Artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method

Abstract

The invention relates to the technical field of patient diagnosis and treatment based on artificial intelligence, in particular to an infertility analysis and diagnosis and treatment auxiliary method based on artificial intelligence, which comprises the following steps: acquiring multi-source data of a patient; semantic analysis is carried out on the multi-source data of the patient, and infertility symptom feature data of the patient is recognized in combination with a machine learning algorithm; constructing a disease risk prediction model based on a deep learning algorithm, and predicting the risk probability of the infertility disease of the patient by taking the infertility symptom characteristic data of the patient as input; and based on the risk probability of the infertility disease prediction of the patient, extracting the potential cause of infertility of the patient through an association rule mining algorithm, and generating a personalized diagnosis and treatment scheme of the patient in combination with a preset medical knowledge base. Through multi-modal time series data analysis, dynamic pathogenesis mining and medical knowledge constraint verification, infertility pathogenesis accurate identification and personalized treatment scheme generation are realized, and diagnosis and treatment accuracy and clinical decision efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of patient diagnosis and treatment based on artificial intelligence. Specifically, it relates to a method for infertility analysis and diagnosis and treatment assistance based on artificial intelligence. Background Art

[0002] Infertility, as a global reproductive health problem, its causes involve multi-dimensional complex mechanisms such as endocrine disorders, reproductive organ abnormalities, genetic factors, and environmental exposures. During the clinical diagnosis and treatment process, it is necessary to integrate multi-modal data such as the patient's hormone level monitoring data over several months, ultrasound image features, medical history records, and lifestyle. It not only has significant time-dynamic characteristics but also has a non-linear coupling relationship between symptom manifestations and causal associations. Traditional medical decision-making has significant limitations in multi-modal data processing. On the one hand, a single prediction model often only focuses on static feature analysis and ignores the temporal relevance of symptom evolution. On the other hand, traditional association rule mining algorithms lack the constraint of medical prior knowledge and are prone to generating a large number of pseudo-associations that do not conform to clinical pathological mechanisms, which not only reduces the credibility of cause identification but may also mislead the design of treatment plans. More critically, in the existing technology chain, the prediction model and the treatment plan generation module are mutually separated, lacking a closed-loop verification mechanism from cause identification to clinical intervention, resulting in limited personalization and executability of the diagnosis and treatment plan. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for infertility analysis and diagnosis and treatment assistance based on artificial intelligence, which realizes accurate identification of infertility causes and generation of personalized treatment plans through multi-modal time-series data analysis, dynamic cause mining, and medical knowledge constraint verification, significantly improving the accuracy of diagnosis and treatment and the efficiency of clinical decision-making.

[0004] The present invention is realized through the following technical solutions: A method for infertility analysis and diagnosis and treatment assistance based on artificial intelligence, the steps of the method include: Obtain multi-source data of the patient; Perform semantic analysis on the multi-source data of the patient, and combine machine learning algorithms to identify the symptom feature data of the patient's infertility; Build a disease risk prediction model based on a deep learning algorithm, using the symptom feature data of the patient's infertility as input to predict the risk probability of the patient's infertility disease; Based on the predicted risk probability of the patient's infertility disease, extract the potential causes of the patient's infertility through an association rule mining algorithm, and generate a personalized diagnosis and treatment plan for the patient in combination with a preset medical knowledge base.

[0005] Optionally, the semantic analysis of the multi-source data of the patient is specifically: Perform natural language preprocessing on the multi-source data of the patient; Build a bidirectional long short-term memory network model, add inter-layer residual connections in the bidirectional long short-term memory network architecture, and integrate the attention mechanism to perform semantic analysis on the preprocessed multi-source patient data to generate a semantic vector for the patient's infertility; Construct a vector comparison space based on a preset medical knowledge base, calculate the similarity between the patient's infertility semantic vector and the medical knowledge base vector through the weighted cosine similarity algorithm, and introduce a semantic association correction factor to optimize the similarity calculation result, generating a set of candidate symptoms for the patient's infertility with confidence scores; Screen the set of candidate symptoms for the patient's infertility through a dynamic threshold filtering module, and output the symptom feature data for the patient's infertility.

