Fallopian tube auxiliary prediction method based on large model

Abstract

The application provides an oviduct auxiliary prediction method based on a large model, and belongs to the technical field of AI assistance, and comprises the following steps: acquiring first data before a patient undergoes oviduct dredging and second data after the patient undergoes oviduct dredging, and taking the data as a group of sample data; based on a multi-modal embedding model, each group of sample data is subjected to data vectorization and stored in a vector database as a basic knowledge base; based on a large model and in combination with the basic knowledge base, preoperative examination results of the current patient before oviduct dredging are analyzed to obtain an auxiliary judgment result; if the current patient undergoes oviduct dredging, then the postoperative examination results of the current patient after oviduct dredging are continuously acquired, and the basic knowledge base is updated in combination with the preoperative examination results. The auxiliary prediction of the current patient is realized, and effective pregnancy guidance is facilitated.

Description

Technical Field

[0001] This invention relates to the field of AI-assisted technology, and in particular to an AI-based method for predicting fallopian tube defects. Background Technology

[0002] Fallopian tube recanalization is an interventional treatment for fallopian tube stenosis or blockage, involving fluid flushing and guidewire recanalization under DSA guidance. It is primarily for patients who intend to conceive but have blocked fallopian tubes.

[0003] Currently, it is impossible to determine in advance whether a patient is suitable for fallopian tube recanalization based on the condition of their fallopian tubes, and it is also impossible to predict the surgical outcome after fallopian tube recanalization.

[0004] After tubal recanalization, it is impossible to determine whether the patient can conceive naturally within 6 months, and it is impossible to estimate the approximate time of natural pregnancy. Therefore, it is impossible to provide the patient with the most accurate pregnancy guidance, thereby reducing the efficiency of natural pregnancy and prolonging the patient's natural pregnancy time.

[0005] Therefore, this invention proposes a fallopian tube-assisted prediction method based on a large model. Summary of the Invention

[0006] This invention provides a large-scale model-based method for fallopian tube assisted prediction. By acquiring preoperative test results from historical patients and combining them with postoperative examinations and pregnancy outcomes, a large-scale model is trained to assist in the prediction of current patients, facilitating effective pregnancy guidance, indirectly improving patient examination efficiency, and reducing waiting time.

[0007] This invention provides a fallopian tube-assisted prediction method based on a large model, comprising:

[0008] Step 1: Obtain the first data before the fallopian tube recanalization surgery and the second data after the fallopian tube recanalization surgery for historical patients, and use them as a set of sample data;

[0009] Step 2: Based on the multimodal embedding model, vectorize each set of sample data and store it in a vector database as a basic knowledge base;

[0010] Step 3: Based on the large model and combined with the basic knowledge base, analyze the preoperative examination results of the current patient before the fallopian tube recanalization procedure to obtain auxiliary judgment results;

[0011] Step 4: If the current patient undergoes fallopian tube recanalization, then continue to obtain the postoperative examination results of the current patient after the fallopian tube recanalization, and update the basic knowledge base in combination with the preoperative examination results.

[0012] Preferably, the first data includes: preoperative hysterosalpingography, preoperative vaginal secretions, and preoperative gynecological ultrasound.

[0013] The second set of data includes: postoperative hysterosalpingography, postoperative vaginal secretions, postoperative gynecological ultrasound, and whether a natural pregnancy was achieved within the set cycle after the surgery.

[0014] Preferably, each set of sample data is vectorized based on a multimodal embedding model, including:

[0015] Feature extraction models are selected from the modality-model lookup table based on the data modality, and features are extracted from the preoperative and postoperative information based on the same data modality in the same sample data to obtain the corresponding first feature and second feature;

[0016] The feature differences are obtained by comparing and analyzing all the first features before the operation and all the second features after the operation in the same sample data. The preoperative feature vector, postoperative feature vector and the assignment result of whether the pregnancy occurred in the natural cycle after the operation are combined with the corresponding sample data and input into the multimodal embedding model to construct the vectorized representation of the corresponding sample data.

[0017] Preferably, the preoperative examination results of the current patient prior to the fallopian tube recanalization procedure are analyzed, including:

[0018] The basic knowledge base is trained and validated based on the large model to obtain the current model. The large model takes the vector result corresponding to the first feature as input and the vector result corresponding to the second feature, feature difference and assignment result as output.

