Intelligent cervical cancer cell detection method based on liquid-based thin-layer cell detection technology

Abstract

According to the intelligent cervical cancer cell detection method based on the liquid-based thin-layer cell detection technology, high-resolution imaging, deep learning, a neural network and Bayesian network decision support are combined, and accurate recognition and risk assessment of cervical cancer cells are achieved. Through automatic sample processing and multispectral image capturing and preprocessing, cell morphology and spectral features are extracted; analyzing a spatial relationship between cells by using a convolutional neural network based on an attention mechanism and a graph neural network, and identifying an abnormal mode; and carrying out risk grading through a Bayesian network in combination with clinical information of the patient. Meanwhile, the invention further relates to a memory, a computer and a server which are used for the method, and data safety, efficient processing and continuous model optimization are ensured. According to the method, the accuracy and efficiency of cervical cancer detection are improved, personalized treatment suggestions are provided for doctors, and the method has important clinical application value.

Description

Technical Field

[0001] The invention belongs to the field of cancer cell detection, and specifically discloses an intelligent detection method for cervical cancer cells based on liquid-based thin-layer cell detection technology. Background Art

[0002] Cervical cancer is one of the most common malignant tumors in women worldwide. Early detection and treatment are crucial to improving the survival rate of patients. At present, liquid-based thin-layer cell detection technology (such as ThinPrep or SurePath, etc.) has become the main means of cervical cancer screening. This technology can effectively detect early cancerous cells by collecting cervical cell samples and performing thin-layer smears, combined with microscopic observation and diagnosis by cytologists.

[0003] However, existing technologies have many limitations. First, the diagnosis of cytologists is highly dependent on personal experience and subjective judgment, and is easily affected by factors such as fatigue and distraction, which makes it difficult to ensure the accuracy and consistency of the diagnostic results. Secondly, the traditional manual film reading method is inefficient and difficult to meet the needs of large-scale screening. In addition, the sensitivity of manual diagnosis is limited for the identification of subtle changes in cell morphological characteristics or early lesions, and some potential high-risk cases may be missed.

[0004] With the rapid development of artificial intelligence and deep learning technology, although some automated auxiliary diagnosis systems based on image processing have been proposed, these systems often focus on the analysis of single cell features, ignoring the spatial relationship and interaction between cells, as well as the integration of patient clinical information, resulting in limited detection accuracy and robustness. At the same time, data security, processing efficiency and continuous model optimization capabilities are also challenges facing current technologies.

[0005] Therefore, there is an urgent need for an intelligent detection method that can comprehensively consider cell morphological characteristics, spectral characteristics, spatial relationships between cells, and patient clinical information to improve the accuracy, efficiency, and personalization level of cervical cancer detection. Summary of the invention

[0006] To solve the above problems, we disclose an intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology.

[0007] The present invention includes the following specific technical solutions: An intelligent detection method for cervical cancer cells based on liquid-based thin-layer cell detection technology, the method comprising: S1 collects liquid-based thin-layer cell samples and obtains multispectral images of cells using high-resolution imaging technology; S2 uses a deep learning algorithm, including a convolutional neural network based on an attention mechanism, to extract features from the multispectral image, wherein the network is specially trained to identify cell morphological features and spectral features associated with cervical cancer, including but not limited to the size, shape, chromatin distribution of the cell nucleus, and fluorescence intensity at a specific wavelength; S3 combines the spatial relationship information between cells and further analyzes the extracted features through a neural network to identify abnormal aggregation patterns or tissue structure changes between cells; Based on the above analysis, S4 generates a preliminary test report containing suspected cervical cancer cells, and applies a Bayesian network-based decision support system that integrates the patient's age and medical history clinical information to perform a probability assessment of the test results and provide personalized risk grading.

[0008] Furthermore, the above-mentioned intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology also includes: The S5 feedback learning mechanism continuously optimizes the deep learning model according to the clinical diagnosis results, and improves the detection accuracy and adaptability through transfer learning and online learning technology.

