A device, method, system for diagnosing gastric-type cervical adenocarcinoma and application thereof

Abstract

The application discloses a device, method, system and application thereof for diagnosing stomach-type cervical adenocarcinoma, and belongs to the technical field of medical diagnosis. The evaluation device comprises a data acquisition device and a data processing device. The data acquisition device is used for acquiring a detection value of a stomach-type cervical adenocarcinoma marker detected from cervical tissue of an evaluation object, wherein the detection value is a staining intensity of immunohistochemistry and a percentage of positive cells of each intensity; and the marker is PGC, a combination of PGC and DNMT1 or a combination of the three. The data processing device is used for calculating an immunohistochemistry (IHC) score (i.e., H-score) of the marker according to the detection value, and judging a risk range based on the score.

Description

Technical Field

[0001] This invention belongs to the field of biological detection technology, specifically relating to devices, methods, systems, and applications for diagnosing or evaluating gastric cervical adenocarcinoma. Background Technology

[0002] Cervical adenocarcinoma is a highly biologically and clinically heterogeneous gynecological malignancy, mainly divided into two categories: human papillomavirus (HPV)-associated adenocarcinoma (HPV-CA) and HPV-unassociated adenocarcinoma. With the widespread adoption of HPV vaccination and screening, the incidence of HPV-driven cervical cancer is declining. However, HPV-unassociated subtypes, especially gastric-type cervical adenocarcinoma (GAS), have become an increasingly serious challenge for diagnosis and treatment due to their unique clinicopathological features and poor prognosis. GAS is characterized by high invasiveness, a tendency to metastasize, insensitivity to conventional radiotherapy and chemotherapy, and extremely poor prognosis. Therefore, accurate and early diagnosis of GAS is crucial for developing individualized treatment plans and improving patient survival outcomes.

[0003] Currently, the clinical diagnosis of GAS mainly relies on morphological observation by pathologists, combined with traditional auxiliary examination methods, such as high-risk HPV DNA testing, liquid-based thin-layer cytology (TCT), and colposcopy biopsy. However, the existing diagnostic system faces significant challenges in dealing with GAS: First, well-differentiated GAS (such as microadenocarcinoma) has a cell morphology that is extremely similar to normal cervical glands, making it easily misdiagnosed as benign hyperplasia or inflammation; second, some cases present mixed morphological characteristics of GAS and HPV-CA, leading to ambiguous classification and low diagnostic consistency; more importantly, GAS is mainly driven by non-HPV infection factors, resulting in very low sensitivity of routine HPV testing and cytology screening, causing a large number of missed diagnoses.

[0004] At the biomarker level, clinical trials currently utilize immunohistochemical (IHC) markers such as MUC6, HIK1083, and PAX8 to assist in the identification of gastric atherosclerosis (GAS). However, these markers have limited diagnostic performance, exhibiting low sensitivity (approximately 51% for MUC6 and 64% for HIK1083) and insufficient specificity, failing to meet the demands of precise diagnosis. Regarding biomarker discovery methods, traditional strategies often rely on simple statistical analyses of proteomics data (such as t-tests) to screen differentially expressed proteins. This approach fails to fully consider the complex interaction networks and biological pathway information between proteins, resulting in biomarkers with limited clinical translational value. In recent years, while some studies have attempted to apply machine learning (such as random forests) or deep learning models for biomarker discovery, these methods are generally plagued by the "curse of dimensionality" for rare malignant tumors like GAS, which have extremely scarce sample sizes. They are prone to overfitting, exhibit poor model generalization ability, and are often "black box" models lacking necessary biological interpretability, hindering their translation into clinical practice.

[0005] In summary, the diagnosis of gastric cervical adenocarcinoma currently faces multiple technical bottlenecks, including low accuracy, poor performance of existing biomarkers, and the failure of advanced algorithms due to sample scarcity. Therefore, there is an urgent need to develop a rapid, accurate, and effective device, method, or system for identifying gastric cervical adenocarcinoma. Summary of the Invention

[0006] This invention aims to address one, two, or all of the following problems existing in current diagnostic techniques for gastric-type cervical adenocarcinoma (GAS): 1. Low diagnostic accuracy: Reliance on subjective morphological observation and the insufficient sensitivity and specificity of existing immunohistochemical markers (such as MUC6 and HIK1083) make GAS easy to be misdiagnosed or missed.

[0007] 2. Difficulty in modeling with small sample sizes: GAS is a rare disease with scarce samples. Traditional machine learning and deep learning methods are prone to overfitting and have poor generalization ability under these conditions.

[0008] 3. Lack of interpretability: Conventional "black box" models are difficult to explain their decision-making basis at the biological network level, and the clinical translation of the screened biomarkers is difficult.