[0019] The preoperative examination results are vectorized based on a multimodal embedding model and input into the current model to obtain analysis results, which include: predicted second features, predicted feature differences, and predicted assignment results.

[0020] Preferably, before obtaining the auxiliary judgment result, the following steps are also included:

[0021] Using any sample data as a benchmark, establish association pairs between the first preoperative feature and the first postoperative feature in each data modality of the corresponding sample data and the first and second comparison features in the same data modality of each remaining sample. The association pairs include the first association coefficient between the corresponding sample data before the fallopian tube recanalization procedure and the corresponding remaining sample based on each data modality, and the second association coefficient between the corresponding sample data after the fallopian tube recanalization procedure and the corresponding remaining sample based on each data modality.

[0022] Based on each association, construct a judgment value to determine whether the corresponding sample data is within the set period after surgery;

[0023]

[0024] Wherein, R1 represents the corresponding sample data showing natural pregnancy within a set period after surgery; This indicates that the corresponding sample data did not result in natural pregnancy within the set cycle after the procedure; g1 i1 g2 represents the first correlation coefficient in the i1th data mode of the corresponding correlation pair; i1 This represents the second correlation coefficient under the i1th data modality in the corresponding correlation pair; N1 represents the total number of data modalities, and its value is 3; This indicates the number of coefficients in the corresponding association pair where the first association coefficient is less than 0.7; This indicates the number of coefficients in the corresponding association pair where the second association coefficient is <0.7; max indicates the maximum value sign; min indicates the maximum value sign; Pj indicates the judgment value for whether the corresponding sample data is within the set period after surgery based on the j-th association pair.

[0025] Based on all the judgment values, determine the sample reference recommendation coefficient Yr for the corresponding sample data;

[0026]

[0027] Where sum(Pj≥0) represents the sum of judgment values ​​that satisfy Pj≥0 among all Pj related to the corresponding sample data; Dn represents the number of judgment values ​​that satisfy Pj≥0 among all Pj related to the corresponding sample data; Ln represents the number of judgment values ​​that satisfy Pj<0 among all Pj related to the corresponding sample data; and sum(Pj<0) represents the sum of judgment values ​​that satisfy Pj<0 among all Pj related to the corresponding sample data.

[0028] The sample reference recommendation coefficient Yr is stored in correspondence with the corresponding sample data to construct a repository.

[0029] Preferably, the auxiliary judgment result is obtained, including:

[0030] Based on the patient's requests, establish a probabilistic relationship with second data from historical patients;

[0031] The analysis results are supplemented by the aforementioned probability relationships and sample reference recommendation coefficients to obtain auxiliary judgment results.

[0032] Preferably, establishing a probabilistic relationship between the patient's requests and historical patient data includes:

[0033] The patient's requests are analyzed to determine the current patient's desire to become pregnant in each sub-cycle within a specified period, and a desire sequence is constructed, wherein: 1 represents a desire, 0 represents no desire, and the number of sub-cycles is 6;

[0034] Based on the predicted second feature and the predicted feature difference of the current patient, a similarity analysis is performed with the second feature and feature difference of the second data of the historical patients to obtain the first similarity coefficient and the second similarity coefficient.

[0035] Historical requests from patients whose first similarity coefficient is greater than a threshold and whose second similarity coefficient is greater than a threshold are selected, and the historical requests are parsed to obtain historical sequences.

[0036] When the assignment result of the historical patient corresponding to the historical request is 0, the historical sequence is retained;

[0037] When the assignment result of the historical patient corresponding to the historical request is 1, an implementation label is set for the corresponding sequence in the corresponding historical sequence according to the pregnancy time, and retained;

[0038] Based on all the retained historical sequences, determine the predicted realization probability of each sequence in the desired sequence as a probability relationship;

[0039]

[0040] Where Yu represents the predicted realization probability of the u-th sequence; Bsu represents the number of realization tags in all retained historical sequences that are consistent with the sub-period of the u-th sequence; Ls represents the total number of all retained historical sequences; eu = 0 indicates that the u-th sequence represents no intention; eu = 1 indicates that the u-th sequence represents intention; ! represents the factorial symbol.

