Intelligent cervical cancer screening method aiming at multi-mode traditional Chinese and western medicine combined data

Abstract

The invention relates to an intelligent cervical cancer screening method for multi-mode traditional Chinese medicine and western medicine combined data, which comprises the following steps: S1, data integration: respectively collecting text data in traditional Chinese medicine and western medicine diagnosis and treatment processes, and combining the two kinds of data to form a comprehensive patient health text file; s2, carrying out fine tuning training on the obtained traditional Chinese and western medicine combined data, and carrying out preliminary screening and diagnosis on cervical cancer; s3, based on an image segmentation and classification model, performing effective segmentation on the image of the cervical cancer, giving a model recommendation category, and promoting a doctor to re-judge the cancer condition; and S4, based on output results of the text and the image data, giving a related treatment scheme. According to the method, automatic classification and focus segmentation auxiliary diagnosis of cervical cancer images are realized, the medical history of a patient, instructions given by a doctor and the like are integrated, a comprehensive diagnosis platform is constructed, preliminary diagnosis suggestions are provided, a personalized treatment scheme is generated according to the specific condition of the patient, and a treatment plan is dynamically adjusted.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent screening method for cervical cancer based on multimodal integrated traditional Chinese and Western medicine data. Background Art

[0002] The combination of traditional Chinese and Western medicine has potential advantages in the screening and diagnosis of cervical cancer. The theoretical basis includes the following aspects: TCM theory: Collect and organize TCM theories related to cervical cancer, including etiology, pathogenesis, syndrome differentiation and treatment, etc. The five internal organs (liver, heart, spleen, lung, and kidney) of the human body are interrelated, and disorders can lead to diseases. The treatment of cervical cancer can achieve the purpose of treatment by regulating the functions of related internal organs. According to the theory, syndrome differentiation and treatment are carried out. For example, if cervical cancer patients show liver depression and qi stagnation, liver depression can be relieved; if they show spleen deficiency and dampness, spleen invigoration and dampness removal can be carried out.

[0003] Western medicine theory: Integrate existing Western medicine diagnostic standards and treatment methods, especially the three-step screening method (cervical cytology, HPV testing and colposcopy). Or conduct molecular biological analysis, such as gene mutation and expression. Western medicine reveals the molecular mechanism of cervical cancer by studying changes in gene mutation and gene expression. For example, the E6 and E7 proteins of the HPV virus can cause the inactivation of tumor suppressor genes (such as p53 and Rb) and promote carcinogenesis.

[0004] Theory of integrating Chinese and Western medicine: The theory of integrating Chinese and Western medicine advocates combining the comprehensive differentiation of Chinese medicine with the precise diagnosis of Western medicine to comprehensively and accurately assess the condition. It advocates promoting patient recovery, reducing side effects, and improving the quality of life through holistic conditioning of Chinese medicine while conducting local treatment. A comprehensive artificial intelligence model is developed based on the above theory. Summary of the invention

[0005] The purpose of the present invention is to overcome the problems existing in the prior art and provide an intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data.

[0006] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions: An intelligent screening method for cervical cancer based on multimodal Chinese and Western medicine data. The model includes: Step S1: Data integration, collecting text data from TCM and Western medicine diagnosis and treatment processes respectively, combining TCM and Western medicine data to form a comprehensive patient health text file; Step S1.1: Collect text data from traditional Chinese medicine and Western medicine diagnosis and treatment processes, and divide these text data into two categories: one category includes the patient's medical history and physical examination report data, which are used as input for cervical cancer prediction; the other category includes the doctor's description and instructions, which are used for medical questions and answers; Step S2: Based on the large language model, fine-tune the obtained integrated Chinese and Western medicine data, and use the feature results obtained from the text description to make preliminary screening and diagnosis of cervical cancer; Step S2.1: Text preprocessing: First, the collected text data of integrated Chinese and Western medicine are manually cleaned to find out the missing and incomplete data; then, the word embedding algorithm is used to convert the text into a vector representation as the input of the model; Step S2.2: Use the pre-trained model with set parameters to fine-tune the Chinese and Western medicine text data, and use the features obtained from the contextual semantic description to make preliminary screening and diagnosis of cervical cancer; Step S3: Based on the commonly used medical image segmentation model U-NET++, the cervical cancer image is effectively segmented to facilitate further judgment and guidance of doctors on the cancer situation. During the segmentation task, the graphics are classified and trained so that the model itself can distinguish between normal images and images containing cervical cancer. Step S3.1: For medical image data containing cervical cancer, data cleaning is performed to remove fuzzy or unusable data, and image preprocessing and data enhancement operations are performed; Step S3.2: Perform image segmentation training based on the U-NET++ model to perform image segmentation on cervical cancer medical image data to facilitate doctors' observation and further judgment of the images; Step S3.3: Based on the Resnet-50 model, the segmented cervical cancer data is classified and trained to automatically identify and classify whether the image contains symptoms.