[0009] To this end, the present invention provides an objective and quantitative diagnostic scheme based on a specific combination of biomarkers to achieve accurate and interpretable diagnosis of GAS.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, an assessment device for evaluating gastric-type cervical adenocarcinoma is provided, characterized in that it comprises: A data acquisition device is used to acquire detection data of the evaluation subject, wherein the detection data is the detection value of gastric cervical adenocarcinoma markers obtained from the human cervical tissue of the evaluation subject suspected of having gastric cervical adenocarcinoma, which is the staining intensity of the immunohistochemistry of the human cervical tissue and the percentage of positive cells of each staining intensity among all evaluation cells, and the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1, two markers. A data processing device is used to calculate the immunohistochemical (IHC) score of the biomarker based on the detection value, and to determine whether the biomarker is within the risk range based on the IHC score.

[0011] Preferably, the IHC score of the marker is the H-score, which is calculated as follows: H-score = Σ (Pi × Ii), where Pi represents the percentage of positive cells with staining intensity level i among all evaluated cells, and Ii represents the staining intensity grading value, with i taking values ​​of 0, 1, 2, and 3, corresponding to no staining, weak positive, moderate positive, and strong positive, respectively.

[0012] Furthermore, if the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1, then when the IHC score of PGC is higher than a first preset threshold and the IHC score of DNMT1 is lower than a second preset threshold, a high-risk assessment result for gastric cervical adenocarcinoma is given; if the IHC score of PGC is lower than the first preset threshold and / or the IHC score of DNMT1 is higher than the second preset threshold, a low-risk assessment result for gastric cervical adenocarcinoma is given.

[0013] Furthermore, the data processing device has a built-in trained machine learning classifier for calculating the risk assessment results of gastric cervical adenocarcinoma based on the IHC score of the marker; the machine learning classifier is trained based on an immunohistochemical scoring cohort including patients with gastric cervical adenocarcinoma and patients with human papillomavirus-associated cervical adenocarcinoma.

[0014] Secondly, a computer-executed method is provided for outputting risk assessment results for gastric-type cervical adenocarcinoma, the method comprising the following steps: a) Read the test data of the patient or subject, wherein the test data is the detection value of gastric cervical adenocarcinoma markers obtained from the human cervical tissue of the subject suspected of having gastric cervical adenocarcinoma, which is the staining intensity of the immunohistochemistry of the human cervical tissue and the percentage of positive cells of each staining intensity among all the assessment cells, and the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1 of the two markers; b) Data processing: Calculate the immunohistochemical (IHC) score of the biomarker based on the detection value in step a); c) Output the risk assessment results based on the IHC score.

[0015] Preferably, the data processing in step b) is performed by a trained machine learning model, which is trained based on an immunohistochemical scoring cohort including patients with gastric cervical adenocarcinoma and patients with human papillomavirus-associated cervical adenocarcinoma.

[0016] Thirdly, a computer system is provided, characterized in that it is configured to perform the following operations: receiving an immunohistochemical (IHC) score of a gastric cervical adenocarcinoma marker in the cervical tissue of an assessment subject, wherein the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1, two markers; and generating a risk assessment result for gastric cervical adenocarcinoma based on the IHC score.

[0017] Fourthly, the present invention provides the use of a combination of pepsinogen C (PGC) and DNA methyltransferase 1 (DNMT1) in the preparation of diagnostic reagents or kits for gastric type cervical adenocarcinoma.

[0018] For example, the diagnostic reagent or kit is used to detect the expression levels of PGC and DNMT1 in cervical tissue samples; the expression level is determined by immunohistochemical (IHC) staining intensity, the proportion of positive cells, or the immunohistochemical score (H-score); the H-score is calculated using the following formula: H-score = Σ(Pi × Ii) Where Pi is the percentage of positive cells at staining intensity level i, and Ii is the staining intensity grading value, with i taking values ​​of 0, 1, 2, and 3, corresponding to no staining, weak positive, moderate positive, and strong positive, respectively.

[0019] Furthermore, the diagnostic reagent or kit is used to differentiate between gastric-type cervical adenocarcinoma and HPV-associated cervical adenocarcinoma.

[0020] Preferably, a sample is considered to be at high risk for gastric-type cervical adenocarcinoma if it meets the following conditions: The PGC's H-score is higher than the first preset threshold, and The H-score of DNMT1 is lower than the second preset threshold.

[0021] For example, the first preset threshold is 0.01, and the second preset threshold is 1.20.

[0022] Furthermore, the diagnostic reagent or kit comprises: At least one of the following: a specific antibody against PGC, a specific antibody against DNMT1, an antigen retrieval solution, a chromogenic reagent, a positive control slide, and a negative control reagent.

[0023] This invention also provides the use of pepsinogen C (PGC) as a gastric type differentiation marker in the preparation of diagnostic reagents for gastric type cervical adenocarcinoma.

[0024] This invention also provides the use of DNA methyltransferase 1 (DNMT1) as an auxiliary diagnostic marker for gastric cervical adenocarcinoma in the preparation of diagnostic reagents.

[0025] For example, the diagnostic reagent is an immunohistochemical detection reagent.