[0041] Preferably, the analysis results are supplemented according to the aforementioned probability relationship and sample reference recommendation coefficient to obtain auxiliary judgment results, including:

[0042] Obtain the sample reference recommendation coefficients for the sample data corresponding to each historical request, and then select the first sample corresponding to the largest coefficient.

[0043] The predicted probability of each sequence is added to the judgment value of the first sample to obtain the first probability of each sequence;

[0044] If the last sequence with intention 1 is located, then the maximum value of the first probability among all sequences before and after the last sequence that are related to intention 1 is obtained;

[0045] The analysis results are supplemented based on the maximum value and the sub-period in which the last sequence is located, to obtain auxiliary judgment results.

[0046] Compared with the prior art, the beneficial effects of this application are as follows:

[0047] By acquiring preoperative test results from historical patients and combining them with postoperative examinations and pregnancy outcomes, a large-scale model can be trained to assist in predicting the outcomes of current patients, facilitating effective pregnancy guidance, indirectly improving patient examination efficiency, and reducing waiting time.

[0048] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0049] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a flowchart of a fallopian tube-assisted prediction method based on a large model, as described in an embodiment of the present invention. Detailed Implementation

[0052] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0053] This invention provides a fallopian tube-assisted prediction method based on a large model, such as... Figure 1 As shown, it includes:

[0054] Step 1: Obtain the first data before the fallopian tube recanalization surgery and the second data after the fallopian tube recanalization surgery for historical patients, and use them as a set of sample data;

[0055] Step 2: Based on the multimodal embedding model, vectorize each set of sample data and store it in a vector database as a basic knowledge base;

[0056] Step 3: Based on the large model and combined with the basic knowledge base, analyze the preoperative examination results of the current patient before the fallopian tube recanalization procedure to obtain auxiliary judgment results;

[0057] Step 4: If the current patient undergoes fallopian tube recanalization, then continue to obtain the postoperative examination results of the current patient after the fallopian tube recanalization, and update the basic knowledge base in combination with the preoperative examination results.

[0058] Preferably, the first data includes: preoperative hysterosalpingography, preoperative vaginal secretions, and preoperative gynecological ultrasound.

[0059] The second set of data includes: postoperative hysterosalpingography, postoperative vaginal secretions, postoperative gynecological ultrasound, and whether a natural pregnancy was achieved within the set cycle after the surgery.

[0060] In this embodiment, the first and second data include the direct results of relevant imaging, secretions, and ultrasound, as well as the judgment results. The direct results of preoperative hysterosalpingography show: the uterine cavity is regularly visualized, both fallopian tubes are fully visualized, their course is natural, and contrast agent diffuses at the fimbriae. At this point, the test result indicates that both fallopian tubes are patent. The direct results of preoperative vaginal secretions show: cleanliness grade II, negative for trichomonas, negative for yeast, and negative for bacterial vaginosis (BV). At this point, the test result indicates that cleanliness grade II is within the normal range, and the negative results for trichomonas, yeast, and bacterial vaginosis indicate that the vagina is not infected by these pathogens, the vaginal microecological environment is basically normal, suitable for surgery, and unlikely to cause postoperative infection. The direct results of preoperative gynecological ultrasound show: the uterus is of normal size and shape, the myometrium has uniform echogenicity, the endometrium is of moderate thickness, and no obvious abnormal echoes are seen in the uterine cavity. At this point, the test result shows that both ovaries are of normal size, and no obvious abnormal masses are seen.

[0061] The direct results, both before and after the surgery, are related to the patency of the fallopian tubes on both sides. The patency of each fallopian tube can be categorized into three types: patent, blocked, and partially patent.

[0062] In this embodiment, sample data = {first data and second data corresponding to the patient}.

[0063] In this embodiment, the multimodal embedding model is used to vectorize the results under different detection types, so as to obtain the vectorized results and store them in a vector database (a storage space). The multimodal embedding model is a known existing technology.

[0064] In this embodiment, the large model can be a large model that can be privately deployed, such as Llama3, Qwen2, or DeepSeek.