[0006] Optionally, the natural language preprocessing of the multi-source patient data is specifically as follows: Standardize the format of the multi-source patient data to generate a text data stream with a unified encoding; Perform word segmentation on the multi-source patient data and synchronously remove invalid stop words; Determine the medical semantic attributes of the multi-source patient data through part-of-speech tagging; Output the multi-source patient data after completing natural language preprocessing.

[0007] Optionally, the construction of the disease risk prediction model based on the deep learning algorithm is specifically as follows: Input the symptom feature data for the patient's infertility; Run the first convolutional neural network model and the second convolutional neural network model in parallel to respectively extract features from the symptom feature data for the patient's infertility; Input the feature extraction result of the first convolutional neural network model into the forward LSTM model to calculate the forward hidden state, and input the feature extraction result of the second convolutional neural network model into the backward LSTM model to calculate the backward hidden state; Jointly splice the forward hidden state and the backward hidden state, and apply the multi-head attention mechanism to obtain the context semantic vector; Aggregate the features of the context semantic vector, and output the predicted risk probability of the patient's infertility disease after activation by the Sigmoid function.

[0008] Optionally, the parallel running of the first convolutional neural network model and the second convolutional neural network model is specifically as follows: Define the first convolutional neural network model, which uses a 3×1 convolutional kernel to extract local features of the symptom feature data for the patient's infertility; Define the second convolutional neural network model, which uses a 7×1 convolutional kernel to extract global features of the symptom feature data for the patient's infertility; Moreover, both the first convolutional neural network model and the second convolutional neural network model are one-dimensional convolutional neural network models.

[0009] Optionally, the step of jointly concatenating the forward hidden state and the backward hidden state and applying a multi-head attention mechanism to obtain a context semantic vector is specifically as follows: Jointly concatenate the forward hidden state and the backward hidden state to form a joint hidden state matrix containing bidirectional temporal information; Split the joint hidden state matrix into multiple attention heads for parallel computing: Each attention head independently calculates the correlation weights of the query vector, the key vector, and the value vector; Calculate the similarity matrix between the query vector and the key vector, and successively perform scaling processing and softmax normalization, and perform weighted aggregation on the value vector to generate a context semantic vector.

[0010] Optionally, the step of aggregating the features of the context semantic vector is specifically as follows: Perform mean pooling on the context semantic vector in the time dimension to extract global semantic features; Perform max pooling on the local features and global features of the patient's infertility symptom feature data in the time dimension respectively; Concatenate the global semantic features, the local features and global features of the patient's infertility symptom feature data to form a comprehensive feature vector, and output the predicted risk probability of the patient's infertility disease after activation by the Sigmoid function.

[0011] Optionally, the step of extracting potential causes of the patient's infertility through an association rule mining algorithm is specifically as follows: Extract the patient's multi-source data through an association rule mining algorithm to obtain a set of candidate symptom-cause pairs for the patient's infertility; Verify the time matching of the set of candidate symptom-cause pairs for the patient's infertility according to the time window of the disease incubation period in the preset medical knowledge base, and output the potential causes of the patient's infertility, including the cause name and the temporal confidence level.

[0012] Optionally, the step of generating a personalized diagnosis and treatment plan for the patient by combining the preset medical knowledge base is specifically as follows: Input the potential causes of the patient's infertility, the predicted risk probability of the patient's infertility disease, and the benchmark incidence data of each cause in the preset medical knowledge base; Calculate a comprehensive score for each cause in the potential causes of the patient's infertility. The scoring basis includes: the predicted risk probability of the patient's infertility disease, the temporal confidence level of the symptom and the cause, and the benchmark incidence of the selected cause; Generate a priority list of the causes of the patient's infertility and sort them in descending order according to the scores; Execute a preset hierarchical decision according to the priority list of the causes of the patient's infertility, and output a personalized diagnosis and treatment plan for the patient.

