System for assessing risk of post-operative recurrence of upper urinary tract stones based on multi-modal data

Abstract

The present application relates to the technical field of urinary calculus risk assessment, and particularly relates to a kind of upper urinary tract calculus postoperative recurrence risk assessment system based on multi-modal data.The present application trains neural network according to the training stone state in patient body and previous examination vector, determines the expected comprehensive load index of verification patient and the expected local load index under each mode using trained neural network, and obtains the prediction contribution weight of each mode;Using prediction contribution weight to adjust the expected local load index of all modes of the patient to be tested during follow-up, obtain the initial recurrence risk degree, and combine the life habits similarity and disease characteristics similarity of the patient to be tested and historical recurrence patients, obtain the calculus recurrence risk degree;Based on calculus recurrence risk degree, the recurrence risk of upper urinary tract calculus of the patient to be tested after operation is evaluated.The present application combines the static basic risk and dynamic disease risk of patient, reduces the influence of multi-modal data loss during follow-up on recurrence risk assessment.

Description

Technical Field

[0001] This invention relates to the field of urinary stone risk assessment technology, specifically to a system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data. Background Technology

[0002] Upper urinary tract stones are a common disease in urology, with a high recurrence rate, causing long-term health problems for patients. The main causes of urinary tract stone formation are usually a supersaturated state of salts in the urine that form stone crystals, insufficient substances in the urine that inhibit crystal formation, and an increased matrix. However, many factors contribute to these conditions, such as dietary habits, metabolic genetics, and occupation. Therefore, it is necessary to integrate multimodal data to assess the risk of urinary stone recurrence to improve the accuracy of the assessment results.

[0003] Currently, assessing the recurrence risk of upper urinary tract stones usually requires multiple medical examinations for diagnosis. However, in actual postoperative follow-up, considering the patient's medical costs and the convenience of follow-up, it is usually impossible to repeat the comprehensive examinations as before the operation. During the follow-up period, the medical examination items for patients are only a few of the preoperative examination items, resulting in a significant reduction in the data modalities that can be used for assessment, which in turn makes the reliability of postoperative risk assessment of upper urinary tract stones poor. Summary of the Invention

[0004] To address the technical problem of poor reliability in recurrence risk assessment due to missing multimodal data during follow-up, the present invention aims to provide a multimodal data-based system for assessing the recurrence risk after upper urinary tract stone surgery. The specific technical solution adopted is as follows: This invention proposes a system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data, the system comprising: The data acquisition module is used to obtain the preliminary examination vectors of each patient with upper urinary tract stones, as well as the follow-up examination vectors of the patients to be tested and each patient with a history of recurrence. Both types of vectors are composed of medical examination data of various modalities. Patients with upper urinary tract stones are divided into training patients and validation patients. The modal contribution analysis module is used to train the neural network based on the patient's stone status and previous examination vectors; and to obtain the predicted contribution weight of each modality based on the expected comprehensive load index of the verification patient and the expected local load index under each modality determined by the trained neural network. The recurrence risk analysis module is used to adjust the expected local burden index of the patient under the follow-up period in all modalities using the predicted contribution weight, and to obtain the initial recurrence risk of the patient under the test; based on the similarity of the patient under the test to the similarity of lifestyle and symptom characteristics with the patient under the test and the initial recurrence risk, the risk of stone recurrence of the patient under the test is obtained. The recurrence risk assessment module is used to assess the risk of recurrence after surgery for upper urinary tract stones in the tested patients based on the stone recurrence risk level.

[0005] Furthermore, the method for obtaining the preliminary inspection vector and the follow-up inspection vector includes: Acquire medical examination data of various modalities in patients with upper urinary tract stones before surgery, and arrange all the medical examination data of the preoperative modalities in a predetermined order to form a preoperative examination vector; Acquire medical examination data of several modalities of the patient to be tested during the follow-up period. The modalities during the follow-up period are a subset of the modalities before surgery. According to the predetermined order, place the medical examination data of all modalities during the follow-up period in the corresponding positions, and set the medical examination data of each remaining modality as the mean of the medical examination data of all training patients in that modality to generate a follow-up examination vector.