[0007] Step S4: combining the feature data obtained by the context input of the text data with the output of the image data returned by the training as the basis for judging the disease, comprehensively deriving the category of the disease, and finally providing relevant treatment plans; Step S4.1: Combining the result of semantic analysis of the patient's description of the disease with the classification result of cervical cancer output by the image classification model, taking different weights for the two and combining them to judge the patient's symptoms; Step S4.2: Based on the obtained output and combined with the doctor's judgment on the segmented image, the final symptom result is determined, and finally the corresponding treatment plan is output.

[0008] Furthermore, the model also includes: Establish a large model evaluation standard system applicable to the medical field to ensure the reliability and safety of the model.

[0009] Furthermore, the evaluation indicators of the large model evaluation standard system include: Accuracy: including sensitivity, specificity, positive predictive value, and negative predictive value; Security: Ensure the stability of the model in different populations and scenarios; Furthermore, the evaluation method of the large model evaluation standard system includes: Classification and image part of medical image data of cervical cancer in traditional Chinese and Western medicine: split the data set, allocate the training set and validation set in a ratio of 8:2, perform cross-validation, and evaluate the model performance; Practical application stage: Use independent medical data sets for verification to ensure the applicability and robustness of the model in real scenarios; Text data part: PPL perplexity and BLEU bilingual evaluation are used for verification. The lower the PPL perplexity value, the more accurate the model prediction; the higher the BLUE bilingual evaluation value, the better the generated text.

[0010] Comparative analysis: Compare the model with existing screening and diagnostic methods to evaluate its strengths and weaknesses.

[0011] Furthermore, in step S2.1, the cleaning of text data relies on simple text gap detection and manual cleaning, and the word embedding algorithm used to convert text data into vector data is Word2Vec.

[0012] Furthermore, in step S2.2, a pre-trained GPT-3.5 model is used to train the text data.

[0013] Furthermore, in step S3.1, the image preprocessing includes image scaling, image cropping, image flipping and rotation, and data standardization.

[0014] Furthermore, in step S4.2, the function integration is achieved based on PyQt in Python.

[0015] The beneficial effects of the present invention are: The present invention is based on a dialogue system of a large language model and utilizes a GPT model to construct an intelligent consultation system to analyze the symptom descriptions of cervical cancer patients and provide preliminary diagnosis suggestions.

[0016] The present invention is based on deep neural network image classification and segmentation, and realizes automatic classification of cervical cancer images and lesion segmentation by respectively training the U-NET++ image segmentation model and the Resnet-50 image classification model.

[0017] The present invention fuses multimodal data, integrates the patient's medical history, physical examination report, imaging data, etc., builds a comprehensive diagnosis platform, and improves diagnostic accuracy.

[0018] The present invention combines traditional Chinese and Western medicine treatment plans, uses models to generate personalized traditional Chinese and Western medicine treatment plans based on the specific conditions of the patients, and dynamically adjusts the treatment plan. DETAILED DESCRIPTION

[0019] The present invention will be described in detail below with reference to embodiments.