[0026] Compared with the prior art, the present invention has one of the following beneficial effects: 1. High diagnostic accuracy: Based on a key biomarker combination (PGC-DNMT1) screened using an advanced deep learning framework (WEDGE), objective quantitative diagnostic evidence is provided. Experiments show that the area under the diagnostic curve (AUC) of this dual biomarker combination in the immunohistochemical validation cohort reaches 0.919, significantly outperforming existing single biomarkers and effectively reducing the rates of misdiagnosis and missed diagnosis.

[0027] 2. Solving the small sample problem: The core biomarker relied upon by this invention comes from the WEDGE framework, which is specifically designed for diseases with small sample sizes. This framework overcomes the modeling bottleneck of traditional methods under conditions of scarce samples by data augmentation and integration of biological prior knowledge, thus ensuring the reliability of screening results.

[0028] 3. Combining interpretability and clinical applicability: The diagnosis is based on the immunohistochemical staining score (H-score), a scoring method widely used in pathology departments with clear standards and strong operability. Simultaneously, the combination of biomarkers has clear biological significance, facilitating clinical understanding and acceptance, and greatly promoting the clinical translation and application of the technology.

[0029] 4. Achieving intelligence and integration: By integrating machine learning classifiers, automatic analysis and risk assessment of detection data can be achieved. Furthermore, it can be integrated with automated testing equipment to form an integrated "detection-analysis-reporting" solution, improving diagnostic efficiency and standardization.

[0030] The PGC and DNMT1 described in this invention are briefly introduced as follows: (1) Pepsinogen C (PGC), also known as pepsinogen C, is a digestive enzyme precursor secreted by chief cells and mucus neck cells of the gastric mucosa, and belongs to the aspartic protease family. Under physiological conditions, PGC is activated in an acidic environment and converted into pepsin C with proteolytic activity, participating in the digestion of food proteins. In this invention, by detecting the expression level of PGC in cervical tissue, it can be used to characterize whether the tissue has gastric differentiation characteristics, thereby providing a basis for risk assessment of gastric type cervical adenocarcinoma.

[0031] (2) DNMT1 (DNA methyltransferase 1) DNMT1 (DNMT1) is a key DNA methyltransferase primarily responsible for maintaining genomic methylation during DNA replication, thus playing a central role in epigenetic regulation. DNMT1 participates in gene expression silencing, chromatin structure maintenance, and cell differentiation regulation by catalyzing the methylation of cytosine residues. In various tumors, aberrant DNMT1 expression is closely related to tumor suppressor gene silencing and tumor development. This invention utilizes immunohistochemical detection of DNMT1 expression in cervical tissue, combined with its expression relationship with other biomarkers, to provide an auxiliary assessment of the risk of gastric cervical adenocarcinoma. Attached Figure Description

[0032] Figure 1 This is a confusion matrix comparing the performance of the ablation WGAN-GP model (left) and the proposed WEDGE model (right) in classifying HPV-CA and GAS groups.

[0033] Figure 2 It is a confusion matrix that demonstrates the classification performance of the best combination of dual protein biomarkers (PGC and DNMT1) in the test queue.

[0034] Figure 3 It is a confusion matrix that demonstrates the classification performance of the best combination of dual protein biomarkers (PGC and DNMT1) in the external validation queue.

[0035] Figure 4 This is a quantitative comparison chart of IHC scores for three biomarkers (PGC and DNMT1) in GAS, HPV-CA, and the control group.

[0036] Figure 5 This is a comparison chart of IHC scores for PGC, where Tukey's multiple comparison test was used to determine statistical significance (***p<0.001, **p<0.01, *p<0.05, ns: no significance).

[0037] Figure 6This is a comparison chart of IHC scores for DNMT1, where statistical significance was determined using Tukey's multiple comparison test (***p<0.001, **p<0.01, *p<0.05, ns: no significance).

[0038] Figure 7 These are ROC curves comparing the diagnostic performance of different combinations of biomarkers in the IHC cohort.

[0039] Figure 8 It is a confusion matrix that demonstrates the classification performance of the dual-protein combination model.

[0040] Figure 9 This is a schematic diagram of the integrated assessment device used in Example 4 for evaluating gastric-type cervical adenocarcinoma.

[0041] Figure 10 This is a flowchart of the computer implementation method for evaluating gastric-type cervical adenocarcinoma in Example 5.

[0042] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Detailed Implementation

[0043] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Other implementation methods obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the protection scope of the present invention.

[0044] In one embodiment, the evaluation device of the present invention includes: The present invention will further demonstrate the selection principle, process and effect of the above-mentioned biomarkers through a specific embodiment, so as to enable those skilled in the art to understand the essence of the present invention.

[0045] Example 1: Screening of specific biomarkers for gastric cervical adenocarcinoma This embodiment illustrates the screening process for gastric cervical adenocarcinoma (GAS) specific biomarkers in this invention. The screening process is based on the analysis of proteomics data using a deep learning model.

[0046] I. Sample Sources and Data Acquisition A training cohort was constructed, comprising 114 patients with cervical adenocarcinoma, among whom: 73 patients with gastric-type cervical adenocarcinoma (GAS); 71 patients with human papillomavirus-associated cervical adenocarcinoma (HPV-CA) were included.