[0006] Furthermore, training the neural network includes: Obtain the volume of each stone in the body of the patient with upper urinary tract stones; normalize the sum of the volumes of all stones in the patient with upper urinary tract stones, and use the normalized result as the stone burden index. The preliminary examination vector of the patient with upper urinary tract stones and the stone burden index are used as input data and output data of a training sample, respectively; the training samples of all training patients constitute the training set of the neural network; and the neural network is trained using the training set.

[0007] Furthermore, the method for obtaining the patient's expected comprehensive burden index and the expected local burden index under each modality includes: The patient's preliminary examination vectors are input into the trained neural network, and the neural network outputs the expected comprehensive workload index. Choose any modality and denote it as the example modality. Set the remaining elements in the early examination vector of the verification patient, except for the elements corresponding to the example modality, to the mean of the medical examination data of all training patients in that modality, to obtain the local early vector under the example modality. Input the local early vector into the trained neural network, and the neural network outputs the expected local load index of the verification patient under the example modality.

[0008] Further, obtaining the prediction contribution weight for each modality includes: Patients whose absolute difference between the stone burden index and the expected comprehensive burden index is less than a preset difference threshold are selected as the analysis patients; A risk matrix is ​​constructed from the expected local burden indices of all analyzed patients across all modalities. The number of rows and columns of the risk matrix are the number of analyzed patients and the number of modalities in the preoperative period, respectively. A column matrix is ​​constructed from the expected comprehensive burden indices of all analyzed patients, denoted as the comprehensive result matrix. Set up a contribution weight column matrix, where the number of rows in the column matrix is ​​equal to the number of modalities in the preoperative period; Multiply the risk matrix by the contribution weight column matrix to obtain a new matrix. Calculate the sum of squared differences between the corresponding elements of the new matrix and the comprehensive result matrix, and use this sum as the error function. Use each element in the contribution weight column matrix corresponding to the minimum value of the error function as the prediction contribution weight of the corresponding mode.

[0009] Furthermore, obtaining the initial relapse risk of the patient to be tested includes: Obtain the expected local burden index for each modality of the patient under test during the follow-up period; sum the expected local burden index of the patient under test in all modalities during the follow-up period with the predicted contribution weight, and use the ratio of the sum to the sum of the predicted contribution weights of all modalities during the follow-up period as the initial recurrence risk.

[0010] Furthermore, obtaining the risk of stone recurrence in the patient under test includes: When the number of modalities of the patient during the follow-up period is equal to the number of modalities before the operation, the initial recurrence risk is taken as the stone recurrence risk. When the number of modalities of the patient under test during the follow-up period is less than the number of modalities before surgery, it is determined whether the initial recurrence risk is greater than or equal to the preset recurrence threshold. If so, the initial recurrence risk is used as the stone recurrence risk. If not, a matching patient is selected from the patients with historical recurrence. Based on the similarity of the lifestyle and symptom characteristics between the patient under test and each matching patient, the similarity of the symptoms between the two patients is obtained. The product of the difference between the preset recurrence threshold and the initial recurrence risk of each matching patient of the patient under test and the similarity of the symptoms is averaged to obtain the recurrence risk of the patient under test. The sum of the initial recurrence risk and the disease recurrence risk of the patient to be tested is taken as the recurrence risk of the patient's stones.

[0011] Furthermore, obtaining the similarity of symptom presentation between two corresponding patients includes: Obtain the lifestyle vectors and symptom vectors of the patient to be tested and their matching patients, and denote them as analysis vectors; Calculate the cosine similarity of each analysis vector between the test patient and each of their matched patients, and use the mean of all cosine similarities as the similarity of disease manifestations between the test patient and each of their matched patients.

[0012] Furthermore, the absolute difference between the elements of each dimension in the follow-up examination vector of the matched patient and the patient to be tested is less than the preset element difference threshold of the corresponding dimension, and the initial recurrence risk of the matched patient is less than the preset recurrence threshold.