[0009] Furthermore, the above-mentioned intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology, The step S1 comprises the following specific steps: S11 applies an automated sample processing flow, including sample dilution, centrifugation, and sedimentation steps, so that cells are evenly distributed and the background is clear; S12 uses multispectral imaging technology to capture cell images under multiple preset spectral bands, including but not limited to visible light, ultraviolet light, and specific fluorescent excitation light; S13 implements image preprocessing algorithms, including noise filtering, contrast enhancement, and image smoothing, to improve image quality and facilitate subsequent feature extraction.

[0010] Furthermore, in the above-mentioned intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology, step S2 comprises the following specific steps: S21 builds a large-scale annotated image database containing samples of cervical cancer cells and non-cancerous cells for model training; S22 designed and trained a convolutional neural network model based on the attention mechanism, which can automatically focus on the key cell areas in the image and extract morphological and spectral features related to cervical cancer; S23 applies data enhancement strategies, including image rotation, scaling, and flipping, to increase the diversity of training data and improve the generalization ability of the model; S24 uses cross-validation method to evaluate model performance, and adjusts model structure and parameters according to the evaluation results to optimize the detection effect.

[0011] Furthermore, the above-mentioned intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology, The step S3 comprises the following specific steps: S31 constructs a spatial relationship graph between cells, treating each cell as a node in the graph and the distances or interactions between cells as edges between nodes; S32 applies graph neural network algorithm to encode spatial relationship graph and learn the spatial layout and correlation patterns between cells; S33 designs a graph convolution operation to update the feature representation of each node by aggregating the feature information of adjacent nodes, thereby capturing abnormal aggregation or tissue structure changes between cells; S34 fuses the features output by the graph neural network with the cell features extracted by the convolutional neural network to improve the accuracy and robustness of detection.

[0012] Furthermore, the above-mentioned intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology, The step S4 comprises the following specific steps: S41 determines the location and number of suspected cervical cancer cells based on the output of the deep learning model and generates a preliminary report containing cell images and test results; S42 builds a Bayesian network model of clinical data containing patient age, medical history, and family genetic information; S43 combines the cell features in the preliminary test report with the clinical data in the Bayesian network model to perform a joint probability analysis to assess the patient's risk level for cervical cancer; S44 provides doctors with personalized treatment recommendations or guidance for further examinations based on risk stratification results.

[0013] Furthermore, the above-mentioned intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology, The step S5 comprises the following specific steps: S51 Establish a database of clinical diagnosis results to store cervical cancer cell detection cases verified by pathology gold standard; S52 regularly compares clinical diagnosis results with the prediction results of the deep learning model to identify misdiagnosis or missed diagnosis in the model; S53 applies transfer learning and online learning techniques to fine-tune or retrain the deep learning model based on the comparison results to correct errors and improve model performance; S54 monitors the changing trends of model performance and adjusts learning strategies and parameter settings so that the model can continuously adapt to new clinical data and medical research progress.

[0014] The present invention also discloses a memory for a cervical cancer cell intelligent detection system based on a liquid-based thin layer cell detection technology, wherein the memory is configured as follows: 1) Store massive amounts of cell image data, model training data, clinical data, and test result data, and use a distributed storage architecture to ensure data security and reliability; 2) Design an efficient data access and retrieval mechanism to support fast data reading and writing during deep learning model training and real-time detection; 3) Includes data backup and recovery functions to quickly restore data integrity in the event of hardware failure or data loss; 4) Integrate data encryption and secure access control modules to protect patient privacy and data security, enabling compliance.

[0015] The present invention also discloses a computer for executing a cervical cancer cell intelligent detection method based on a liquid-based thin layer cell detection technology, the computer comprising: 1) High-performance processing units, including GPUs or TPUs, to accelerate the training and inference tasks of deep learning models; 2) Large-capacity memory for storing cell image database, model training data, intermediate calculation results and final detection model; 3) High-speed network interface, enabling efficient communication between computers and external devices or server clusters; 4) Specialized software systems, including deep learning frameworks, image processing algorithm libraries, Bayesian network decision support systems, and algorithm modules for data preprocessing, feature extraction, spatial relationship analysis, risk grading, and feedback learning.