[0065] In this embodiment, based on a large model and combined with a basic knowledge base, the preoperative examination results of the current patient before the fallopian tube recanalization procedure can be directly analyzed to obtain relevant prediction results. These results are considered as analysis results. By supplementing the analysis results with probabilities, auxiliary judgment results are obtained.

[0066] In this embodiment, the cycle is set to 6 months, and each month is considered as a sub-cycle.

[0067] In this embodiment, the purpose of updating the basic knowledge base is to provide more samples to enrich the large model.

[0068] The beneficial effects of the above technical solution are: by acquiring preoperative test results from historical patients and combining them with postoperative examinations and pregnancy outcomes, a large model can be trained to assist in the prediction of current patients, making it easier to provide effective pregnancy guidance, indirectly improving the efficiency of patient examinations and reducing waiting time.

[0069] This invention provides a fallopian tube-assisted prediction method based on a large model, which vectorizes each set of sample data based on a multimodal embedding model, including:

[0070] Feature extraction models are selected from the modality-model lookup table based on the data modality, and features are extracted from the preoperative and postoperative information based on the same data modality in the same sample data to obtain the corresponding first feature and second feature;

[0071] The feature differences are obtained by comparing and analyzing all the first features before the operation and all the second features after the operation in the same sample data. The preoperative feature vector, postoperative feature vector and the assignment result of whether the pregnancy occurred in the natural cycle after the operation are combined with the corresponding sample data and input into the multimodal embedding model to construct the vectorized representation of the corresponding sample data.

[0072] In this embodiment, the data modalities include: contrast imaging modal, secretion modal, and ultrasound modal. Since contrast imaging is presented in video format, secretion is presented in tabular format, and ultrasound is presented in image format, these three modalities constitute multimodality.

[0073] In this embodiment, the modality-model lookup table includes feature extraction models under different data modalities. All of these are pre-trained neural network models and are trained using data information under the corresponding modality and feature results for that information as samples. The training samples are greater than 1,000.

[0074] In this embodiment, each historical patient has first data and second data, and the preoperative and postoperative information is data extracted from the first data and second data under the same modality. For example, the characteristics of hysterosalpingography can be: the first characteristic is: the left fallopian tube is completely blocked and the right fallopian tube is not blocked; the second characteristic is: the left fallopian tube is partially blocked and the right fallopian tube is not blocked.

[0075] At this point, the key difference is between complete blockage of the left fallopian tube and partial blockage of the left fallopian tube.

[0076] In this embodiment, if pregnancy is predicted, the value is 1; if pregnancy is predicted, the value is 0.

[0077] In this embodiment, the multimodal embedding model is obtained by training the neural network model with samples based on the expression of preoperative feature vectors, postoperative feature vectors, and the assignment results of whether pregnancy occurs within the natural cycle under different data modalities. Therefore, the vectorized representation of sample data can be directly realized.

[0078] In this embodiment, the vectorized representation is defined as {the vector result corresponding to the first feature, the vector result corresponding to the second feature and the feature difference, and the vector result corresponding to the assignment result}.

[0079] The beneficial effects of the above technical solution are: by selecting models from the comparison table, it is convenient to extract features of information under different modalities, and based on the multimodal embedding model, it is convenient to realize the vectorized representation of data, providing simplified samples for subsequent large model training and prediction, and improving training efficiency.

[0080] This invention provides a large-model-based method for predicting fallopian tube patency, which analyzes the preoperative examination results of a patient before undergoing fallopian tube recanalization surgery, including:

[0081] The basic knowledge base is trained and validated based on the large model to obtain the current model. The large model takes the vector result corresponding to the first feature as input and the vector result corresponding to the second feature, feature difference and assignment result as output.

[0082] The preoperative examination results are vectorized based on a multimodal embedding model and input into the current model to obtain analysis results, which include: predicted second features, predicted feature differences, and predicted assignment results.

[0083] In this embodiment, the amount of sample data is 1000 cases or more. It should be noted that during the training and validation of the large model, it is either a specialized training based on the existing basic knowledge base, or a fine-tuning process based on the existing large model and the basic knowledge base.

[0084] In this embodiment, the preoperative test results are consistent with the content types contained in the first data of historical patients.