[0013] Furthermore, the number of patients analyzed is greater than the number of modalities in the preoperative period.

[0014] The present invention has the following beneficial effects: In this embodiment of the invention, a neural network is trained based on the patient's pre-examination vectors and stone status. Validation patient data is used to quantify the expected comprehensive load index and the expected local load index under each modality. The importance of medical examination data for each modality in predicting recurrence risk is analyzed to obtain predictive contribution weights, thus improving the reliability of risk assessment. In cases where modality data is missing during follow-up, the expected local load index is adjusted based on limited follow-up examination data using predictive contribution weights. This approach respects the independent value of each modality while considering its relative importance in specific clinical problems, resulting in a more reliable initial recurrence risk. Simultaneously, symptom descriptions and lifestyle habits can indirectly reflect the patient's stone recurrence status. By combining the similarity in lifestyle habits and symptom characteristics between the tested patient and historical recurrence patients, the initial recurrence risk is optimized, ultimately obtaining an accurate stone recurrence risk. By integrating the patient's static baseline risk and dynamic symptom risk, a more comprehensive and accurate quantitative assessment of individual recurrence probability is achieved. This effectively overcomes the problems of large assessment errors and poor reliability caused by incomplete follow-up data, significantly improving the reliability and robustness of postoperative recurrence risk assessment for upper urinary tract stones. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a system structure diagram of a system for assessing the risk of recurrence after surgery for upper urinary tract stones based on multimodal data, provided in one embodiment of the present invention. Figure 2 This is a system structure diagram of a stone recurrence risk provided in one embodiment of the present invention; Figure 3 This is a schematic diagram of a computer device for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data, provided as an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a multimodal data-based upper urinary tract stone postoperative recurrence risk assessment system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the postoperative recurrence risk assessment system for upper urinary tract stones based on multimodal data provided by this invention.

[0020] Example 1: Please see Figure 1 The diagram illustrates a system block diagram of a multimodal data-based system for assessing the risk of recurrence after surgery for upper urinary tract stones, according to an embodiment of the present invention. The system includes: a data acquisition module 110, a modal contribution analysis module 120, a recurrence risk analysis module 130, and a recurrence risk assessment module 140.

[0021] The data acquisition module 110 is used to acquire the preliminary examination vectors of each patient with upper urinary tract stones, as well as the follow-up examination vectors of the patient to be tested and each patient with a history of recurrence. The two vectors are composed of medical examination data of various modalities; the patients with upper urinary tract stones are divided into training patients and validation patients.

[0022] The preoperative period refers to the time from when a patient decides to undergo surgical treatment until the end of the surgery. Medical imaging and biochemical data at any point during the preoperative period are extracted from electronic medical records for each patient with upper urinary tract stones. Medical imaging includes computed tomography (CT) images, Doppler ultrasound images, and X-ray images; biochemical data includes urinalysis indicators such as pH and red blood cell count, blood routine indicators such as white blood cell count and red blood cell count, and renal function indicators such as creatinine and blood urea nitrogen concentrations. Each independent source of data represents a modality; that is, Doppler ultrasound images, pH, white blood cell count, and creatinine concentration each represent a modality.

[0023] Numerical data such as pH value, white blood cell count, and creatinine concentration are directly standardized, and the standardized results are used as medical examination data for each modality. In this embodiment, range standardization is used, but other methods such as Z-score standardization, decimal scaling standardization, and sum-normalization can also be used. For image data, CT images and other image data are input into a pre-trained convolutional neural network, which outputs a fixed-dimensional feature vector. This feature vector represents the CNN features of the high-level semantic content of the image, and it is a numerical vector where each element represents the activation intensity of a specific visual feature. Then, the feature vector corresponding to each type of image data is L2 normalized, and the normalized result is used as medical examination data for each modality. Extracting image feature vectors using a convolutional neural network is a technique well-known to those skilled in the art and will not be elaborated upon here.

[0024] All patients with upper urinary tract stones are randomly divided into training patients and validation patients, with the number of training patients exceeding the number of validation patients. In one implementation of this invention, the ratio of training patients to validation patients is 9:1.