[0016] The present invention also discloses a server for supporting intelligent detection of cervical cancer cells based on liquid-based thin layer cell detection technology, the server comprising: 1) At least one high-performance computing node equipped with a high-performance processor and sufficient memory to perform training and inference of deep learning models; 2) Large-capacity storage systems, using distributed or centralized architectures, for secure and reliable storage of cell image data, clinical data, and model files; 3) High-speed network interconnection equipment, enabling low-latency, high-bandwidth communication between components within the server and between the server and external clients; 4) System management software, responsible for resource scheduling, load balancing, fault recovery and data security management, so that the server can run efficiently and stably; 5) Specialized software services provide functions such as image processing, feature extraction, spatial relationship analysis, risk grading and feedback learning to support intelligent detection of cervical cancer cells.

[0017] Compared with the prior art, the present invention has the following outstanding beneficial effects: 1. Improve detection accuracy: By combining high-resolution imaging technology and deep learning algorithms, especially convolutional neural networks based on attention mechanisms, the present invention can automatically focus on key cell areas in the image, accurately extract morphological and spectral features related to cervical cancer, significantly improve the ability to identify cancer cells, and reduce the rates of misjudgment and missed judgments.

[0018] 2. Enhanced detection robustness: By introducing graph neural networks to analyze the spatial relationship between cells, the present invention can capture abnormal aggregation patterns or tissue structure changes between cells. This information is crucial for the identification of early cancer, thereby enhancing the robustness and sensitivity of detection.

[0019] 3. Personalized risk assessment: This invention is not only based on cell image features, but also integrates clinical information such as the patient's age and medical history. It performs joint probability analysis through the Bayesian network to provide patients with personalized cervical cancer risk grading, which helps doctors develop more accurate treatment plans.

[0020] 4. Improve detection efficiency: The automated sample processing process, efficient image processing algorithm and fast model reasoning capability enable the present invention to complete the detection of a large number of samples in a short period of time, significantly improving the detection efficiency and meeting the needs of large-scale screening.

[0021] 5. Continuous learning and optimization: Through the feedback learning mechanism, the present invention can continuously optimize the deep learning model according to the clinical diagnosis results, ensuring that the model always remains in the best state, adapting to new clinical data and medical research progress, and improving the long-term availability and accuracy of the system.

[0022] 6. Ensure data security and privacy: The present invention adopts advanced encryption and secure access control technology during data storage, access and transmission, ensuring the security and privacy protection of patient data and meeting the compliance requirements of medical data. DETAILED DESCRIPTION

[0023] An intelligent detection method for cervical cancer cells based on liquid-based thin-layer cell detection technology, the method comprising: S1 collects liquid-based thin-layer cell samples and obtains multispectral images of cells using high-resolution imaging technology; The specific steps include: S11 applies an automated sample processing flow, including sample dilution, centrifugation, and sedimentation steps, so that cells are evenly distributed and the background is clear; S12 uses multispectral imaging technology to capture cell images under multiple preset spectral bands, including but not limited to visible light, ultraviolet light, and specific fluorescent excitation light; S13 implements image preprocessing algorithms, including noise filtering, contrast enhancement, and image smoothing, to improve image quality and facilitate subsequent feature extraction.

