.

[0049] The inclusion criteria were: (1) age ≥18 years and ≤70 years, (2) newly diagnosed patients with thyroid nodules who had not received drug treatment, (3) thyroid fine needle biopsy performed before surgery, with cytological pathological diagnosis of Bethesda II-VI, (4) if it was a Bethesda III-VI sample, follow-up information or total / partial thyroidectomy was required, and a histological report of the corresponding cytological biopsy nodule was required, (5) the patient voluntarily participated after giving informed consent.

[0050] Exclusion criteria were: (1) patients who had not undergone surgery, and (2) insufficient FNA sample size (as determined by the laboratory).

[0051] This study used ultrasound-guided fine-needle aspiration biopsy to precisely obtain target thyroid cells, and sampled from different sites to reduce heterogeneity interference. Patient information was collected, including gender, age, and BRAF V600E mutation information. After filtering out samples with missing data, a total of 731 cases were included in the analysis.

[0052] The obtained puncture samples were lysed with three times the amount of red blood cell lysis buffer. Red blood cells were washed away by centrifugation, retaining thyroid cells. Protein extraction and lysis were performed using pressure cycling technology to convert the proteins into linear structures. Further digestion with LysC and Trypsin protease at an enzyme:protein ratio of 1:80 was used to prepare thyroid tumor polypeptide samples. The obtained polypeptides were washed with C18 solution, dried, and then reconstituted with 0.1% formic acid aqueous solution. Concentration was measured at A280 absorbance.

[0053] 2. Quantitative analysis of peptides Fifteen standard relabeled peptides corresponding to peptide segments were synthesized (each peptide's C-terminus K or R was isotopically labeled, serving as internal standard peptides). The linear range of each peptide segment was detected, and the lower limit concentration of the range was determined based on the linear range. The content of each peptide was optimized by 10-50 times, and the final combined concentration was determined. A standard relabeled mixed peptide sample of thyroid tissue was constructed (15 peptide segments diluted to a certain concentration (median of the sample range), with 0.1% FA as the solvent).

[0054] The concentration of peptides extracted from thyroid cells was adjusted to 0.5 µg / µL, and one mass of relabeled peptide mixture was added to each sample. The mixed samples were then subjected to peptide separation using a liquid chromatography system. A Hypersil GOLD™ 100 x 2.1 mm x 1.9 mm column was used, with elution at a flow rate of 0.2 mL / min and a HPLC gradient of 12 min or 25 min, as shown in Table 1. Phase A was an aqueous solution containing 0.1% formic acid, and Phase B was an acetonitrile solution containing 0.1% formic acid. The AQUA (refer to https: / / www.nature.com / articles / nprot.2010.196) relabeled peptides (corresponding to the 15 peptides mentioned above) and their corresponding endogenous peptides (peptides contained in the sample, i.e., the 15 peptides provided above) were simultaneously eluted. The eluted peptides were then collected using a TripleQuad 4500MD mass spectrometer to obtain mass spectrometry data containing peptide quantitative information. The mass spectrometry parameter settings are shown in Table 2. Three ion pairs were collected from each of the lightly and heavily labeled peptides for subsequent quantification. The detected peptide spectra are shown below. Figure 2 As shown.

[0055] Table 1. Liquid phase gradient parameter settings

[0056] Table 2. Mass Spectrometry Parameter Settings

[0057] 3. Quantitative analysis of peptide mass spectrometry data First, the mass spectrometry data was converted to .mzML format. Then, the OpenSwathAssayGenerator module in OpenMS software was used to screen and optimize the peptide chromatogram library file. This module annotates all fragment ion information according to predetermined criteria (i.e., fragment ions, retention time, mass-to-charge ratio, and signal information). Fragment ions are filtered to a minimum of 1 to improve the sensitivity of peptide detection, ultimately generating a TraML format file. The MRMMAPper module was used to match the .mzML format mass spectrometry data with the information in the TraML file. With the precursor ion error set to 1, a multi-mapping algorithm was used to match one or more mapping targets to the mzML chromatogram, generating an mzML format file containing the mappings. Further, the OpenSwathAnalyzer module was used to perform peak selection in a non-strict mode on the .mzML file containing the mapping metadata, and quantification was performed by scoring. The result with the largest peak area was used to characterize the expression level of the corresponding peptide.

[0058] 4. Feature Selection and Model Building First, the acquired dataset of 731 cases was divided into two parts: a discovery set of N=584 data points and an external validation set of M=147 data points. The discovery set was used for model building and related data mining, while the external validation set was used for model performance evaluation.