[0085] The beneficial effects of the above technical solution are: by training and refining the large model, it is convenient to analyze the preoperative examination results, obtain the analysis results, and provide a basis for subsequent auxiliary guidance.

[0086] This invention provides a fallopian tube-assisted prediction method based on a large model, which further includes the following steps before obtaining the auxiliary judgment result:

[0087] Using any sample data as a benchmark, establish association pairs between the first preoperative feature and the first postoperative feature in each data modality of the corresponding sample data and the first and second comparison features in the same data modality of each remaining sample. The association pairs include the first association coefficient between the corresponding sample data before the fallopian tube recanalization procedure and the corresponding remaining sample based on each data modality, and the second association coefficient between the corresponding sample data after the fallopian tube recanalization procedure and the corresponding remaining sample based on each data modality.

[0088] Based on each association, construct a judgment value to determine whether the corresponding sample data is within the set period after surgery;

[0089]

[0090] Wherein, R1 represents the corresponding sample data showing natural pregnancy within a set period after surgery; This indicates that the corresponding sample data did not result in natural pregnancy within the set cycle after the procedure; g1 i1 g2 represents the first correlation coefficient in the i1th data mode of the corresponding correlation pair; i1 This represents the second correlation coefficient under the i1th data modality in the corresponding correlation pair; N1 represents the total number of data modalities, and its value is 3; This indicates the number of coefficients in the corresponding association pair where the first association coefficient is less than 0.7; This indicates the number of coefficients in the corresponding association pair where the second association coefficient is <0.7; max indicates the maximum value sign; min indicates the maximum value sign; Pj indicates the judgment value for whether the corresponding sample data is within the set period after surgery based on the j-th association pair.

[0091] Based on all the judgment values, determine the sample reference recommendation coefficient Yr for the corresponding sample data;

[0092]

[0093] Where sum(Pj≥0) represents the sum of judgment values ​​that satisfy Pj≥0 among all Pj related to the corresponding sample data; Dn represents the number of judgment values ​​that satisfy Pj≥0 among all Pj related to the corresponding sample data; Ln represents the number of judgment values ​​that satisfy Pj<0 among all Pj related to the corresponding sample data; and sum(Pj<0) represents the sum of judgment values ​​that satisfy Pj<0 among all Pj related to the corresponding sample data.

[0094] The sample reference recommendation coefficient Yr is stored in correspondence with the corresponding sample data to construct a repository.

[0095] In this embodiment, the first correlation coefficient and the second correlation coefficient are both implemented based on similarity functions, such as sim(preoperative features of the corresponding data sample and the preoperative features of the corresponding remaining sample under data modality A1), and the value range is from 0 to 1.

[0096] In this embodiment, the association pair = {the first association coefficient for each data modality before surgery, and the second association coefficient for each data modality after surgery}.

[0097] In this embodiment, the second feature to be compared and the first feature to be compared are the second feature and the first feature obtained based on the feature extraction model under the corresponding modality, which are postoperative and preoperative features.

[0098] In this embodiment, the repository is used to store the sample reference recommendation coefficients and the sample data corresponding to those coefficients, so that they can be easily retrieved and used later.

[0099] The beneficial effects of the above technical solution are: by suggesting association pairs based on the first association coefficient and the second association coefficient based on any sample data, it is convenient to calculate the judgment value, and further, by comparing the judgment value with the value of 0.7, the reference recommendation coefficient for different samples can be determined, thereby determining the reliability of subsequent supplementary analysis results and indirectly improving the accuracy of auxiliary guidance.

[0100] This invention provides a fallopian tube-assisted prediction method based on a large model, which obtains auxiliary judgment results, including:

[0101] Based on the patient's requests, establish a probabilistic relationship with second data from historical patients;

[0102] The analysis results are supplemented by the aforementioned probability relationships and sample reference recommendation coefficients to obtain auxiliary judgment results.

[0103] In this embodiment, the probabilistic relationship refers to the relationship obtained based on the current patient's needs.

[0104] In this embodiment, "supplement" refers to supplementing the more accurate pregnancy cycle and pregnancy probability.

[0105] The beneficial effects of the above technical solution are: it determines supplementary information for the analysis results based on probabilistic relationships, ensuring the comprehensiveness of the auxiliary judgment results.