[0013] Optionally, the preset hierarchical decision includes: First-priority decision: Define that this decision is executed when the score is greater than or equal to the first threshold, and match the emergency diagnosis and treatment plan in the preset medical knowledge base; Second-priority decision: Define that this decision is executed when the score is greater than or equal to the second threshold and lower than the first threshold, and select a treatment plan that balances the efficacy and cost factors to represent the personalized diagnosis and treatment plan for the patient; Third-priority decision: Define that this decision is executed when the score is lower than the second threshold, mark the cause of the patient's infertility to be confirmed, generate a list of manually reviewed causes, and trim the list of manually reviewed causes in combination with the medical record data in the patient's multi-source data, and output the final diagnosis and treatment plan for the patient.

[0014] The technical solution of the present invention has at least the following advantages and beneficial effects: The present invention constructs a complete technical closed-loop from risk prediction to cause mining to treatment decision-making by deeply integrating time-series feature extraction and association analysis under medical constraints. On the one hand, by using bidirectional LSTM and multi-head attention mechanisms, it accurately captures the local fluctuations and global trends of symptom evolution, significantly improving the time-series sensitivity of risk prediction; on the other hand, it innovatively combines the dynamic support degree adjustment mechanism with the verification of the medical knowledge base to ensure that the cause mining not only meets statistical significance but also satisfies clinical pathological logic. Through the multi-level feature pooling strategy, it effectively integrates the semantic features and key event response values across time steps, enabling the model to have both global trend perception and local anomaly detection capabilities. In addition, the treatment plan generation module not only strictly verifies the proportion of indication and contraindication indicators but also enforces the physiological cycle time window constraint to ensure the safety and operability of clinical intervention. The present invention forms a self-optimizing intelligent diagnosis and treatment ecosystem by automatically triggering multi-disciplinary consultations and iterative updates of the knowledge base, and finally outputs a personalized plan document to achieve seamless docking between precision medicine and clinical practice. Brief Description of the Drawings

[0015] Figure 1 It is a schematic flow chart of an infertility analysis and diagnosis and treatment assistance method based on artificial intelligence provided by the present invention; Figure 2 It is a schematic logical diagram of semantic analysis provided by the present invention; Figure 3 It is a schematic logical diagram of a disease risk prediction model provided by the present invention. Detailed Embodiments

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention that are usually described and shown in the accompanying drawings here can be arranged and designed in various different configurations.

[0017] As Figure 1 shown, the present invention provides one of the embodiments: an artificial intelligence-based infertility analysis and diagnosis and treatment assistance method, and the steps of the method include: Obtain multi-source data of the patient; Perform semantic analysis on the multi-source data of the patient, and combine machine learning algorithms to identify the symptom characteristic data of the patient's infertility; Build a disease risk prediction model based on a deep learning algorithm, use the symptom characteristic data of the patient's infertility as input, and predict the risk probability of the patient's infertility disease; Based on the predicted risk probability of the patient's infertility disease, extract the potential causes of the patient's infertility through an association rule mining algorithm, and generate a personalized diagnosis and treatment plan for the patient in combination with a preset medical knowledge base.

[0018] In this embodiment, this embodiment mainly includes Step 1: Collect multi-source data of the patient, including symptom descriptions, age, gender, medical history, examination results, etc.; Step 2: Use natural language processing technology to perform semantic analysis on the patient's symptom descriptions, and combine machine learning algorithms to identify the key symptoms related to infertility; Step 3: Build a disease risk prediction model based on a deep learning algorithm, use multi-dimensional data such as the patient's age, gender, medical history, and examination results as input, and train the model to predict the risk probability of the patient suffering from infertility diseases such as polycystic ovary syndrome, varicocele, and endometriosis; Step 4: Use the association rule mining algorithm to analyze the association relationship between the patient's multi-source data, and find out the potential causes of infertility; Step 5: According to the analysis results of Step 2, Step 3, and Step 4, combine the clinical practice guidelines and the expert experience knowledge base to generate personalized diagnosis and treatment suggestions for doctors, including diagnosis suggestions, treatment suggestions, and follow-up suggestions.