[0059] Feature selection was performed on the discovery dataset. During feature selection, the discovery set was randomly partitioned into 2 / 8 segments, and eight data points and their corresponding peptide quantification information were extracted. Four algorithms were used for feature selection, including... a) LASSO algorithm; b) Recursive Feature Elimination (RFE) algorithm based on XGBoost and its model importance; c) Recursive Feature Elimination (RFE) algorithm based on XGBoost and its Shapley Value; d) Based on XGBoost and its Shapely Value, the Boruta algorithm is used, and the frequency of occurrence is considered.

[0060] Each feature selection method was performed 100 times, for a total of 400 times. Features were then ranked by frequency of occurrence, and the 15 most frequent features were selected for subsequent model construction.

[0061] Finally, using the selected 15 peptide quantitative data, patient gender, age, and BRAF V600E information as input, XGBoost was used as the prediction model algorithm, and a genetic algorithm was combined to optimize the model's hyperparameters: a grid search strategy was used to fine-tune five hyperparameters, including the learning rate (learning_rate), the random sampling ratio (subsample) for each tree, the proportion of columns randomly sampled per tree (colsample_bytree), the proportion of columns sampled per node split (colsample_bylevel), and the balancing parameter (scale_pos_weight). In this step, the dataset was further divided into two subsets: a training set and an internal validation set. Based on the cross-validation results of the internal validation set, the model with the optimal parameters was finally determined and the model parameters were fixed. After the model optimization was completed, tests were performed on the validation set, and prediction results were obtained, such as... Figure 3 , Figure 4 And as shown in Table 3; for comparison, Figure 3 , Figure 4 Table 3 also shows the prediction results when modeling using only BRAF mutation, age, and sex. Figure 3 , Figure 4 In Table 3, the BRAF V600E mutation + final characteristic peptides specifically refer to three final characteristic peptides selected from the aforementioned 15 peptides. In the scheme with three or more final characteristic peptides, the sensitivity of Bethesda III to V nodules obtained by these three final characteristic peptides is the lowest, at 81.6%. It can be seen that adding some of the final characteristic peptides from the 15 peptides mentioned above to the model can effectively improve the model's performance, and this improvement is very obvious in the validation set.

[0062] Table 3. Validation set results of the model

[0063] Based on the validation set results above, as further validation of the final characteristic peptide and BRAF V600E mutation, EGFR mutation was further considered in addition to BRAF V600E mutation. The AUC of the validation set results reached 0.943 (Bethesda II to VI nodules), 0.916 (Bethesda III to V nodules), and 0.901 (Bethesda III and IV nodules), respectively. Therefore, from the perspective of prediction accuracy, BRAF V600E mutation and EGFR mutation can be combined into the model, system, and prediction method of the final characteristic peptide to predict the nature of thyroid nodules, so as to assess and classify the stage and nature of thyroid nodules, assist clinicians in the identification of thyroid nodules, and thus alleviate the problem of inaccurate diagnosis in clinical practice.

[0064] Example 2: Based on the foregoing embodiments, a method for constructing a model to predict the nature of thyroid nodules is provided, comprising: Obtain training datasets and internal validation sets. Each training or validation dataset contains quantitative information on peptides, BRAF V600E information, and whether the thyroid nodules are malignant. Based on the training set, the second processor uses the quantitative information of the 15 peptides obtained in the aforementioned embodiment, along with BRAFV600E information, the subject's gender and age as feature information, to construct and train an actual prediction model based on the second AI algorithm. The second processor uses the internal validation set to perform cross-validation on the trained actual prediction model to determine the final actual prediction model.

[0065] This embodiment also provides a model for predicting the nature of thyroid nodules based on the construction method described above.

[0066] Example 3: Based on the foregoing embodiments, a system for evaluating the nature of thyroid nodules using AI algorithms is provided, comprising: The acquisition module is configured to acquire fine-needle aspiration (FNA) tissue biopsy samples of the thyroid gland of the target subject. A quantitative polymerase chain reaction (PCR) detection module configured to detect BRAF V600E information in tissue biopsy samples; The MRM MS detection module is configured to detect quantitative information of at least 3 final characteristic peptides from the following 15 peptides in tissue biopsy samples: ELSDFISYLQR, FSMVVQDGIVK, GESGPSGPAGPTGAR, GETGPSGPVGPAGAVGPR, GLGTDEDTLIEILASR, GQTLVVQFTVK, IYLYTLNDNAR, LFDQAFGLPR, LGPLVEQGR, LMEDLDR, LQDAEIAR, MQQQLDEYQELLDIK, SFVLNLGK, SGVYQHVTGEMMGGHAIR, and TVAAPSVFIFPPSDEQLK. The evaluation module includes a third processor configured to evaluate the nature of the thyroid nodules using a model that predicts the nature of the thyroid nodules, taking quantitative information of the final characteristic peptide along with BRAF V600E information, the gender and age of the target subject as characteristic information.