[0025] Upper urinary tract stones have a high risk of recurrence in the short term, and patients need regular follow-up examinations after surgery. "Patients to be tested" refers to patients who have previously had upper urinary tract stones and underwent surgery, while "patients with a history of recurrence" refers to patients who have previously had upper urinary tract stones and underwent surgery, but subsequently developed stones again. For patients with a history of recurrence, medical examination data in at least one modality is collected at any point during the follow-up period, both before the recurrence of stones after surgery and at each follow-up examination of patients to be tested. The method for obtaining the same modality of medical examination data during the follow-up period is the same as that before surgery. Typically, a comprehensive diagnosis of urinary stones requires a combination of medical data. Patients undergo detailed and comprehensive medical examinations before surgery to assess surgical effectiveness and risk. The comprehensive multimodal medical examination data before surgery can be considered the comprehensive basis for diagnosing the patient with stones. Due to economic and convenience factors, the medical examination items during the follow-up period are only a few of those used in the pre-operative period. Therefore, the modalities during the follow-up period are a subset of those used in the pre-operative period.

[0026] In this embodiment of the invention, the method for obtaining the pre-operative examination vector and the follow-up examination vector is as follows: Medical examination data of various modalities in patients with upper urinary tract stones before surgery are acquired; all modalities of medical examination data before surgery are arranged in a predetermined order to form the pre-operative examination vector; medical examination data of several modalities of the patient to be tested during the follow-up period are acquired; according to the predetermined order, all modalities of medical examination data during the follow-up period are placed in their corresponding positions, and the remaining modalities of medical examination data are set to the mean of the medical examination data of all training patients in that modality, thus generating the follow-up examination vector. It should be noted that in this embodiment, the predetermined order can be CT images, Doppler ultrasound images and X-ray images, pH value, red blood cell count, white blood cell count, creatinine concentration and urea nitrogen concentration. The follow-up examination vector and the pre-operative examination vector have the same dimension. Image-type medical examination data is a vector, and each element in this vector has one dimension in either the follow-up examination vector or the pre-operative examination vector. As an example, suppose the preoperative modalities include CT images and pH values. The medical examination data for the CT image modality is X = (a, b), and the medical examination data for the pH value is c. Then the preoperative examination vector is Q = (a, b, c), and the corresponding elements of the CT image modality in vector Q are a and b. Suppose the follow-up modality is pH values, then the follow-up examination vector is S = (…). , c) These are the mean values ​​of the elements in the first feature dimension (the dimension containing element a) and the second feature dimension (the dimension containing element b) of the CT image modality, respectively, from the preliminary examination vectors of all training patients.

[0027] The modal contribution analysis module 120 is used to train the neural network based on the stone status in the patient and the vector from previous examinations; and to obtain the predicted contribution weight of each modality based on the expected comprehensive load index of the verification patient and the expected local load index under each modality determined by the trained neural network.

[0028] Patients with upper urinary tract stones typically undergo comprehensive examinations before surgery. Data from different modalities of medical examinations can complement and validate each other, serving as direct evidence for assessing the patient's stone condition. Furthermore, the medical examination data for upper urinary tract patients is crucial for diagnosing urinary stones and can be used to train neural networks using vectors from these preliminary examinations. Neural networks are essentially environment-load mapping models. Their input is the physiological and biochemical stone-forming environment represented by multimodal medical examination data, and their output is the expected comprehensive load index generated under the corresponding environment. Through training with pre-operative data, the model learns and quantifies the correspondence between a specific bodily environment and its actual stone-forming capacity.

[0029] However, during follow-up, patients typically only undergo a few of the pre-operative examinations. This incompleteness of data directly leads to poor reliability of recurrence risk assessment based on follow-up examination data (i.e., the follow-up examination vector). Therefore, it is necessary to utilize the expected local burden index for patients in different modalities, analyze the importance of medical examination data for each modality in predicting recurrence risk, obtain the predictive contribution weight, and improve the reliability of risk assessment.