[0020] An intelligent screening method for cervical cancer based on multimodal Chinese and Western medicine data. The model includes: Step S1: Data integration, collecting text data from TCM and Western medicine diagnosis and treatment processes, such as the patient's medical history, physical examination report, instructions given by the physician, etc., combining TCM and Western medicine data to form a comprehensive patient health text file; Step S1.1: Collect text data from traditional Chinese medicine and Western medicine diagnosis and treatment processes, and divide these text data into two categories: one category includes the patient's medical history and physical examination report data, which are used as input for cervical cancer prediction; the other category includes the doctor's description and instructions, which are used for medical questions and answers; Step S2: Based on the large language model, fine-tune the obtained integrated Chinese and Western medicine data, and use the feature results obtained from the text description to make preliminary screening and diagnosis of cervical cancer; Step S2.1: Text preprocessing: First, the collected text data of integrated Chinese and Western medicine are manually cleaned to find out the missing and incomplete data; then, the word embedding algorithm is used to convert the text into a vector representation as the input of the model; Step S2.2: Use the pre-trained GPT-3.5 model with set parameters to fine-tune the Chinese and Western medicine text data, and use the features obtained from the contextual semantic description to make preliminary screening and diagnosis for cervical cancer; Step S3: Based on the commonly used medical image segmentation model U-NET++, the cervical cancer image is effectively segmented to facilitate further judgment and guidance of doctors on the cancer situation. During the segmentation task, the graphics are classified and trained so that the model itself can distinguish between normal images and images containing cervical cancer. Step S3.1: For medical image data containing cervical cancer, data cleaning is performed to remove fuzzy or unusable data, and image preprocessing and data enhancement operations are performed; Step S3.2: Perform image segmentation training based on the U-NET++ model to perform image segmentation on cervical cancer medical image data to facilitate doctors' observation and further judgment of the images; Step S3.3: Based on the Resnet-50 model, the segmented cervical cancer data is classified and trained to automatically identify and classify whether the image contains symptoms.

[0021] Step S4: combining the feature data obtained by the context input of the text data with the output of the image data returned by the training as the basis for judging the disease, comprehensively deriving the category of the disease, and finally providing relevant treatment plans; Step S4.1: Combining the result of semantic analysis of the patient's description of the disease with the classification result of cervical cancer output by the image classification model, taking different weights for the two and combining them to judge the patient's symptoms; Step S4.2: Based on the obtained output and combined with the doctor's judgment on the segmented image, the final symptom result is determined, and finally the corresponding treatment plan is output.

[0022] In the specific implementation of clinical applications: Intelligent consultation system: Dialogue system based on large language model: Using BERT, GPT and other models, we build an intelligent consultation system to analyze the symptom descriptions of cervical cancer patients and provide preliminary diagnosis suggestions; Medical image analysis system: CNN-based image classification and segmentation: Automatic classification and lesion segmentation of cervical cancer images can be achieved by training the CNN model; Multimodal diagnostic platform: Fusion of multimodal data: Integrate the patient's medical history, physical examination report, imaging data, etc. to build a comprehensive diagnostic platform to improve diagnostic accuracy; Personalized treatment recommendation system: Combining Chinese and Western medicine treatment plans: Based on the patient's specific situation, the model is used to generate a personalized Chinese and Western medicine treatment plan and dynamically adjust the treatment plan.

[0023] The model also includes: Establish a large model evaluation standard system applicable to the medical field to ensure the reliability and safety of the model.

[0024] The evaluation indicators of the large model evaluation standard system include: Accuracy: including sensitivity, specificity, positive predictive value, and negative predictive value; Safety: Ensure the stability of the model in different populations and scenarios to avoid misdiagnosis and missed diagnosis; The evaluation method of the large model evaluation standard system includes: Classification and image part of medical image data of cervical cancer in traditional Chinese and Western medicine: split the data set, allocate the training set and validation set in a ratio of 8:2, perform cross-validation, and evaluate the model performance; Practical application stage: Use independent medical data sets for verification to ensure the applicability and robustness of the model in real scenarios; Text data part: PPL (Perplexity) and BLEU (Bilingual Evaluation Understudy) bilingual evaluation are used for verification. The lower the PPL value, the more accurate the model prediction, and the higher the BLEU value, the better the generated text.

[0025] In step S2.1, the cleaning of text data relies on simple text gap detection and manual cleaning, and the word embedding algorithm used to convert text data into vector data is Word2Vec.

[0026] In step S3.1, the image preprocessing mainly includes image scaling, image cropping, image flipping and rotation, data standardization, etc.

[0027] In the step S4.1, the function integration is realized based on PyQt in Python.