[0067] Preferably, the assessment of the nature of the thyroid nodule is intended to characterize the malignancy risk of the thyroid nodule.

[0068] Preferably, the evaluation module is configured as follows: Obtain personally identifiable information of the target subjects, including their gender and age; The device is communicatively connected to the quantitative polymerase chain reaction (PCR) detection module and the MRM MS detection module to obtain BRAF V600E information from each subject in the PCR detection module and quantitative information of the final characteristic peptides from each subject in the MRM MS detection module. From the BRAF V600E information and quantitative information of the final characteristic peptide obtained from each subject, BRAF V600E information and quantitative information of the final characteristic peptide that match the personal identification information of the target subject are identified in order to assess the nature of thyroid nodules.

[0069] This embodiment also provides the use of the system in the preparation of an apparatus for assessing the properties of thyroid nodules in subjects based on targeted detection peptides and AI.

[0070] Example 4: A computer-readable storage medium is also provided, which stores a computer program that can be executed by a processor. When the computer program is executed by the processor, it performs at least one step in the aforementioned method for screening detection targets for the nature assessment of thyroid nodules using AI algorithms, and can achieve the same technical effect. To avoid repetition, this embodiment will not elaborate further.

[0071] Example 5: A computer device includes a memory, a processor, a communication interface, and a communication bus; wherein the memory, processor, and communication interface communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to execute the computer programs stored in the memory, wherein when the processor executes the computer programs, it implements at least one step of the aforementioned method for screening detection targets for the nature assessment of thyroid nodules using AI algorithms, and achieves the same technical effect. To avoid repetition, this embodiment will not elaborate further.

[0072] Computer-readable media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0073] The conventional techniques described in the above embodiments are existing technologies known to those skilled in the art, and therefore will not be described in detail here.

[0074] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0075] Although the present invention has been described in detail and specific embodiments have been cited, it will be apparent to those skilled in the art that various changes or modifications can be made without departing from the spirit and scope of the invention.

[0076] While the foregoing detailed descriptions have shown, described, and pointed out novel features applicable to various embodiments, it should be understood that various omissions, substitutions, and changes may be made to the form and details of the described apparatus or methods without departing from the spirit of this disclosure. Furthermore, the various features and methods described above may be used independently of each other or may be combined in various ways. All possible combinations and sub-combinations are intended to fall within the scope of this disclosure. Many of the foregoing embodiments include similar components, and therefore, these similar components are interchangeable in different embodiments. Although the invention has been disclosed in the context of certain embodiments and examples, those skilled in the art will understand that the invention extends beyond the specifically disclosed embodiments to other alternative embodiments and / or applications, as well as their obvious modifications and equivalents. Therefore, the invention is not intended to be limited to the specific disclosure of the preferred embodiments herein.

[0077] All matters not covered in this invention are common knowledge.

Claims

1. A method for screening detection targets for assessing the nature of thyroid nodules using AI algorithms, characterized in that: The method includes: Obtain the training dataset, each training dataset containing the subject's peptide quantification information, BRAF V600E information, and whether the thyroid nodule is malignant; Based on the training set, the first processor uses the peptide quantitative information of different peptide combinations as test feature information to construct and train corresponding test prediction models for the property assessment of thyroid nodules based on the first AI algorithm, compares the performance of each trained test prediction model, and selects the first group of candidate prediction models that meet the performance requirements. Based on the training set, the first processor uses the peptide quantitative information of different peptide combinations as test feature information to construct and train corresponding test prediction models based on at least two derivative algorithms of the second AI algorithm, compares the performance of each trained test prediction model, and selects a second group of candidate prediction models that meet the performance requirements. The first processor selects the quantitative information of the 15 most frequently occurring peptides from the test feature information used in the first group of candidate prediction models and the second group of candidate prediction models, which are used as detection targets for the nature assessment of thyroid nodules.