[0047] Proteomics analysis was performed on all samples to obtain protein expression matrix data. Differential expression analysis identified 1114 differentially expressed proteins as candidate features.

[0048] II. Data Augmentation Processing To address the issue of limited sample size, a Wasserstein generative adversarial network with gradient penalty (WGAN-GP) was used to enhance the original protein expression data.

[0049] The specific steps are as follows: 1. Normalize the original protein expression data to the range [-1, 1]; 2. Construct generator and discriminator networks; 3. Generate simulated protein expression data through adversarial training; 4. Data amplification is performed with an enhancement factor of 5, that is, 5 simulated samples are generated for each real sample.

[0050] To avoid data leakage and ensure the model's generalization ability, this embodiment strictly limits WGAN-GP data augmentation to the internal training loop of 5-fold cross-validation. In each fold of the entire training set of 144 samples (73 GAS, 71 HPV-CA), approximately 115 samples (4 / 5) are used as the training subset for that fold, and the remaining 29 samples (1 / 5) are used as the validation subset. Therefore, WGAN-GP augmentation is performed only on the 115 training samples for that fold. With the augmentation factor set to 5, the model generates 5 synthetic samples for each real sample, resulting in approximately 575 synthetic samples (285 GAS, 290 HPV-CA) in each fold. These synthetic samples are inversely transformed and merged with the original 144 real training samples to finally construct the augmented training dataset, totaling 719 samples, including 358 GAS samples and 361 HPV-CA samples.

[0051] The amplified data exhibited a consistent distribution with the original data in the principal component analysis (PCA) space, indicating that the generated data had good biological rationality.

[0052] III. Modeling Two-Stream Graph Convolutional Networks Based on the amplified data, a dual-stream GCN (D-GCN) was constructed for disease classification and feature selection.

[0053] Specifically, it includes: (1) Graph structure construction Each sample is represented as a graph structure, where: Node: Protein; Node characteristics: protein expression levels; The edges include two types: protein-protein interaction networks (PPIs), sourced from the STRING database; and gene regulatory networks (GRNs), sourced from the TRRUST database.

[0054] (2) Dual-stream network processing PPI flow: handling undirected protein-protein interactions; GRN flow: dealing with directed gene regulatory relationships; Each stream consists of two graph convolutional layers and enhances feature representation through residual connections.

[0055] (3) Feature fusion Perform the following on the outputs of the two streams: Global average pooling; Max pooling; Attention mechanism fusion; The final output is the classification result (GAS or HPV-CA).

[0056] IV. Model Training and Evaluation After obtaining the training dataset amplified by WGAN-GP, the WEDGE model constructed in this invention was systematically trained and its performance was evaluated to verify its effectiveness and stability in the classification task of gastric cervical adenocarcinoma.

[0057] (1) Training set partitioning and cross-validation To fully utilize the limited sample size and reduce the impact of random factors on model performance, this embodiment employs a 5-fold cross-validation strategy for model training. Specifically, the amplified training data is randomly divided into five subsets of roughly equal size and with similar class proportions. In each round of training, four subsets are selected as the training set, and the remaining subset is used as the validation set; this process is repeated five times, ensuring that each subset serves as a validation set exactly once.

[0058] During each round of training, the ratio of gastric cervical adenocarcinoma samples to HPV-related cervical adenocarcinoma samples was kept approximately consistent to avoid class imbalance causing bias in model training. Finally, the average of the five rounds of validation results was used as the model's internal validation performance.

[0059] (2) Model training parameters This embodiment uses the WEDGE model, which employs the Adam optimizer for parameter updates. The initial learning rate is preferably set to 0.001, the batch size to 32, and the training epochs to 200. An early stopping mechanism is used during training: training stops when the validation set loss does not significantly decrease within 20 consecutive epochs, and the model parameters corresponding to the optimal validation performance are saved.

[0060] To improve the model's generalization ability, Dropout regularization is introduced into the graph convolutional and fully connected layers, with the optimal Dropout ratio being 0.5. L2 regularization is also used to suppress overfitting.

[0061] (3) Independent test queue verification After completing five-fold cross-validation, the model parameters with the best validation performance were selected, and external validation was performed on an independent test cohort. The test cohort included 88 clinical samples that had not been trained on the model, including gastric cervical adenocarcinoma samples and HPV-related cervical adenocarcinoma samples.

[0062] During the testing phase, no parameters are updated. The trained model is used to classify and predict test samples to accurately reflect the model's generalization performance.

[0063] (4) Evaluation indicators The following indicators are used to comprehensively evaluate the model performance: Accuracy; Sensitivity; Specificity; Precision; F1 value; Area under the receiver operating characteristic curve (AUC).

[0064] AUC is used to reflect the overall classification ability of the model. The closer the value is to 1, the stronger the model's ability to distinguish between the two classes of samples.

[0065] (5) Verification of data augmentation effect To verify the effectiveness of the WGAN-GP data augmentation strategy, the following were constructed: The base model did not employ data augmentation; The WEDGE model with WGAN-GP data augmentation was adopted.