[0024] S2 uses a deep learning algorithm, including a convolutional neural network based on an attention mechanism, to extract features from the multispectral image, wherein the network is specially trained to identify cell morphological features and spectral features associated with cervical cancer, including but not limited to the size, shape, chromatin distribution of the cell nucleus, and fluorescence intensity at a specific wavelength; The specific steps include: S21 builds a large-scale annotated image database containing samples of cervical cancer cells and non-cancerous cells for model training; S22 designed and trained a convolutional neural network model based on the attention mechanism, which can automatically focus on the key cell areas in the image and extract morphological and spectral features related to cervical cancer; S23 applies data enhancement strategies, including image rotation, scaling, and flipping, to increase the diversity of training data and improve the generalization ability of the model; S24 uses cross-validation method to evaluate model performance, and adjusts model structure and parameters according to the evaluation results to optimize the detection effect.

[0025] S3 combines the spatial relationship information between cells and further analyzes the extracted features through a neural network to identify abnormal aggregation patterns or tissue structure changes between cells; The specific steps include: S31 constructs a spatial relationship graph between cells, treating each cell as a node in the graph and the distances or interactions between cells as edges between nodes; S32 applies graph neural network algorithm to encode spatial relationship graph and learn the spatial layout and correlation patterns between cells; S33 designs a graph convolution operation to update the feature representation of each node by aggregating the feature information of adjacent nodes, thereby capturing abnormal aggregation or tissue structure changes between cells; S34 fuses the features output by the graph neural network with the cell features extracted by the convolutional neural network to improve the accuracy and robustness of detection.

[0026] S4 generates a preliminary test report containing suspected cervical cancer cells based on the above analysis, and applies a Bayesian network-based decision support system to integrate the patient's age and medical history clinical information to conduct a probability assessment of the test results and provide personalized risk grading; The step S4 comprises the following specific steps: S41 determines the location and number of suspected cervical cancer cells based on the output of the deep learning model and generates a preliminary report containing cell images and test results; S42 builds a Bayesian network model of clinical data containing patient age, medical history, and family genetic information; S43 combines the cell features in the preliminary test report with the clinical data in the Bayesian network model to perform a joint probability analysis to assess the patient's risk level for cervical cancer; S44 provides doctors with personalized treatment recommendations or guidance for further examinations based on risk stratification results.

[0027] S5 feedback learning mechanism, which continuously optimizes the deep learning model according to the clinical diagnosis results, and improves the detection accuracy and adaptability through transfer learning and online learning technology; The specific steps include: S51 Establish a database of clinical diagnosis results to store cervical cancer cell detection cases verified by pathology gold standard; S52 regularly compares clinical diagnosis results with the prediction results of the deep learning model to identify misdiagnosis or missed diagnosis in the model; S53 applies transfer learning and online learning techniques to fine-tune or retrain the deep learning model based on the comparison results to correct errors and improve model performance; S54 monitors the changing trends of model performance and adjusts learning strategies and parameter settings so that the model can continuously adapt to new clinical data and medical research progress.

[0028] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Example 1 Intelligent detection of cervical cancer cells based on liquid-based thin-layer cell detection S1: Collect liquid-based thin-layer cell samples and obtain multispectral images; S11: The sample processing process uses automated equipment. First, the collected cervical cell sample is diluted with preservation solution at a ratio of 1:10, and then centrifuged at 1500rpm for 5 minutes. After removing the supernatant, the preservation solution is added again and the centrifugation is repeated. Finally, the cells are deposited on a glass slide to form a thin layer of cell sample.

[0030] S12: Use a multispectral imaging system to capture cell images at 405nm (ultraviolet light), 488nm (blue light), 532nm (green light), 633nm (red light), and specific fluorescent excitation light (such as 358nm for DAPI). The resolution of the image in each band is 4096x4096 pixels.

[0031] S13: Apply Gaussian filter to remove noise, use histogram equalization to enhance contrast, and perform image smoothing through median filtering to prepare for subsequent feature extraction.

[0032] S2: Deep Learning Feature Extraction S21: Build a database of 5,000 annotated images, of which 2,500 are cervical cancer cell samples and 2,500 are non-cancerous cell samples. Each image is annotated with characteristics such as cell nucleus size, shape, and chromatin distribution.