2. The method according to claim 1, characterized in that: The 15 peptides are as follows: ELSDFISYLQR, FSMVVQDGIVK, GESGPSGPAGPTGAR, GETGPSGPVGPAGAVGPR, GLGTDEDTLIEILASR, GQTLVVQFTVK, IYLYTLNDNAR, LFDQAFGLPR, LGPLVEQGR, LMEDLDR, LQDAEIAR, MQQQLDEYQELLDIK, SFVLNLGK, SGVYQHVTGEMMGGHAIR, TVAAPSVFIFPPSDEQLK.

3. The method according to claim 1 or 2, characterized in that: The first AI algorithm includes the LASSO algorithm; and / or The second AI algorithm includes the XGBoost algorithm; and / or The derivative algorithms of the second AI algorithm include: a recursive feature elimination algorithm based on XGBoost and its model importance, a recursive feature elimination algorithm based on XGBoost and its Shapley value, and a Boruta algorithm based on XGBoost and its Shapley value.

4. The method according to claim 1 or 2, characterized in that: After the model selects 15 peptides, at least 3 are selected as the final feature peptides based on operational complexity and industrialization costs. The final feature peptides, together with BRAF V600E information, serve as the microbiological feature information of the actual prediction model.

5. The method according to claim 1 or 2, characterized in that: The method further includes: Based on the training set, the second processor uses the quantitative information of the final feature peptide, along with BRAF V600E information, the subject's gender and age as feature information, to construct and train an actual prediction model based on the second AI algorithm. The second processor uses the validation set to perform cross-validation on the trained actual prediction model to determine the final actual prediction model.

6. A method for constructing a model to predict the nature of thyroid nodules, characterized in that... include: Obtain training datasets and internal validation sets. Each training or validation dataset contains quantitative information on peptides, BRAF V600E information, and whether the thyroid nodules are malignant. Based on the training set, the second processor uses the quantitative information of the 15 peptides obtained in the method of any one of claims 1-5, together with BRAF V600E information, the subject's gender and age as feature information, to construct and train an actual prediction model based on the second AI algorithm. The second processor uses the internal validation set to perform cross-validation on the trained actual prediction model to determine the final actual prediction model.

7. A model for predicting the properties of thyroid nodules obtained by the construction method of claim 6.

8. A system for evaluating the nature of thyroid nodules using AI algorithms, characterized in that... include: The acquisition module is configured to collect tissue biopsy samples from the thyroid gland of the target subject via fine-needle aspiration. A quantitative polymerase chain reaction (PCR) detection module configured to detect BRAF V600E information in tissue biopsy samples; The MRM MS detection module is configured to detect quantitative information of at least 3 of the following 15 peptides in tissue biopsy samples as the final characteristic peptides: ELSDFISYLQR, FSMVVQDGIVK, GESGPSGPAGPTGAR, GETGPSGPVGPAGAVGPR, GLGTDEDTLIEILASR, GQTLVVQFTVK, IYLYTLNDNAR, LFDQAFGLPR, LGPLVEQGR, LMEDLDR, LQDAEIAR, MQQQLDEYQELLDIK, SFVLNLGK, SGVYQHVTGEMMGGHAIR, and TVAAPSVFIFPPSDEQLK. The evaluation module includes a third processor configured to evaluate the nature of the thyroid nodules using a model that predicts the nature of the thyroid nodules, taking quantitative information of the final characteristic peptide along with BRAF V600E information, the gender and age of the target subject as characteristic information.

9. Use of the system of claim 8 in the preparation of an apparatus for assessing the properties of thyroid nodules in a subject based on targeted detection peptides and AI.

10. The use of a reagent for detecting a targeted detection peptide in the preparation of an evaluation product, said evaluation product being used to assess the properties of thyroid nodules in a subject based on the targeted detection peptide and an AI, characterized in that: The targeted detection peptide is selected from at least 3 of the following 15 peptides: ELSDFISYLQR, FSMVVQDGIVK, GESGPSGPAGPTGAR, GETGPSGPVGPAGAVGPR, GLGTDEDTLIEILASR, GQTLVVQFTVK, IYLYTLNDNAR, LFDQAFGLPR, LGPLVEQGR, LMEDLDR, LQDAEIAR, MQQQLDEYQELLDIK, SFVLNLGK, SGVYQHVTGEMMGGHAIR, and TVAAPSVFIFPPSDEQLK.

Images