[0106] This invention provides a fallopian tube assisted prediction method based on a large model, which establishes a probabilistic relationship between the patient's needs and second-generation data from historical patients, including:

[0107] The patient's requests are analyzed to determine the current patient's desire to become pregnant in each sub-cycle within a specified period, and a desire sequence is constructed, wherein: 1 represents a desire, 0 represents no desire, and the number of sub-cycles is 6;

[0108] Based on the predicted second feature and the predicted feature difference of the current patient, a similarity analysis is performed with the second feature and feature difference of the second data of the historical patients to obtain the first similarity coefficient and the second similarity coefficient.

[0109] Historical requests from patients whose first similarity coefficient is greater than a threshold and whose second similarity coefficient is greater than a threshold are selected, and the historical requests are parsed to obtain historical sequences.

[0110] When the assignment result of the historical patient corresponding to the historical request is 0, the historical sequence is retained;

[0111] When the assignment result of the historical patient corresponding to the historical request is 1, an implementation label is set for the corresponding sequence in the corresponding historical sequence according to the pregnancy time, and retained;

[0112] Based on all the retained historical sequences, determine the predicted realization probability of each sequence in the desired sequence as a probability relationship;

[0113]

[0114] Where Yu represents the predicted realization probability of the u-th sequence; Bsu represents the number of realization tags in all retained historical sequences that are consistent with the sub-period of the u-th sequence; Ls represents the total number of all retained historical sequences; eu = 0 indicates that the u-th sequence represents no intention; eu = 1 indicates that the u-th sequence represents intention; ! represents the factorial symbol.

[0115] In this embodiment, the demand analysis is achieved through an analytical model, which is trained on a neural network model using different patient demands and the patient's pregnancy intention within the same cycle in response to the demand as samples, and the number of training samples is greater than 1000.

[0116] In this embodiment, for example, the intention sequence is: {0 0 1 1 1 1}.

[0117] In this embodiment, the second feature and feature difference of the second data of the historical patients are known and extracted from the historical data. The predicted second feature and the predicted feature difference are obtained based on the prediction of the first feature of the current patient by the large model. The two acquisition methods are different.

[0118] In this embodiment, the similarity analysis is also calculated based on the similarity function, and the first similarity coefficient = sim(predicted second feature, second feature of second data under historical patients);

[0119] Second similarity coefficient = sim(predicted feature difference, feature difference of second data under historical patients).

[0120] It should be noted that the values ​​of the first similarity coefficient and the second similarity coefficient both range from 0 to 1.

[0121] In this embodiment, the coefficient threshold is 0.8.

[0122] In this embodiment, as long as the obtained similarity coefficient is greater than the coefficient threshold, the historical request can be directly locked, and the historical sequence is obtained by analyzing the historical request based on the analytical model.

[0123] In this embodiment, for example, there are months 1, 2, 3, 4, 5, and 6, and the cycle is set to 6 months starting from the moment the surgery ends. Suppose the pregnancy is in the 3rd month, then an implementation label is set for the sequence corresponding to the 3rd month (the 3rd sequence) (mainly to mark which month of pregnancy occurred after the surgery).

[0124] In this embodiment, for example, there are 3 sequences, which are {0 0 1 (implementation label) 1 1 1}, {1 0 1 0 11}, and {1 1 1 (implementation label) 1 1 1}. In this case, the Bsu for the third sequence is 2 and the Ls is 3.

[0125] The beneficial effects of the above technical solution are: by analyzing patient needs to construct a willingness sequence, and by performing similarity analysis between the predicted second feature and feature difference and the second feature and feature difference of historical patients, the probability of each sequence in the willingness sequence can be set by comparing the magnitude of the two coefficients and assigning labels, thus providing a reliable analytical basis for subsequent supplementation of the analysis results.

[0126] This invention provides a fallopian tube-assisted prediction method based on a large model. The analysis results are supplemented according to the aforementioned probability relationship and sample reference recommendation coefficients to obtain auxiliary judgment results, including:

[0127] Obtain the sample reference recommendation coefficients for the sample data corresponding to each historical request, and then select the first sample corresponding to the largest coefficient.