[0033] S22: Design a convolutional neural network (Attention-CNN) based on the attention mechanism. The network consists of 10 convolutional layers, each followed by a ReLU activation function and a maximum pooling layer. The last two layers are fully connected layers, and the output is a vector representation of cell features. The attention mechanism automatically focuses on key cell areas by calculating the importance weight of each position.

[0034] S23: The data augmentation strategy includes random rotation (0-360 degrees), scaling (0.8-1.2 times), and horizontal flipping. Each transformation generates 5 new images to increase the diversity of training data.

[0035] S24: 5-fold cross validation was used to evaluate the model performance, using accuracy, precision, recall and F1 score as evaluation indicators. Based on the validation results, the network structure and learning rate were adjusted, and the F1 score of the final model on the test set reached 95%.

[0036] S3: Analysis of spatial relationships between cells S31: Construct a cell spatial relationship graph, where nodes represent cells and edges represent relationships between cells with a Euclidean distance less than 50 pixels.

[0037] S32: Apply graph convolutional network (GCN) to encode the spatial relationship graph. Each layer of GCN updates the node features by aggregating the features of neighboring nodes. The number of layers is 3.

[0038] S33: Design a graph convolution operation, use ReLU as the activation function, and update the feature representation of each node by aggregating the feature information of adjacent nodes to capture abnormal aggregation patterns between cells.

[0039] S34: The features output by GCN are fused with the cell features extracted by Attention-CNN by splicing, and the fully connected layer is used for classification to improve the accuracy and robustness of detection.

[0040] S4: Generate preliminary test report and risk classification S41: Based on the fusion results of Attention-CNN and GCN, the location and number of suspected cervical cancer cells are determined, and a preliminary report containing cell images and detection results is generated.

[0041] S42: Construct a Bayesian network model, including nodes such as patient age (X1), medical history (X2, such as whether HPV is infected), family genetic information (X3, such as whether there is a family history of cervical cancer), and the probabilistic relationship between these nodes and cervical cancer risk (Y).

[0042] S43: Combine the cell characteristics (such as the number of abnormal cells) in the preliminary test report with the clinical data in the Bayesian network model, and use the Bayesian formula to calculate the posterior probability of the patient having cervical cancer for risk stratification.

[0043] S44: Provide personalized treatment recommendations to doctors based on risk stratification results, such as recommending regular follow-up for low-risk patients and further pathological examination for high-risk patients.

[0044] S5: Feedback Learning Mechanism S51: Establish a clinical diagnosis result database, including 1000 cases of cervical cancer cell detection verified by pathology gold standard, of which 500 are positive and 500 are negative.

[0045] S52: Compare the clinical diagnosis results with the model prediction results every quarter and calculate the misdiagnosis rate and missed diagnosis rate.

[0046] S53: Apply transfer learning technology and use the newly collected case data to fine-tune the model. Adjust the learning rate to 0.001 and iterate 10 rounds to correct misjudgments and missed judgments and improve model performance.

[0047] S54: Monitor the performance changes of the model on the validation set. If the accuracy drops by more than 2% continuously, trigger the retraining process and update the model with the latest clinical data and medical research results to ensure the continuous adaptability of the model.

[0048] Through the above specific embodiments, the intelligent detection method of cervical cancer cells based on liquid-based thin layer cell detection technology of the present invention can achieve efficient and accurate detection of cervical cancer cells, and at the same time provide personalized risk grading in combination with the patient's clinical information, providing strong support for doctors' diagnosis and treatment. Example 2 Enhanced version of the intelligent identification system for cervical cancer cells based on liquid-based thin layer cell detection.

[0049] S1: Sample collection and multispectral image acquisition S11: Sample processing uses a semi-automated process. First, a cervical brush is used to collect cell samples, which are then mixed with a special preservation solution at a ratio of 1:5. The samples are centrifuged at 1200rpm for 8 minutes in an automatic centrifuge, and the supernatant is removed. This process is repeated twice to ensure cell cleanliness. Finally, an automatic deposition device is used to evenly deposit the cells on a special slide to form a thin layer of cell samples.