[0030] The recurrence risk analysis module 130 is used to adjust the expected local burden index of the patient under all modalities during the follow-up period using the prediction contribution weight to obtain the initial recurrence risk of the patient under the test; and to obtain the stone recurrence risk of the patient under the test based on the similarity of the patient under the test to the similarity of lifestyle and symptom characteristics with the patient under the test and the initial recurrence risk, as well as the initial recurrence risk.

[0031] Predictive contribution weights are used to measure the importance of medical examination data for each modality in predicting the final recurrence risk. The expected local load index is the risk of urinary stone recurrence determined based on medical examination data for each modality. Different modalities usually have different predictive abilities for recurrence risk. Adjusting the expected local load index using predictive contribution weights respects the independent value of each modality while considering its relative importance in specific clinical problems, thereby improving overall predictive performance and obtaining a more reliable initial recurrence risk level.

[0032] Patient descriptions of symptoms and lifestyle habits can indirectly reflect the recurrence rate of kidney stones. For example, unhealthy habits such as excessive consumption of sugary drinks and meat can increase uric acid levels in the body. Uric acid is a major component in the formation of kidney stones, thus increasing the risk of recurrence. Therefore, based on the initial recurrence risk, and combined with the similarity of lifestyle habits and symptom characteristics between the tested patient and patients with a history of recurrence, the risk of kidney stone recurrence can be determined.

[0033] The recurrence risk assessment module 140 is used to assess the risk of recurrence after surgery for upper urinary tract stones in the tested patients based on the stone recurrence risk level.

[0034] The risk of recurrence of kidney stones is based on a limited number of medical examinations, symptom descriptions, and lifestyle analysis of the patients during the follow-up period. It integrates the patient's static baseline risk and dynamic disease risk, thereby achieving a more comprehensive and accurate quantitative assessment of the individual's recurrence probability and effectively overcoming the problem of unreliable recurrence risk due to incomplete follow-up data.

[0035] Preferably, in some possible implementations of the embodiments of the present invention, the training method of the neural network includes: obtaining the volume of each stone in the body of a patient with upper urinary tract stones; normalizing the sum of the volumes of all stones in the body of the patient with upper urinary tract stones, and using the normalization result as a stone load index; taking the previous examination vector and the stone load index of the patient with upper urinary tract stones as input data and output data of a training sample in sequence; constructing a training set of the neural network from the training samples of all trained patients; and training the neural network using the training set.

[0036] The neural network takes a pre-examination vector as input and outputs a stone load index ranging from 0 to 1. During training, the neural network calculates predicted values ​​via forward propagation and uses a loss function to calculate the error between the predicted values ​​and the stone load index. Based on the mean squared error loss, it iteratively updates the connection weights and bias parameters of each layer of the neural network using backpropagation until the model converges. Essentially, the neural network constructs a mapping relationship from biochemical and imaging features of the body to the stone formation result. Since the physical stones in the patient's body have been removed post-surgery, the expected stone load index output by the model during the follow-up phase represents the maintenance of the current biochemical environment, which is theoretically sufficient to support the intensity of the induced stone load. The neural network is a deep neural network, and the loss function is the mean squared error function. The deep neural network and its training process are well-known techniques to those skilled in the art and will not be elaborated upon here.

[0037] It should be noted that since the volume and number of stones directly reflect the degree of mechanical obstruction of the urinary tract and the potential risk of damage to the renal parenchyma, they are core clinical indicators for assessing disease severity. Therefore, a stone burden index is analyzed based on the morphology, volume, and number of the patient's stones. A higher stone burden index indicates a more severe stone burden in the preoperative period, reflecting a stronger driving force of stone formation in the patient's physiological metabolism and biochemical environment, i.e., a higher potential for stone formation in the body. The volume and number of stones in the upper urinary tract can be determined through examinations such as ultrasound, CT, and MRI.

[0038] In this embodiment of the invention, the sum of the volumes of all stones in a patient with upper urinary tract stones is divided by a preset maximum stone reference volume to achieve normalization of the sum. The preset maximum stone reference volume is the maximum value of the sum of the accumulated volumes of all stones in a single patient within the neural network training sample set.