[0028] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data, characterized in that: The model includes: Step S1: Data integration, collecting text data from TCM and Western medicine diagnosis and treatment processes respectively, combining TCM and Western medicine data to form a comprehensive patient health text file; Step S1.1: Collect text data from traditional Chinese medicine and Western medicine diagnosis and treatment processes, and divide these text data into two categories: one category includes the patient's medical history and physical examination report data, which are used as input for cervical cancer prediction; the other category includes the doctor's description and instructions, which are used for medical questions and answers; Step S2: Based on the large language model, fine-tune the obtained integrated Chinese and Western medicine data, and use the feature results obtained from the text description to make preliminary screening and diagnosis of cervical cancer; Step S2.1: Text preprocessing: First, the collected text data of integrated Chinese and Western medicine are manually cleaned to find out the missing and incomplete data; then, the word embedding algorithm is used to convert the text into a vector representation as the input of the model; Step S2.2: Use the pre-trained model with set parameters to fine-tune the Chinese and Western medicine text data, and use the features obtained from the contextual semantic description to make preliminary screening and diagnosis of cervical cancer; Step S3: Based on the commonly used medical image segmentation model U-NET++, the cervical cancer image is effectively segmented to facilitate further judgment and guidance of doctors on the cancer situation. During the segmentation task, the graphics are classified and trained so that the model itself can distinguish between normal images and images containing cervical cancer. Step S3.1: For medical image data containing cervical cancer, data cleaning is performed to remove fuzzy or unusable data, and image preprocessing and data enhancement operations are performed; Step S3.2: Perform image segmentation training based on the U-NET++ model to perform image segmentation on cervical cancer medical image data to facilitate doctors' observation and further judgment of the images; Step S3.3: Based on the Resnet-50 model, the segmented cervical cancer data is classified and trained to automatically identify and classify whether the image contains symptoms; Step S4: combining the feature data obtained by the context input of the text data with the output of the image data returned by the training as the basis for judging the disease, comprehensively deriving the category of the disease, and finally providing relevant treatment plans; Step S4.1: Combining the result of semantic analysis of the patient's description of the disease with the classification result of cervical cancer output by the image classification model, taking different weights for the two and combining them to judge the patient's symptoms; Step S4.2: Based on the obtained output and combined with the doctor's judgment on the segmented image, the final symptom result is determined, and finally the corresponding treatment plan is output.

2. The intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data according to claim 1, characterized in that: The model also includes: Establish a large model evaluation standard system applicable to the medical field to ensure the reliability and safety of the model.

3. The intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data according to claim 2, characterized in that: The evaluation indicators of the large model evaluation standard system include: Accuracy: including sensitivity, specificity, positive predictive value, and negative predictive value; Security: Ensure the stability of the model in different populations and scenarios.

4. The intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data according to claim 3 is characterized in that: The evaluation method of the large model evaluation standard system includes: Classification and image part of medical image data of cervical cancer in traditional Chinese and Western medicine: split the data set, allocate the training set and validation set in a ratio of 8:2, perform cross-validation, and evaluate the model performance; Practical application stage: Use independent medical data sets for verification to ensure the applicability and robustness of the model in real scenarios; Text data part: PPL perplexity and BLEU bilingual evaluation are used for verification. The lower the PPL perplexity value, the more accurate the model prediction, and the higher the BLUE bilingual evaluation value, the better the generated text. Comparative analysis: Comparative analysis with existing screening and diagnosis methods is conducted to evaluate the advantages and disadvantages of the model.

5. The intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data according to claim 1, characterized in that: In step S2.1, the cleaning of text data relies on simple text gap detection and manual cleaning, and the word embedding algorithm used to convert text data into vector data is Word2Vec.

6. The intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data according to claim 1, characterized in that: In step S2.2, a pre-trained GPT-3.5 model is used to train the text data.

7. The intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data according to claim 1, characterized in that: In step S3.1, the image preprocessing includes image scaling, image cropping, image flipping and rotation, and data standardization.

8. The intelligent screening method for cervical cancer based on multimodal integrated Chinese and Western medicine data according to claim 1, characterized in that: In step S4.2, the function integration is realized based on PyQt in Python.