[0128] The predicted probability of each sequence is added to the judgment value of the first sample to obtain the first probability of each sequence;

[0129] If the last sequence with intention 1 is located, then the maximum value of the first probability among all sequences before and after the last sequence that are related to intention 1 is obtained;

[0130] The analysis results are supplemented based on the maximum value and the sub-period in which the last sequence is located, to obtain auxiliary judgment results.

[0131] In this embodiment, historical demands refer to the demands of historical patients whose first similarity coefficient is greater than the coefficient threshold and whose second similarity coefficient is greater than the coefficient threshold.

[0132] In this embodiment, there are 6 sequences in each current patient's wish sequence, and the prediction probability of each sequence plus the judgment value of the first sample equals the first probability.

[0133] In this embodiment, for example, if the last sequence in the located intention sequence {0 0 1 1 1 1} is the 6th month, then the first probability associated with the 6th sequence and the sequences preceding the 6th sequence that are related to the intention of 1 is obtained, and the maximum probability is determined as the maximum value.

[0134] In this embodiment, the supplementary result is: the probability of pregnancy is at its maximum by the 6th month.

[0135] In this embodiment, the auxiliary judgment result is a combination of the analysis result and the supplementary result.

[0136] In this embodiment, if it is predicted that natural pregnancy will not be possible within the set period, or if the probability in the supplementary results is less than 30%, reproductive measures such as in vitro fertilization will be suggested.

[0137] The beneficial effects of the above technical solution are: to obtain the first sample by starting with historical demands, to obtain the first probability by adding the predicted probability and the judgment value, and then to construct the supplementary result by using the maximum value and the sub-cycle of the last sequence with a willingness of 1, so as to obtain the auxiliary judgment result and ensure its accuracy.

[0138] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A fallopian tube-assisted prediction method based on a large model, characterized in that, include: Step 1: Obtain the first data before the fallopian tube recanalization surgery and the second data after the fallopian tube recanalization surgery of historical patients, and use them as a set of sample data; Step 2: Based on the multimodal embedding model, vectorize each set of sample data and store it in a vector database as a basic knowledge base; Step 3: Based on the large model and combined with the basic knowledge base, analyze the preoperative examination results of the current patient before the fallopian tube recanalization procedure to obtain auxiliary judgment results; Step 4: If the current patient undergoes fallopian tube recanalization, then continue to obtain the postoperative examination results of the current patient after the fallopian tube recanalization, and update the basic knowledge base in combination with the preoperative examination results; The first data includes: preoperative hysterosalpingography, preoperative vaginal secretions, and preoperative gynecological ultrasound. The second set of data includes: postoperative hysterosalpingography, postoperative vaginal secretions, postoperative gynecological ultrasound, and whether the patient conceived naturally within the set cycle after the surgery; Among them, each set of sample data is vectorized based on a multimodal embedding model, including: Feature extraction models are selected from the modality-model lookup table based on the data modality, and features are extracted from the preoperative and postoperative information based on the same data modality in the same sample data to obtain the corresponding first feature and second feature; The feature differences are obtained by comparing and analyzing all the first features before the operation and all the second features after the operation in the same sample data. The preoperative feature vector, postoperative feature vector and the assignment result of whether the pregnancy occurred in the natural cycle after the operation are combined with the corresponding sample data and input into the multimodal embedding model to construct the vectorized representation of the corresponding sample data. Before obtaining the auxiliary judgment result, it also includes: Using any sample data as a benchmark, establish association pairs between the first preoperative feature and the first postoperative feature in each data modality of the corresponding sample data and the first and second comparison features in the same data modality of each remaining sample. The association pairs include the first association coefficient between the corresponding sample data before the fallopian tube recanalization procedure and the corresponding remaining sample based on each data modality, and the second association coefficient between the corresponding sample data after the fallopian tube recanalization procedure and the corresponding remaining sample based on each data modality. Based on each association, construct a judgment value to determine whether the corresponding sample data is within the set period after surgery; Wherein, R1 represents the corresponding sample data showing natural pregnancy within a set period after surgery; This indicates that the corresponding sample data did not result in natural pregnancy within the set cycle after the procedure; g1 i1 g2 represents the first correlation coefficient in the i1th data mode of the corresponding correlation pair; i1 This represents the second correlation coefficient under the i1th data modality in the corresponding correlation pair; N1 represents the total number of data modalities, and its value is 3; This indicates the number of coefficients in the corresponding association pair where the first association coefficient is less than 0.7; This indicates the number of coefficients in the corresponding association pair where the second association coefficient is <0.7; max indicates the maximum value sign; min indicates the maximum value sign; Pj indicates the judgment value for whether the corresponding sample data is within the set period after surgery based on the j-th association pair. Based on all the judgment values, determine the sample reference recommendation coefficient Yr for the corresponding sample data; Where sum(Pj≥0) represents the sum of judgment values ​​that satisfy Pj≥0 among all Pj related to the corresponding sample data; Dn represents the number of judgment values ​​that satisfy Pj≥0 among all Pj related to the corresponding sample data; Ln represents the number of judgment values ​​that satisfy Pj<0 among all Pj related to the corresponding sample data; and sum(Pj<0) represents the sum of judgment values ​​that satisfy Pj<0 among all Pj related to the corresponding sample data. The sample reference recommendation coefficient Yr is stored in correspondence with the corresponding sample data to construct a repository.