[0039] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the expected comprehensive load index and the expected local load index includes: inputting the pre-examination vector of the verification patient into a trained neural network, and the neural network outputting the expected comprehensive load index; selecting any modality as the example modality, setting all elements in the pre-examination vector of the verification patient except for the element corresponding to the example modality to zero, obtaining the local pre-examination vector under the example modality; inputting the local pre-examination vector into the trained neural network, and the neural network outputting the expected local load index of the verification patient under the example modality. It should be noted that the expected comprehensive load index simulates the process of integrating multiple medical examination results to quantify the overall stone-forming potential of the patient's current body environment; the larger the value, the more likely the urinary environment and biochemical state in the patient's body have the potential to support the formation of high-load stones. Setting all elements in the pre-examination vector except for the element corresponding to the example modality to the mean of the medical examination data of all trained patients in the corresponding modality can mask the feature information of other modalities, forcing the neural network to infer only based on the data of the example modality, thereby quantifying the local stone-forming potential reflected by the example modality alone, and obtaining the expected local load index. The higher the expected local load index, the stronger the internal driving force that triggers stone recurrence.

[0040] It should be noted that the expected local load indices for all modalities in the preoperative period were obtained using the same method as those for the example modalities.

[0041] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the predicted contribution weight includes: selecting verification patients whose absolute difference between the stone burden index and the expected comprehensive burden index is less than a preset difference threshold, and denoting them as analysis patients; constructing a risk matrix from the expected local burden indices of all analysis patients in all modalities, wherein the number of rows and columns of the risk matrix are respectively the number of analysis patients and the number of modal types in the preoperative period; constructing a column matrix from the expected comprehensive burden indices of all analysis patients, denoted as the comprehensive result matrix; setting a contribution weight column matrix, wherein the number of rows of the column matrix is ​​the number of modal types in the preoperative period; multiplying the risk matrix and the contribution weight column matrix to obtain a new matrix, calculating the sum of squares of the differences between the corresponding elements of the new matrix and the comprehensive result matrix, and using it as an error function; and taking each element in the contribution weight column matrix corresponding to the minimum value of the error function as the predicted contribution weight of the corresponding modality.

[0042] It should be noted that the expected comprehensive load index can be approximated by a linearly weighted combination of the expected local load indices of each modality. The neural network model performs reliably in patient analysis; using relevant patient data to analyze and predict contribution weights can reduce the impact of noise and outliers, ensuring the reliability of weight estimation. The contribution weight column matrix represents the weight of each modality, with elements representing unknowns. The sum of squares error is the most commonly used loss function in linear regression, measuring the overall deviation between the predicted and target values. Minimizing the error function finds the best-fit weights. The contribution weight column matrix that minimizes the error function can be determined using gradient descent or the normal equation method. Predicting contribution weights quantifies the importance of each modality in predicting stone recurrence, providing a key indicator of model interpretability. The initial recurrence risk is comprehensively quantified as the theoretically achievable stone load level under the current biochemical and physiological conditions during follow-up. The higher this value, the stronger the body's motivation to form stones again, i.e., the higher the initial risk.

[0043] Multiplying the risk matrix and the contribution weight column matrix yields the comprehensive result matrix, which results in an overdetermined system of equations—that is, the number of equations exceeds the number of unknowns. Solving this system of equations requires analyzing a larger number of patients than the number of modalities in the preoperative period. It is important to note that when minimizing the error function to solve for the contribution weight column matrix, the constraint is that each element in the contribution weight column matrix is ​​greater than or equal to 0, and the sum of all elements is 1. This maintains a consistent numerical scale before and after weighted fusion, ensuring that the initial recurrence risk remains within the original severity scale range.

[0044] In one implementation of this invention, the preset difference threshold is set as the 10th percentile of the absolute difference between the stone burden index and the expected comprehensive burden index of all verification patients. If the number of patients selected for analysis is less than or equal to the number of modalities in the preoperative period, then verification patients who do not meet the above conditions are selected as analysis patients in ascending order of absolute difference until the number of analysis patients is greater than the number of modalities.