[0066] The two sets of models were compared under the same training conditions and test queue.

[0067] As attached Figure 1As shown, data augmentation significantly improves model accuracy and reduces performance fluctuations between different repeated experiments. This indicates that generating high-quality simulated samples can effectively alleviate the problem of insufficient model training under small sample conditions, thereby improving classification stability and robustness.

[0068] (6) Comparison with existing machine learning methods To further verify the advancement of the method of this invention, several existing mainstream methods were selected as comparisons, including: Random Forest (RF); DIABLO (Data Integration Analysis for Biomarker discovery using LatentcOmponents); BINN (Biologically Informed Neural Networks).

[0069] All the above comparison methods use the same training set, test set, and protein feature input, and are optimized according to their respective recommended parameters to ensure the fairness of the comparison results.

[0070] (7) Comparison results As attached Figure 3 As shown: The WEDGE model achieved the highest accuracy in the test queue; The overall ROC curve is higher than that of other methods; The AUC value was significantly better than RF, DIABLO, and BINN.

[0071] Meanwhile, the WEDGE model achieves a better balance between sensitivity and specificity, and can more accurately distinguish between gastric cervical adenocarcinoma and HPV-related cervical adenocarcinoma.

[0072] (8) Technical effect analysis The above results demonstrate that the present invention combines: WGAN-GP data augmentation; Two-stream graph convolutional networks; Protein-protein interaction networks and gene regulatory networks; It effectively improved the utilization efficiency of small-sample proteomics data and significantly enhanced the accuracy and reliability of gastric cervical adenocarcinoma classification.

[0073] Compared with traditional machine learning methods, this invention not only has higher predictive performance, but also stronger biological interpretability, providing a more reliable technical foundation for subsequent biomarker screening.

[0074] The model was trained using 5-fold cross-validation, and its performance was evaluated on a test queue (88 cases).

[0075] As attached Figure 1 As shown, the WEDGE model proposed in this invention significantly outperforms the model without data augmentation in terms of classification accuracy.

[0076] V. Biomarker Selection Based on the trained model, the contribution of each protein to the classification results is calculated using the feature attribution method (integral gradient), and further screening is performed by combining logistic regression analysis.

[0077] The final selection yielded a combination of markers with significant discriminative power: PGC (Pepsinogen C) DNMT1 (DNA methyltransferase 1) As attached Figure 2 and 3 As shown, the above-mentioned combination of markers exhibits good classification ability in both the test queue and the external validation queue.

[0078] This embodiment introduces a graph neural network modeling method to effectively screen a combination of biologically significant and clinically valuable biomarkers for gastric cervical adenocarcinoma under small sample conditions, providing a reliable basis for subsequent immunohistochemical detection and risk assessment.

[0079] Example 2. Clinical Validation This embodiment is used to verify the diagnostic performance of the biomarker combination screened in Example 1 at the clinical immunohistochemistry (IHC) level. I. Sample Source Construct an immunohistochemistry validation cohort, including: Fifty patients with gastric-type cervical adenocarcinoma (GAS); Sixty patients with HPV-related cervical adenocarcinoma; Ten healthy control samples were included.

[0080] All samples were derived from paraffin-embedded cervical tissue specimens.

[0081] II. Immunohistochemical Detection Methods Standard immunohistochemical (IHC) detection procedure was used: 1. The thickness of the tissue sections is 3–4 μm; 2. Fix with 4% neutral buffered formalin for 6–24 hours; 3. Perform high-temperature and high-pressure antigen retrieval (citrate buffer or Tris-EDTA buffer). 4. Staining was performed using specific antibodies against the following markers: Pepsinogen C (PGC) and DNMT1; 5. Employ a tyramine signal amplification system (TSA) to enhance the signal.

[0082] III. IHC Scoring Method The staining results were semi-quantitatively evaluated using the H-score method: Staining intensity grading: 0 (no), 1 (weak), 2 (medium), 3 (strong); Record the proportion of positive cells corresponding to each intensity; The calculation formula is: H-score = Σ (Pi × Ii) Where: Pi represents the percentage of positive cells corresponding to different staining intensities; Ii represents the corresponding intensity grading value.

[0083] IV. Results Analysis (1) Single marker analysis As attached Figures 4-6 As shown: PGC was significantly highly expressed in the GAS group; DNMT1 was significantly underexpressed in the GAS group; The difference between the two was statistically significant (p<0.05).

[0084] (2) Analysis of combined markers As attached Figure 7 As shown: The area under the ROC curve (AUC) of the PGC + DNMT1 combination was significantly higher than that of the single marker; This indicates that the combined model has higher diagnostic performance.

[0085] (3) Classification performance As attached Figure 8 As shown, the dual-marker combination model can effectively distinguish between GAS and HPV-CA; the confusion matrix shows a significant improvement in classification accuracy.