[0050] S12: Using advanced multispectral imaging system, four key spectral bands are selected: 450nm (blue light), 546nm (green light), 633nm (red light) and DAPI fluorescence excitation light (365nm). The resolution of the image in each band is 5120x5120 pixels to ensure high resolution and detail capture.

[0051] S13: Image preprocessing includes using an adaptive median filter to remove noise, applying CLAHE (contrast limited adaptive histogram equalization) to enhance image contrast, and using a bilateral filter for image smoothing to preserve edge details.

[0052] S2: Deep Learning Feature Extraction and Classification S21: Build a database of 8,000 images, of which 4,000 are images of cervical cancer cells and 4,000 are images of normal or benign lesion cells. Each image is annotated by experts, including the size and shape of the cell nucleus, chromatin distribution, nuclear-cytoplasmic ratio, and fluorescence intensity under a specific spectrum.

[0053] S22: Design a deep convolutional neural network (Deep CNN) model, combining the attention mechanism (AttentionMechanism) and the residual network (ResNet) structure to improve feature extraction capabilities and model depth. The model contains 15 convolutional layers, each followed by a ReLU activation function and a batch normalization layer, and 4 residual blocks to reduce the gradient vanishing problem during training.

[0054] S23: Data augmentation strategies include random cropping (cropping to 80%-100% of the original size), random rotation (0-360 degrees), random color jitter (brightness, contrast, saturation changes), and horizontal and vertical flipping. Each transformation generates 3 new images to enrich the training dataset.

[0055] S24: K-fold cross validation (K=10) was used to evaluate the model performance, using AUC (area under the curve), accuracy, precision, recall, and F1 score as evaluation indicators. Based on the validation results, the model hyperparameters were adjusted using grid search and random search methods, and the final model achieved an AUC of 0.98 and an F1 score of 96% on the test set.

[0056] S3: Spatial relationships between cells and recognition of abnormal patterns S31: Construct a cell spatial relationship graph and use the Delaunay triangulation method to determine the spatial proximity relationship between cells. Each cell is regarded as a node in the graph, and the edges between nodes indicate that the spatial distance between cells is less than a certain threshold (such as 75 pixels).

[0057] S32: Apply Graph Attention Network (GAT) to encode the spatial relationship graph and learn the spatial layout and correlation patterns between cells through a multi-head attention mechanism. The number of layers is 4 and the number of heads in each layer is 8.

[0058] S33: Design a graph convolution operation, use ELU (exponential linear unit) as the activation function, and update the feature representation of each node by aggregating the feature information of adjacent nodes to capture abnormal aggregation, irregular arrangement or tissue structure changes between cells.

[0059] S34: The features output by GAT are fused with the cell features extracted by Deep CNN through a feature fusion network (FFN). FFN contains two fully connected layers, uses the ReLU activation function, and finally performs classification through a Softmax layer to improve the accuracy and robustness of detection.

[0060] S4: Preliminary test report generation and risk assessment S41: Based on the fusion results of Deep CNN and GAT, a preliminary detection report is generated containing the location, number, type (such as squamous cell carcinoma, adenocarcinoma, etc.) and confidence level of suspected cervical cancer cells.

[0061] S42: Construct a Bayesian network model that includes clinical data such as patient age, medical history (including history of HPV infection, previous history of cervical disease), reproductive history, smoking history, family genetic history, and the probabilistic relationship between these data and the risk of cervical cancer.

[0062] S43: Combine the cell features in the preliminary test report with the clinical data in the Bayesian network model, use the Bayesian inference algorithm to calculate the posterior probability of the patient having cervical cancer, and divide the patients into three groups: low risk, medium risk, and high risk according to the posterior probability.

[0063] S44: Based on the risk stratification results, provide doctors with personalized treatment recommendations or further examination plans, such as regular screening for low-risk patients, more detailed cytological examination for medium-risk patients, and direct histological biopsy for high-risk patients.