[0045] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the initial recurrence risk includes: obtaining the expected local burden index of the patient under each modality during the follow-up period; weighting the expected local burden index of the patient under all modalities during the follow-up period with the predicted contribution weight, and using the ratio of the summation result to the sum of the predicted contribution weights of all modalities during the follow-up period as the initial recurrence risk. In a specific implementation of the embodiments of the present invention, the initial recurrence risk is expressed by the formula: In the formula, F represents the initial relapse risk of the patient being tested; N represents the number of modalities of the patient being tested during the follow-up period; The expected local burden index for the patient under the nth modality during the follow-up period; The predictive contribution weight for the nth modality is assigned to the patient during follow-up. It should be noted that modalities with higher predictive contribution weights are more important for predicting the risk of recurrence after urolithiasis surgery. The weighted sum of the expected local burden index and the predictive contribution weight for each modality is considered. The larger the value, the stronger the stone-forming driving force reflected by that mode, and the more significant its contribution to the recurrence risk. By weighting and ensuring that the recurrence assessment process focuses more on modes with high predictive value and reducing the influence of noisy modes, a more reliable initial recurrence risk score is obtained. The initial recurrence risk score comprehensively characterizes the stone load intensity that the patient's current stone-forming environment can theoretically support; the larger the value, the stronger the body's stone-forming potential, that is, the higher the risk of recurrence after upper urinary tract stone surgery in the patient being tested. This is used to renormalize the remaining weights after missing modes, so as to avoid the total weights involved in the calculation being less than 1 due to the missing parts of the follow-up period.

[0046] It is important to note that the methods for obtaining the expected local burden indices for both the test and validation patients during the follow-up period are the same; if If the value is zero, then the initial recurrence risk of the patient to be tested will be set to zero.

[0047] Please see Figure 2 The diagram illustrates a structural diagram of a stone recurrence risk level provided by an embodiment of the present invention. The system includes: a stone recurrence first situation analysis unit 131, a stone recurrence second situation analysis unit 132, a disease recurrence analysis unit 133, and a stone recurrence third situation analysis unit 134.

[0048] The first case analysis unit 131 for stone recurrence is used to determine the initial recurrence risk as the stone recurrence risk when the number of modal types of the patient during the follow-up period is equal to the number of modal types before surgery.

[0049] It should be noted that when the number of modal types during the follow-up period is equal to the number of modal types before surgery, it means that the patient under test has the same comprehensive data during the follow-up period as before surgery, and the initial recurrence risk is a reliable assessment result that can be directly used as the risk of stone recurrence.

[0050] The second case analysis unit 132 for stone recurrence is used to determine whether the initial recurrence risk is greater than or equal to the preset recurrence threshold when the number of modal types of the patient during the follow-up period is less than the number of modal types before the operation. If so, the initial recurrence risk is used as the stone recurrence risk.

[0051] It should be noted that when the number of modalities during the follow-up period is less than the number of modalities before surgery, if the initial recurrence risk is greater than or equal to the preset recurrence threshold, it means that although the data is incomplete, the model has given a high-risk signal based on the existing information, and the initial recurrence risk can be used as the stone recurrence risk.

[0052] In this embodiment of the invention, the same method as the initial recurrence risk of this scheme is used to obtain the initial recurrence risk of at least 100 patients with follow-up outcomes (recurrence or no recurrence) after upper urinary tract stone surgery. Receiver operating characteristic curve analysis is used to select the initial recurrence risk corresponding to the maximum Youden index as the preset recurrence threshold.

[0053] The disease recurrence analysis unit 133 is used to select matching patients from the historical recurrence patients of the test patient if no, and obtain the similarity of disease manifestations between the test patient and each matching patient based on the similarity of lifestyle habits and disease characteristics. The product of the difference between the preset recurrence threshold and the initial recurrence risk of each matching patient of the test patient and the similarity of disease manifestations is averaged to obtain the disease recurrence risk of the test patient.