[0086] V. Threshold Determination Rules Determining the optimal threshold based on ROC curve analysis: When the IHC score of PGC is higher than the first preset threshold (0.01); Furthermore, the IHC score of DNMT1 was lower than the second preset threshold (1.20). This indicates a high risk of gastric-type cervical adenocarcinoma. Otherwise, it is considered low risk.

[0087] This embodiment validates the clinical effectiveness of the immunohistochemical assessment method based on the combination of PGC and DNMT1. The results show that this combination can significantly improve the identification accuracy of gastric-type cervical adenocarcinoma, exhibiting superior sensitivity and specificity compared to traditional single biomarkers, and can provide a reliable basis for clinical diagnosis.

[0088] Example 3: Kit for rapid diagnosis of gastric-type cervical adenocarcinoma This embodiment provides an immunohistochemical detection kit for the auxiliary diagnosis of gastric cervical adenocarcinoma (GAS). Based on the combined detection of PGC and DNMT1 dual markers, it can quickly and accurately distinguish between GAS and HPV-related cervical adenocarcinoma.

[0089] I. Reagent Kit Components The kit includes at least: 1. Specific antibodies against Pepsinogen C; 2. Specific antibodies against DNMT1; 3. Positive control slides; 4. Negative control reagent; 5. Antigen retrieval solution; 6. Sealing liquid; 7. Enzyme-labeled secondary antibody; 8. Colorimetric solution; 9. Instruction manual.

[0090] Preferably, the antibody is a monoclonal antibody suitable for immunohistochemical detection of paraffin sections.

[0091] II. Detection Methods The following steps are used to complete the test: (1) Dewaxing and hydrating paraffin sections of cervical tissue; (2) Perform antigen retrieval; (3) Add PGC and DNMT1 primary antibodies and incubate separately; (4) Add secondary antibody and develop color; (5) Hematoxylin redyeing; (6) Observe and score under a microscope.

[0092] III. Result Determination Calculated based on H-score: PGC H-score > 0.01; DNMT1 H-score < 1.20; The sample was then determined to be at high risk for gastric cervical adenocarcinoma.

[0093] Otherwise, it is classified as non-gastric cervical adenocarcinoma or low risk.

[0094] Example 4: An integrated assessment device for evaluating gastric-type cervical adenocarcinoma This embodiment provides an integrated assessment device for the auxiliary diagnosis of gastric-type endocervical adenocarcinoma (GAS). Figure 9 This device can automatically complete the entire process of pathological image acquisition, biomarker identification, expression level calculation, risk assessment, and result output. It is suitable for hospital pathology departments, medical testing laboratories, and digital pathology centers.

[0095] I. Overall Equipment Structure The integrated assessment equipment mainly includes: (1) Sample input module; (2) Digital image acquisition module; (3) Image preprocessing module; (4) Marker recognition module; (5) Expression level calculation module; (6) Risk assessment module; (7) Result output module; (8) Data storage module; (9) Human-computer interaction module.

[0096] The modules mentioned above are connected via bus or network communication and are controlled by a central processing unit.

[0097] II. Functions of Each Module 1. Sample Input Module Used to receive cervical tissue sections to be tested, preferably paraffin sections stained with PGC and DNMT1 immunohistochemical staining.

[0098] Sample information includes: Patient ID; Test date; Sample type; Dyeing batch.

[0099] 2. Digital Image Acquisition Module High-resolution full-field images were obtained using a fully automated pathological slide scanner.

[0100] The preferred parameters are as follows: Magnification: 20× or 40×; Resolution: Not less than 0.25 μm / pixel; Image format: SVS, TIFF or NDPI.

[0101] The acquired images are automatically uploaded to the analysis server.

[0102] 3. Image preprocessing module Used to improve image quality and eliminate background interference, including: White balance correction; Color normalization; Background removal; Noise reduction processing; The organization area is automatically segmented.

[0103] This step can improve the accuracy of subsequent identification.

[0104] 4. Marker Recognition Module A convolutional neural network is used to automatically identify positive and negative cells.

[0105] in: Pepsinogen C positive results are characterized by brownish-yellow staining of the cytoplasm; DNMT1 positivity is characterized by brownish-yellow staining of the cell nucleus.

[0106] The system automatically outputs: Number of positive cells; The number of negative cells; The percentage of positive areas; Staining intensity grade.

[0107] III. H-score Calculation The expression level calculation module calculates the H-score of each biomarker according to the following formula: The staining results were semi-quantitatively evaluated, and the staining intensity was graded as follows: 0 (no staining), 1 (weak staining), 2 (medium staining), and 3 (strong staining). Record the proportion of positive cells corresponding to each intensity; The calculation formula is: H-score = Σ (Pi × Ii), where Pi is the percentage of positive cells corresponding to different staining intensities; Ii is the corresponding intensity grade value.

[0108] IV. Risk Assessment Model The risk assessment module has a built-in classification model based on PGC and DNMT1.

[0109] Preferably, a logistic regression model is used.

[0110] V. Results Output The results output module automatically generates a standardized test report, which includes at least the following: PGC H-score; DNMT1 H-score; GAS risk probability; Positive / negative determination; Recommendations.