[0019] As Figure 2 shown, in a further implementation of the above embodiment, the semantic analysis of the multi-source data of the patient is specifically: Perform natural language preprocessing on the multi-source data of the patient; Build a bidirectional long short-term memory network model, add an inter-layer residual connection in the bidirectional long short-term memory network architecture, and integrate an attention mechanism to perform semantic analysis on the preprocessed multi-source data of the patient to generate a semantic vector of the patient's infertility; Construct a vector comparison space based on a preset medical knowledge base, calculate the similarity between the semantic vector of the patient's infertility and the vector of the medical knowledge base through the weighted cosine similarity algorithm, and introduce a semantic association correction factor to optimize the similarity calculation result, generating a set of candidate symptoms of the patient's infertility with confidence scores; Filter the set of candidate symptoms of the patient's infertility through a dynamic threshold filtering module, and output the symptom feature data of the patient's infertility.

[0020] In implementation, the algorithm description of the optimized cosine similarity algorithm in this embodiment: Weight adjustment: The original cosine similarity algorithm treats each dimension of the vector equally, but in the analysis of infertility symptoms, the importance of different symptom dimensions varies. For example, the importance of symptoms related to "abnormal ovulation" in infertility diagnosis may be higher than that of "occasional abdominal pain". Through expert experience and statistical analysis of a large number of case data, weights are assigned to different dimensions of the semantic vector of symptom descriptions and the semantic vector of infertility symptoms in the medical knowledge base. Let the semantic vector of symptom descriptions be , where is the semantic vector of symptom descriptions, with a dimension of n, , are respectively the nth and ith elements of the vector, and the semantic vector of symptoms in the medical knowledge base is , where is the semantic vector of symptoms in the medical knowledge base, with a dimension of n, , are respectively the nth and ith elements of the vector, and the weight vector is , where , are respectively the nth and ith elements of the weight vector, and the optimized dot product calculation becomes , where k is K>1 (indicating a strong semantic association,[[]] k=1 indicating a general association,[[]] 0<k<1 indicating a weak semantic association), and the optimized cosine similarity formula is , where It is the optimized weighted cosine similarity, which makes the similarity calculation more in line with the medical semantic logic.

[0021] In this embodiment, the improved bidirectional long short-term memory network (Bi-LSTM) structure is applied: in the Bi-LSTM model based on the attention mechanism, an inter-layer residual connection is added. A direct connection is established between different hidden layers, allowing information to be transmitted more smoothly, alleviating the problem of gradient disappearance, and enabling the model to better learn the complex dependencies in the long sequence of symptom descriptions. For example, in a multi-layer Bi-LSTM, the output of the layer is not only passed to the layer for normal processing, but also directly added to the input of the layer, that is, , where represents the index of the current layer, is the hidden layer output of the layer, h is the hidden layer output, W is the weight matrix, is the weight matrix of the layer, b is the bias, is the bias term of the layer, is the activation function, which enhances the model's ability to learn symptom semantics. Using a language model pre-trained on a large-scale medical text dataset, the medical semantic features learned by the pre-trained model are transferred to the current infertility symptom analysis model. Freeze the parameters of some layers of the pre-trained model and fine-tune the model on the infertility symptom data. In this way, the model can utilize the general medical semantic knowledge captured by the pre-trained model to learn the relevant feature representations of infertility symptoms faster and more accurately, improving the key symptom recognition effect, especially with obvious advantages when the data volume is limited.

[0022] Furthermore, the natural language preprocessing of the patient's multi-source data is specifically as follows: Standardize the format of the patient's multi-source data to generate a text data stream with a unified encoding; Segment the words of the patient's multi-source data and synchronously remove invalid stop words; Determine the medical semantic attributes of the patient's multi-source data through part-of-speech tagging; Output the patient's multi-source data after completing natural language preprocessing.