[0064] S5: Feedback Learning and Model Optimization S51: Establish a clinical diagnosis result database and continuously collect cervical cancer cell detection cases verified by the gold standard of pathology, which currently contains 1,500 cases, of which 750 are positive and 750 are negative.

[0065] S52: Compare the clinical diagnosis results with the model prediction results every month, calculate the confusion matrix, and analyze the reasons for misdiagnosis and missed diagnosis.

[0066] S53: Apply online learning technology to incrementally learn the model using newly collected case data, with a learning rate of 0.0001 and the number of iterations dynamically adjusted according to the amount of data to continuously improve model performance. At the same time, apply transfer learning technology to transfer the model's knowledge on similar tasks (such as other types of cancer cell detection) to the cervical cancer cell detection task to improve the generalization ability of the model.

[0067] S54: Monitor the performance changes of the model on the validation set and test set. If the accuracy, AUC and other indicators continuously drop by more than the threshold (such as 1%), the model retraining process will be triggered. During retraining, the training data set is updated with the latest clinical data and medical research results, and the model structure and hyperparameters are adjusted to ensure the continuous adaptability and accuracy of the model.

[0068] Through the above embodiments, the enhanced version of the cervical cancer cell intelligent identification system based on liquid-based thin layer cell detection of the present invention can realize efficient and accurate identification of cervical cancer cells, and at the same time provide personalized risk assessment and treatment recommendations based on the patient's clinical information, providing strong support for doctors' clinical decision-making.

[0069] The above are only a few preferred embodiments of the present invention, which are described in a more specific and detailed manner, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. An intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology, characterized in that: The method includes: S1 collects liquid-based thin-layer cell samples and obtains multispectral images of cells using high-resolution imaging technology; S2 uses a deep learning algorithm, including a convolutional neural network based on an attention mechanism, to extract features from the multispectral image, wherein the network is specially trained to identify cell morphological features and spectral features associated with cervical cancer, including but not limited to the size, shape, chromatin distribution of the cell nucleus, and fluorescence intensity at a specific wavelength; S3 combines the spatial relationship information between cells and further analyzes the extracted features through a neural network to identify abnormal aggregation patterns or tissue structure changes between cells; Based on the above analysis, S4 generates a preliminary test report containing suspected cervical cancer cells, and applies a Bayesian network-based decision support system that integrates the patient's age and medical history clinical information to perform a probability assessment of the test results and provide personalized risk grading.

2. According to claim 1, a method for intelligent detection of cervical cancer cells based on liquid-based thin layer cell detection technology is characterized in that: The method further comprises: The S5 feedback learning mechanism continuously optimizes the deep learning model according to the clinical diagnosis results, and improves the detection accuracy and adaptability through transfer learning and online learning technologies.

3. According to claim 1, a method for intelligent detection of cervical cancer cells based on liquid-based thin layer cell detection technology is characterized in that: The step S1 comprises the following specific steps: S11 applies an automated sample processing flow, including sample dilution, centrifugation, and sedimentation steps, so that cells are evenly distributed and the background is clear; S12 uses multispectral imaging technology to capture cell images in multiple preset spectral bands, including but not limited to visible light, ultraviolet light, and specific fluorescent excitation light; S13 implements image preprocessing algorithms, including noise filtering, contrast enhancement and image smoothing, to improve image quality and facilitate subsequent feature extraction.

4. The intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology according to claim 1 is characterized in that: The step S2 comprises the following specific steps: S21 builds a large-scale annotated image database containing samples of cervical cancer cells and non-cancerous cells for model training; S22 designed and trained a convolutional neural network model based on the attention mechanism, which can automatically focus on the key cell areas in the image and extract morphological and spectral features related to cervical cancer; S23 applies data enhancement strategies, including image rotation, scaling, and flipping, to increase the diversity of training data and improve the generalization ability of the model; S24 uses cross-validation method to evaluate model performance, and adjusts model structure and parameters according to the evaluation results to optimize the detection effect.