[0111] Report Support: Export to PDF format; Upload to LIS system; Cloud storage.

[0112] Example 5: Computer-based method for evaluating gastric-type cervical adenocarcinoma This embodiment provides a computer-based method for evaluating gastric-type endocervical adenocarcinoma (GAS). Figure 10 This method can be deployed in digital pathology analysis platforms, hospital laboratory information systems (LIS), cloud-based medical auxiliary diagnostic platforms, or dedicated pathology workstations to automatically identify and assist in the diagnosis of gastric cervical adenocarcinoma based on the expression results of immunohistochemical markers.

[0113] The computer implementation method includes the following steps: S501: Acquire detection data Receive the expression data of each marker in the sample to be tested.

[0114] Preferably, the expression data is derived from digital pathological images of paraffin sections of cervical tissue after immunohistochemical staining.

[0115] More preferably, the marker includes: Pepsinogen C (PGC); DNMT1 (DNA methyltransferase 1).

[0116] The system can automatically read or manually enter the following parameters: Percentage of positive cells; Staining strength; H-score; Patient basic information.

[0117] S502: Data Preprocessing The input data is standardized to eliminate systematic errors between different test batches.

[0118] Preferably, the operation includes the following: Missing values ​​filled; Outlier identification; Z-score standardization; The data format is consistent.

[0119] S503: Feature Extraction Extract key features for classification from the input data.

[0120] Preferably, it includes: PGC's H-score; H-score of DNMT1; Percentage of PGC-positive areas; Percentage of DNMT1 positive cells.

[0121] S504: Risk Probability Calculation The feature vector is input into a pre-trained classification model to calculate the probability that a sample belongs to gastric cervical adenocarcinoma.

[0122] S505: Result Determination The calculated probability value is compared with a preset threshold.

[0123] S606: Generate test report Automatically generate a structured report containing the following: Patient information; PGC test results; DNMT1 test results; GAS risk probability; Judgment conclusion; Clinical recommendations.

[0124] The report can be exported as PDF, HTML, or electronic medical record format.

[0125] Example 6: Computer System for Evaluating Gastric Type Cervical Adenocarcinoma This embodiment provides a computer system for evaluating gastric-type endocervical adenocarcinoma (GAS). Based on the expression information of immunohistochemical markers in cervical tissue, this system can automatically complete data acquisition, feature analysis, risk assessment, and diagnostic result output. It can be widely used in digital pathology platforms, hospital laboratory information systems (LIS), pathology workstations, and cloud-based assisted diagnostic platforms.

[0126] I. System Overall Structure The computer system includes at least: 1. Processor; 2. Memory; 3. Input interface; 4. Output interface; 5. Communication interface; 6. Display device.

[0127] The components are electrically connected via a system bus.

[0128] Preferably, the processor can be a central processing unit (CPU), a graphics processing unit (GPU), or a combination of both.

[0129] The memory includes: Volatile memory; Non-volatile memory.

[0130] For example, random access memory (RAM), solid-state drives (SSDs), and read-only memory (ROM).

[0131] II. Program modules stored in memory The memory stores a computer program, which, when executed by a processor, implements the following functional modules: (1) Data acquisition module Used to receive detection data from the sample to be tested.

[0132] Preferably, the detection data includes: Expression value of Pepsinogen C; Expression value of DNMT1; The corresponding H-score; Digital pathology images.

[0133] (2) Preprocessing module Used for preprocessing raw data, including: Missing value correction; Outlier removal; Standardized processing; Format conversion.

[0134] (3) Feature extraction module Feature parameters used to extract features for classification include: PGC H-score; DNMT1 H-score; Percentage of PGC-positive cells; Percentage of DNMT1 positive cells.

[0135] (4) Risk assessment module Used to invoke a pre-trained classification model to assess the risk of gastric cervical adenocarcinoma.

[0136] Preferably, the model is a logistic regression model, a random forest model, or a support vector machine model.

[0137] (5) Result Output Module Used to output evaluation results.

[0138] The results include at least: Risk probability of gastric type cervical adenocarcinoma; Positive / negative determination; Recommended clinical opinion.

[0139] Output methods include: The display screen shows; Export PDF report; Upload to LIS system; Cloud-based database storage.

[0140] III. System Operation Flow When the system is running, perform the following steps: 1. Receive input data; 2. Automatically complete data preprocessing; 3. Extract key features; 4. Call the classification model to calculate the risk probability; 5. Output diagnostic results.

[0141] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0142] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An assessment device for evaluating gastric-type cervical adenocarcinoma, characterized in that, include: A data acquisition device is used to acquire detection data of the evaluation subject, wherein the detection data is the detection value of gastric cervical adenocarcinoma markers obtained from human cervical tissue of the evaluation subject suspected of having gastric cervical adenocarcinoma, which is the staining intensity of immunohistochemistry of the human cervical tissue and the percentage of positive cells of each staining intensity among all evaluation cells, and the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1. The data processing equipment is used to calculate the level value of the marker in the gastric type cervical adenocarcinoma marker based on the test value of the evaluation object, and then calculate whether the level value of the obtained marker is within the risk range of the corresponding marker.