[0023] In specific implementation, this embodiment adopts a word segmentation algorithm that combines rules and statistical learning to segment the text of the patient's symptom description into individual words. For example, "menstrual disorder accompanied by lower abdominal distension" is segmented into "menstrual disorder", "and", "accompanied by", and "lower abdominal distension". Then, stop words such as "de" and "le" that have no practical meaning are removed to reduce data redundancy. Next, part-of-speech tagging is performed to determine the part of speech of each word, such as nouns, verbs, adjectives, etc., in order to more accurately understand the text meaning and prepare for subsequent analysis.

[0024] As Figure 3 shown, in the specific application of this embodiment, the disease risk prediction model constructed based on the deep learning algorithm is specifically as follows: Input the infertility symptom characteristic data of the patient.

[0025] Run the first convolutional neural network model and the second convolutional neural network model in parallel, and respectively perform feature extraction on the infertility symptom characteristic data of the patient.

[0026] The parallel running of the first convolutional neural network model and the second convolutional neural network model is specifically as follows: Define the first convolutional neural network model, which uses a 3×1 convolutional kernel to extract the local features of the infertility symptom characteristic data of the patient; Define the second convolutional neural network model, which uses a 7×1 convolutional kernel to extract the global features of the infertility symptom characteristic data of the patient; And both the first convolutional neural network model and the second convolutional neural network model are one-dimensional convolutional neural network models.

[0027] Input the feature extraction result of the first convolutional neural network model into the forward LSTM model to calculate the forward hidden state, and input the feature extraction result of the second convolutional neural network model into the reverse LSTM model to calculate the reverse hidden state.

[0028] Jointly splice the forward hidden state and the reverse hidden state, and apply the multi-head attention mechanism to obtain the context semantic vector.

[0029] The joint splicing of the forward hidden state and the reverse hidden state, and the application of the multi-head attention mechanism to obtain the context semantic vector is specifically as follows: Jointly splice the forward hidden state and the reverse hidden state to form a joint hidden state matrix containing bidirectional time series information; Split the joint hidden state matrix into multiple attention heads for parallel calculation: Each attention head independently calculates the correlation weights of the query vector, key vector, and value vector; Calculate the similarity matrix between the query vector and the key vector, and successively perform scaling processing and softmax normalization, and perform weighted aggregation on the value vector to generate a context semantic vector.

[0030] Perform feature aggregation on the context semantic vector, and output the predicted risk probability of the patient's infertility disease after activation by the Sigmoid function.

[0031] The feature aggregation of the context semantic vector is specifically as follows: Perform mean pooling in the time dimension on the context semantic vector to extract global semantic features; Perform max pooling in the time dimension on the local features and global features of the patient's infertility symptom feature data respectively; Concatenate the global semantic features, the local features and global features of the patient's infertility symptom feature data to form a comprehensive feature vector, and output the predicted risk probability of the patient's infertility disease after activation by the Sigmoid function.

[0032] During implementation, obtain the infertility symptom characteristic data of the patient as input, including hormone level sequences, ultrasound imaging indicators, and other clinical manifestation information. To fully capture the local patterns and global trends of the data, a first convolutional neural network model and a second convolutional neural network model are respectively defined to perform feature extraction on the patient data in parallel. The first convolutional neural network model uses a 3×1 convolutional kernel to focus on the local associations between short-term or adjacent time steps; the second convolutional neural network model uses a 7×1 convolutional kernel, mainly for detecting global changes within a larger time range. Both belong to one-dimensional convolutional neural network models, aiming to perform efficient and targeted convolutional scans on the time dimension of the infertility symptom characteristic data. After completing the parallel convolutional extraction, input the feature map obtained from the first convolutional neural network model into the forward LSTM model to obtain the forward hidden state; input the feature map obtained from the second convolutional neural network model into the backward LSTM model to obtain the backward hidden state. By jointly splicing the forward and backward hidden states, a joint hidden state matrix containing bidirectional temporal information is formed. In this matrix, each time step aggregates the feature representations from front to back and from back to front, and better retains the complete context of the patient's symptoms evolving over time. Subsequently, in order to further highlight the importance of different time steps or feature dimensions, a multi-head attention mechanism is applied to the joint hidden state matrix. Specifically, the joint hidden state matrix is split into multiple attention heads for parallel computing. Each head generates a query vector, a key vector, and a value vector respectively, and calculates the similarity score. After scaling and softmax normalization, the value vectors are weighted and aggregated, and finally the context semantic vector is output. This process can adaptively emphasize or weaken the contribution of different temporal positions to the model judgment, making the prediction more targeted and interpretable. After obtaining the context semantic vector, deep information fusion is performed based on the feature aggregation strategy. Its core steps include: performing mean pooling on the context semantic vector in the time dimension to extract the overall semantic information across time steps; performing max pooling on the local features and global features output from the first convolutional neural network and the second convolutional neural network respectively to retain their most significant eigenvalue; splicing the global semantic vector, local features, and global features again to form a unified comprehensive feature vector. Finally, activate the comprehensive feature vector through the Sigmoid function to output the predicted risk probability of the patient's infertility disease. This probability value can be further applied to clinical assistant diagnosis, etiology mining, and treatment plan recommendation, providing a more efficient and accurate risk assessment tool for the field of reproductive medicine.