2. The fallopian tube-assisted prediction method based on a large model according to claim 1, characterized in that, An analysis was conducted on the preoperative examination results of the current patient prior to the fallopian tube recanalization procedure, including: The basic knowledge base is trained and validated based on the large model to obtain the current model. The large model takes the vector result corresponding to the first feature as input and the vector result corresponding to the second feature, feature difference and assignment result as output. The preoperative examination results are vectorized based on a multimodal embedding model and input into the current model to obtain analysis results, which include: predicted second features, predicted feature differences, and predicted assignment results.

3. The fallopian tube-assisted prediction method based on a large model according to claim 1, characterized in that, Obtain auxiliary judgment results, including: Based on the patient's requests, establish a probabilistic relationship with second data from historical patients; The analysis results are supplemented by the aforementioned probability relationships and sample reference recommendation coefficients to obtain auxiliary judgment results.

4. The fallopian tube-assisted prediction method based on a large model according to claim 3, characterized in that, Based on the patient's requests, a probabilistic relationship is established with second data from historical patients, including: The patient's requests are analyzed to determine the current patient's desire to become pregnant in each sub-cycle within a specified period, and a desire sequence is constructed, wherein: 1 represents a desire, 0 represents no desire, and the number of sub-cycles is 6; Based on the predicted second feature and the predicted feature difference of the current patient, a similarity analysis is performed with the second feature and feature difference of the second data of the historical patients to obtain the first similarity coefficient and the second similarity coefficient. Historical requests from patients whose first similarity coefficient is greater than a threshold and whose second similarity coefficient is greater than a threshold are selected, and the historical requests are parsed to obtain historical sequences. When the assignment result of the historical patient corresponding to the historical request is 0, the historical sequence is retained; When the assignment result of the historical patient corresponding to the historical request is 1, an implementation label is set for the corresponding sequence in the corresponding historical sequence according to the pregnancy time, and retained; Based on all the retained historical sequences, determine the predicted realization probability of each sequence in the desired sequence as a probability relationship; Where Yu represents the predicted realization probability of the u-th sequence; Bsu represents the number of realization tags in all retained historical sequences that are consistent with the sub-period of the u-th sequence; Ls represents the total number of all retained historical sequences; eu = 0 indicates that the u-th sequence represents no intention; eu = 1 indicates that the u-th sequence represents intention; ! represents the factorial symbol.

5. The fallopian tube-assisted prediction method based on a large model according to claim 4, characterized in that, The analysis results are supplemented based on the aforementioned probability relationships and sample reference recommendation coefficients to obtain auxiliary judgment results, including: Obtain the sample reference recommendation coefficients for the sample data corresponding to each historical request, and then select the first sample corresponding to the largest coefficient. The predicted probability of each sequence is added to the judgment value of the first sample to obtain the first probability of each sequence; If the last sequence with intention 1 is located, then the maximum value of the first probability among all sequences before and after the last sequence that are related to intention 1 is obtained; The analysis results are supplemented based on the maximum value and the sub-period in which the last sequence is located, to obtain auxiliary judgment results.

Images