2. The evaluation device as described in claim 1, characterized in that: The level value of the biomarker is the immunohistochemical (IHC) score of the biomarker in the cervical tissue of the human subject being evaluated, and the IHC score is calculated based on staining intensity and percentage of positive cells; in, The IHC score is called H-score, which is calculated as follows: H-score = Σ (Pi × Ii), where Pi represents the percentage of positive cells with staining intensity level i among all evaluated cells, and Ii represents the staining intensity grading value, with i taking values ​​of 0, 1, 2, and 3, corresponding to no staining, weak positive, moderate positive, and strong positive, respectively. If the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1, then when the IHC score of PGC is higher than the first preset threshold and the IHC score of DNMT1 is lower than the second preset threshold, a high-risk assessment result for gastric cervical adenocarcinoma will be given. If the IHC score of PGC is lower than the first preset threshold and / or the IHC score of DNMT1 is higher than the second preset threshold, a low-risk assessment result for the corresponding gastric type cervical adenocarcinoma will be given.

3. The evaluation device as described in claim 1, characterized in that, The data processing device has a built-in trained machine learning classifier for calculating the risk assessment results of gastric cervical adenocarcinoma based on the level values ​​of the markers; the machine learning classifier is trained based on an immunohistochemical scoring cohort including patients with gastric cervical adenocarcinoma and patients with human papillomavirus-associated cervical adenocarcinoma.

4. The evaluation device as described in claim 1 or 2, characterized in that, The data acquisition device is an input device for inputting data or a communication device for reading data from an external data storage device or the storage device of the detection device via an interface.

5. The evaluation device as described in claim 4, characterized in that, When communication equipment is used, its corresponding external data storage device is the data storage device on the device used to measure biomarkers.

6. The evaluation device as described in claim 4, characterized in that, It is integrated into automated biomarker assay equipment, thereby directly outputting evaluation results after assay, realizing integrated detection and evaluation functions.

7. A computer-executed method for outputting risk assessment results for gastric-type cervical adenocarcinoma, the method comprising the following steps: a) Read the test data of the patient or subject, wherein the test data is the test value of the gastric cervical adenocarcinoma marker obtained from the human cervical tissue of the subject suspected of having gastric cervical adenocarcinoma, which is the staining intensity of the immunohistochemistry of the human cervical tissue and the percentage of positive cells of each staining intensity to all the assessment cells, and the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1. b) Data processing: The level value of the biomarker is calculated based on the detection value in step a). The level value of the biomarker is the immunohistochemical (IHC) score of the biomarker in the cervical tissue of the human subject being evaluated. The IHC score is calculated based on staining intensity and percentage of positive cells. in, The IHC score is called H-score, which is calculated as follows: H-score = Σ (Pi × Ii), where Pi represents the percentage of positive cells with staining intensity level i among all evaluated cells, and Ii represents the staining intensity grading value, with i taking values ​​of 0, 1, 2, and 3, corresponding to no staining, weak positive, moderate positive, and strong positive, respectively. c) Output the evaluation results.

8. The execution method of claim 7, wherein the data processing in step b) is performed by a trained machine learning model trained on an immunohistochemical scoring cohort comprising patients with gastric cervical adenocarcinoma and patients with human papillomavirus-associated cervical adenocarcinoma.

9. The execution method as described in claim 7, wherein in step c), if the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1, then when the IHC score of PGC is higher than a first preset threshold and the IHC score of DNMT1 is lower than a second preset threshold, a high-risk assessment result for gastric cervical adenocarcinoma is given; if the IHC score of PGC is lower than the first preset threshold and / or the IHC score of DNMT1 is higher than the second preset threshold, a low-risk assessment result for gastric cervical adenocarcinoma is given.

10. A computer system, characterized in that, Configured to perform the following operations: receive the immunohistochemical (IHC) score of gastric cervical adenocarcinoma markers in the cervical tissue of the assessment subject, wherein the gastric cervical adenocarcinoma markers are a combination of PGC and DNMT1, two markers; wherein the IHC score is the H-score, which is calculated by the formula: H-score = Σ (Pi × Ii), where Pi represents the percentage of positive cells with staining intensity level i among all assessment cells, and Ii represents the staining intensity grading value, where i takes values ​​of 0, 1, 2, and 3, corresponding to no staining, weak positive, moderate positive, and strong positive, respectively; Based on the IHC score, a risk assessment result for gastric cervical adenocarcinoma is generated, and a risk warning is optionally given. If the gastric cervical adenocarcinoma marker is a combination of PGC and DNMT1, then when the IHC score of PGC is higher than a first preset threshold and the IHC score of DNMT1 is lower than a second preset threshold, a high-risk assessment result for gastric cervical adenocarcinoma is given; if the IHC score of PGC is lower than the first preset threshold and / or the IHC score of DNMT1 is higher than the second preset threshold, a low-risk assessment result for gastric cervical adenocarcinoma is given.

Images