[0033] In a further implementation of this embodiment, the potential causes of the patient's infertility are extracted by the association rule mining algorithm, specifically: Extract the multi-source data of the patient through the association rule mining algorithm to obtain a set of candidate symptom-cause pairs for the patient's infertility; Verify the time match of the set of candidate symptoms-causes for the patient's infertility according to the time window of the incubation period of the cause in the preset medical knowledge base, and output the potential causes of the patient's infertility, including the cause name and the chronological confidence level.

[0034] The generation of the personalized diagnosis and treatment plan for the patient by combining the preset medical knowledge base is specifically as follows: Input the potential causes of the patient's infertility, the predicted risk probability of the patient's infertility disease, and the benchmark incidence data of each cause in the preset medical knowledge base; Calculate the comprehensive score for each cause in the potential causes of the patient's infertility. The scoring basis includes: the predicted risk probability of the patient's infertility disease, the chronological confidence level of the symptom and the cause, and the benchmark incidence of the selected cause; Generate a priority list of the causes of the patient's infertility and sort it in descending order according to the score; Execute the preset hierarchical decision according to the priority list of the causes of the patient's infertility, and output the personalized diagnosis and treatment plan for the patient.

[0035] The preset hierarchical decision includes: The first priority decision: Define that this decision is executed when the score is greater than or equal to the first threshold, and match the emergency diagnosis and treatment plan in the preset medical knowledge base; The second priority decision: Define that this decision is executed when the score is greater than or equal to the second threshold and lower than the first threshold, and select a treatment plan that balances the efficacy and cost factors, which is characterized as the personalized diagnosis and treatment plan for the patient; The third priority decision: Define that this decision is executed when the score is lower than the second threshold, mark the potential causes of the patient's infertility to be confirmed, generate a list of causes for manual review, and trim the list of causes for manual review in combination with the medical record data in the patient's multi-source data, and output the final diagnosis and treatment plan for the patient.