5. The intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology according to claim 1 is characterized in that: The step S3 comprises the following specific steps: S31 constructs a spatial relationship graph between cells, treating each cell as a node in the graph and the distances or interactions between cells as edges between nodes; S32 applies a graph neural network algorithm to encode the spatial relationship graph and learn the spatial layout and correlation patterns between cells; S33 designs a graph convolution operation to update the feature representation of each node by aggregating the feature information of adjacent nodes, thereby capturing abnormal aggregation or tissue structure changes between cells; S34 fuses the features output by the graph neural network with the cell features extracted by the convolutional neural network to improve the accuracy and robustness of detection.

6. The intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology according to claim 1 is characterized in that: The step S4 comprises the following specific steps: S41 determines the location and number of suspected cervical cancer cells based on the output of the deep learning model and generates a preliminary report containing cell images and test results; S42 builds a Bayesian network model of clinical data containing patient age, medical history, and family genetic information; S43 combines the cell features in the preliminary test report with the clinical data in the Bayesian network model to perform a joint probability analysis to assess the patient's risk level for cervical cancer; S44 provides doctors with personalized treatment recommendations or guidance for further examinations based on risk stratification results.

7. The intelligent detection method for cervical cancer cells based on liquid-based thin layer cell detection technology according to claim 2 is characterized in that: The step S5 comprises the following specific steps: S51 Establish a database of clinical diagnosis results to store cervical cancer cell detection cases verified by pathology gold standard; S52 regularly compares clinical diagnosis results with the prediction results of the deep learning model to identify misdiagnosis or missed diagnosis in the model; S53 applies transfer learning and online learning techniques to fine-tune or retrain the deep learning model based on the comparison results to correct errors and improve model performance; S54 monitors the changing trends of model performance and adjusts learning strategies and parameter settings so that the model can continuously adapt to new clinical data and medical research progress.

8. A memory for a cervical cancer cell intelligent detection system based on liquid-based thin layer cell detection technology, characterized in that: The memory is configured as: 1) Store massive amounts of cell image data, model training data, clinical data, and test result data, and use a distributed storage architecture to ensure data security and reliability; 2) Design an efficient data access and retrieval mechanism to support fast data reading and writing during deep learning model training and real-time detection; 3) Includes data backup and recovery functions to quickly restore data integrity in the event of hardware failure or data loss; 4) Integrate data encryption and secure access control modules to protect patient privacy and data security, enabling compliance.

9. A computer for executing an intelligent detection method for cervical cancer cells based on liquid-based thin-layer cell detection technology, characterized in that: The computer includes: 1) High-performance processing units, including GPUs or TPUs, to accelerate the training and inference tasks of deep learning models; 2) Large-capacity memory for storing cell image database, model training data, intermediate calculation results and final detection model; 3) High-speed network interface, enabling efficient communication between computers and external devices or server clusters; 4) Specialized software systems, including deep learning frameworks, image processing algorithm libraries, Bayesian network decision support systems, and algorithm modules for data preprocessing, feature extraction, spatial relationship analysis, risk grading, and feedback learning.

10. A server for supporting intelligent detection of cervical cancer cells based on liquid-based thin layer cell detection technology, characterized in that: The server includes: 1) At least one high-performance computing node equipped with a high-performance processor and sufficient memory to perform training and inference of deep learning models; 2) Large-capacity storage systems, using distributed or centralized architectures, for secure and reliable storage of cell image data, clinical data, and model files; 3) High-speed network interconnection equipment, enabling low-latency, high-bandwidth communication between components within the server and between the server and external clients; 4) System management software, responsible for resource scheduling, load balancing, fault recovery and data security management, so that the server can run efficiently and stably; 5) Specialized software services provide functions such as image processing, feature extraction, spatial relationship analysis, risk grading and feedback learning to support intelligent detection of cervical cancer cells.