[0036] Specifically, in this embodiment, the association rule mining algorithm is first used to extract the multi-source data of patients to obtain a set of candidate symptom-cause pairs. The multi-source data includes multi-dimensional information such as hormone level records, ultrasound detection indicators, medical history documents, and living habits. Through a rule mining model trained and verified under a large-scale clinical sample, a preliminary association identification is performed on the infertility candidate symptoms and the possible corresponding causes to form symptom-cause pairs with basic statistical significance. According to the time window of the cause incubation period defined in the preset medical knowledge base, a time matching verification is performed on the above candidate symptom-cause pair set to ensure that only the causes that can produce corresponding clinical manifestations within the incubation period are retained. The potential causes of the patients' infertility output by this step will be recorded according to the cause name and chronological confidence for use by the subsequent diagnosis and treatment plan generation module. This embodiment takes the potential causes of infertility, the predicted risk probability of the patients' infertility disease, and the benchmark incidence data of each cause in the preset medical knowledge base as inputs, and performs a comprehensive score on each cause. The main scoring basis includes: the predicted risk probability of the patients' infertility disease, the chronological confidence of the symptoms and causes, and the benchmark incidence of this cause in the overall population. After scoring, all causes are sorted from high to low according to the scores to form a priority list of infertility causes. According to this priority list, a hierarchical decision-making process is executed to output the personalized diagnosis and treatment plan for the patient. The hierarchical decision-making process is specifically divided into three priorities: the first priority decision corresponds to the cause with a score greater than or equal to the first threshold, and directly matches the diagnosis and treatment plan marked as emergency treatment or critical intervention in the preset medical knowledge base; the second priority decision corresponds to the cause with a score greater than or equal to the second threshold but lower than the first threshold, and a modified treatment plan that takes into account both efficacy and cost is selected to reasonably allocate medical resources while meeting the medical needs of the patients; the third priority decision corresponds to the cause with a score lower than the second threshold, which is temporarily marked as to be confirmed, a list of manually reviewed causes is generated, and the medical record documents and examination results in the multi-source data of the patient are retrieved for in-depth comparison. If it is found during the expert consultation or further examination and verification process that there is indeed clinical evidence for this cause, it will be included in the final diagnosis and treatment plan for the patient; if the verification is invalid or there is not enough evidence to support it, it will be excluded to ensure the safety and accuracy of the diagnosis and treatment plan. It can be understood that the treatment plans are all existing treatment plans in the medical knowledge base. Finally, through the complete processes of the above-mentioned association rule mining, incubation period verification, comprehensive scoring, and hierarchical decision-making, a personalized diagnosis and treatment plan including diagnostic basis, recommended treatment plan, and auxiliary monitoring indicators is output. This plan can be docked with the patient's medical record system or electronic health record (EHR) to realize the closed-loop management of clinical diagnosis and treatment, which not only improves the accuracy of identifying and intervening in the causes of infertility, but also consolidates the standardized utilization and continuous update of medical data, providing more scientific and efficient decision-making support for clinicians.

[0037] The above are only the preferences of the present invention and are not used to limit the present invention. For those skilled in the art, various modifications and changes can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method, characterized in that: The steps of the method include: Obtain multi-source data on patients; Perform semantic analysis on multi-source patient data and use machine learning algorithms to identify patient infertility symptom feature data; A disease risk prediction model is built based on a deep learning algorithm, using the patient's infertility symptom characteristic data as input to predict the patient's risk probability of infertility disease; Based on the predicted risk probability of infertility diseases in patients, the potential causes of infertility in patients are extracted through association rule mining algorithms, and personalized diagnosis and treatment plans for patients are generated in combination with the preset medical knowledge base.

2. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 1, characterized in that: The semantic analysis of the multi-source data of the patient is specifically as follows: Perform natural language preprocessing on multi-source patient data; Construct a bidirectional long short-term memory network model, add inter-layer residual connections to the bidirectional long short-term memory network architecture, and integrate the attention mechanism to perform semantic analysis on the pre-processed multi-source patient data to generate the patient infertility semantic vector; A vector comparison space is constructed based on a preset medical knowledge base. The similarity between the patient's infertility semantic vector and the medical knowledge base vector is calculated using a weighted cosine similarity algorithm. A semantic association correction factor is introduced to optimize the similarity calculation result, generating a set of candidate patient infertility symptoms with confidence scores. The candidate set of infertility symptoms of patients is screened through the dynamic threshold filtering module, and the characteristic data of the infertility symptoms of patients are output.

3. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 2, characterized in that: The natural language preprocessing of the multi-source patient data is specifically as follows: Standardize the format of multi-source patient data and generate uniformly coded text data streams; Perform word segmentation on multi-source patient data and remove invalid stop words simultaneously; Determine the medical semantic attributes of multi-source patient data through part-of-speech tagging; Output multi-source patient data that has completed natural language preprocessing.

4. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 2, characterized in that: The disease risk prediction model is constructed based on the deep learning algorithm, which is specifically: Enter the patient's infertility symptom characteristics data; The first convolutional neural network model and the second convolutional neural network model are run in parallel to extract features from the patient's infertility symptom feature data; The feature extraction results of the first convolutional neural network model are input into the forward LSTM model to calculate the forward hidden state, and the feature extraction results of the second convolutional neural network model are input into the reverse LSTM model to calculate the reverse hidden state; The forward hidden state and the reverse hidden state are concatenated together, and a multi-head attention mechanism is applied to obtain the contextual semantic vector; The context semantic vector is feature aggregated and activated by the Sigmoid function to output the predicted risk probability of the patient's infertility disease.

5. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 4, characterized in that: The first convolutional neural network model and the second convolutional neural network model are run in parallel, which is specifically: Define the first convolutional neural network model, which uses a 3×1 convolution kernel to extract local features of the patient's infertility symptom feature data; Define the second convolutional neural network model, which uses a 7×1 convolution kernel to extract the global features of the patient's infertility symptom feature data; Moreover, the first convolutional neural network model and the second convolutional neural network model are both one-dimensional convolutional neural network models.

6. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 5, characterized in that: The forward hidden state and the reverse hidden state are jointly concatenated, and a multi-head attention mechanism is applied to obtain a context semantic vector, which is specifically: Jointly concatenate the forward hidden state and the reverse hidden state to form a joint hidden state matrix containing bidirectional temporal information; Split the joint hidden state matrix into multiple attention heads that are computed in parallel: Each attention head independently calculates the association weights of the query vector, key vector, and value vector; The similarity matrix between the query vector and the key vector is calculated, and then scaled and softmax normalized in turn, and the value vector is weightedly aggregated to generate a contextual semantic vector.

7. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 6, characterized in that: The feature aggregation of the context semantic vector is specifically as follows: Perform time dimension mean pooling on the context semantic vector to extract global semantic features; Perform maximum pooling in the time dimension on the local features and global features of the patient's infertility symptom feature data; The global semantic features, local features of the patient's infertility symptom feature data, and the global features are spliced ​​to form a comprehensive feature vector, which is then activated by the Sigmoid function to output the patient's infertility disease predicted risk probability.

8. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 7, characterized in that: The potential causes of infertility of patients are extracted by association rule mining algorithm, which are specifically: By extracting multi-source patient data through association rule mining algorithms, we can obtain a set of candidate symptom-cause pairs of patients’ infertility. According to the time window of the incubation period of the cause in the preset medical knowledge base, the time matching of the candidate symptom-cause pair of the patient's infertility is verified, and the potential cause of the patient's infertility is output, which includes the cause name and the time sequence confidence.

9. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 8, characterized in that: The generation of a personalized diagnosis and treatment plan for a patient in combination with a preset medical knowledge base is specifically as follows: Input the potential causes of the patient's infertility, the patient's predicted risk probability of infertility disease, and the baseline incidence data of each cause in the preset medical knowledge base; A comprehensive score is calculated for each potential cause of infertility in the patient, based on the following: the patient's predicted risk probability of infertility disease, the temporal confidence between symptoms and causes, and the baseline incidence of the selected causes; Generate a priority list of the patient's infertility etiologies and sort them from highest to lowest score; Execute preset hierarchical decisions based on the priority list of the patient's infertility causes and output personalized diagnosis and treatment plans for the patient.

10. The artificial intelligence-based infertility analysis and diagnosis and treatment auxiliary method according to claim 9, characterized in that: The preset grading decision includes: First priority decision: define a decision whose score is greater than or equal to the first threshold, and match the emergency diagnosis and treatment plan in the preset medical knowledge base; Second priority decision: Define the score to be greater than or equal to the second threshold and lower than the first threshold to execute this decision, and select a treatment plan that balances efficacy and cost factors to represent a personalized diagnosis and treatment plan for the patient; The third priority decision: define the score below the second threshold to execute this decision, mark the patient's infertility cause to be confirmed, generate a manually reviewed cause list, modify the manually reviewed cause list based on the patient's medical record data in the patient's multi-source data, and output the final patient diagnosis and treatment plan.

Images