[0054] In this embodiment of the invention, for the non-missing modalities shared by the matched patient and the test patient during the follow-up period, the absolute difference of the elements in the corresponding feature dimension of the follow-up examination vectors of the two patients is less than a preset element difference threshold for the corresponding dimension, and the initial relapse risk of the matched patient is less than a preset relapse threshold. The initial relapse risk of the matched patient and the test patient is obtained using the same method. It should be noted that the absolute difference of the elements in each dimension being less than the preset element difference threshold for the corresponding dimension excludes differences in relapse risk caused by different underlying conditions, ensuring that the matched patient and the test patient are highly similar in clinical indicators; the initial relapse risk of the matched patient being less than the preset relapse threshold reveals risk factors that are ignored by the risk model, hidden, or subsequently emerging, in order to more accurately identify the risk of relapse. It should be noted that if no matched patient meeting the matching criteria is found, the relapse risk of the test patient is set to zero.

[0055] In one implementation of this invention, the preset element difference threshold for each dimension in the follow-up examination vector is set to one-third of the range of elements in the same dimension in the follow-up examination vectors of all historical relapse patients.

[0056] In this embodiment of the invention, the method for obtaining the similarity of symptoms includes: obtaining the lifestyle vector and symptom vector of the patient to be tested and its matched patients, denoted as analysis vectors; calculating the cosine similarity of each analysis vector of the patient to be tested and each matched patient, and taking the mean of all cosine similarities as the similarity of symptoms between the patient to be tested and each matched patient.

[0057] It should be noted that the methods for obtaining the lifestyle habit vector and symptom vector are as follows: Lifestyle habit information is described through questionnaires, such as "How many milliliters of water do you drink daily?" and "What is the main source of your daily water intake? Options include beer, beverages, purified water, tea, etc." Patients need to fill out the questionnaires. A keyword database is built based on the questionnaire text, and the text data is converted into numerical feature vectors using a bag-of-words model, denoted as the lifestyle habit vector. Patients describe their symptom characteristics during the follow-up period, resulting in symptom text, such as pain often located in the lower back or flank, described as knife-like or colicky, accompanied by paleness, nausea, vomiting, and difficulty urinating. Symptom vectors are generated using the bag-of-words model. The greater the similarity of symptom manifestations, the more similar the symptom characteristics and lifestyle habits of the tested patient are to all matched patients. It should be noted that if all elements in the lifestyle habit vector and symptom vector are non-negative, the cosine similarity of the same analysis vector will range from 0 to 1.

[0058] When the initial recurrence risk is less than the preset recurrence threshold, external information is introduced for compensation. Stone recurrence is strongly correlated with lifestyle habits and symptom characteristics. By analyzing the similarity of lifestyle habits and symptom characteristics between the test patient and the matched patient, the potential recurrence tendency of urinary stones in the test patient is estimated. The assessment expands from relying solely on incomplete patient data to utilizing group experience and clinical knowledge, thus bridging the information gap. It is known that the more consistent the symptom characteristics and lifestyle habits are between the test patient and the matched patient, the higher the risk of recurrence after upper urinary tract stone surgery in the test patient. If a matched patient is initially diagnosed with recurrent kidney stones even when the initial recurrence risk is less than the preset recurrence threshold, and the sum of the actual recurrence risk based on symptom description and lifestyle changes and the initial recurrence risk exceeds the preset recurrence threshold, then the difference between the preset recurrence threshold and the initial recurrence risk of each matched patient can be considered as the actual recurrence risk based on symptom description and lifestyle changes. The more similar the symptom description and lifestyle of the matched patient and the tested patient, i.e., the greater the similarity of the disease presentation, the greater the proportion of these characteristics being converted into actual risk. The higher the recurrence risk of the tested patient after upper urinary tract stone surgery, and consequently the greater the disease recurrence risk.

[0059] The third analysis unit 134 for stone recurrence is used to take the sum of the initial recurrence risk and the disease recurrence risk of the patient as the stone recurrence risk of the patient.

[0060] It should be noted that the initial recurrence risk score is the risk of stone recurrence determined based on a small number of medical examinations performed on the patient during the follow-up period. The symptom recurrence risk score is the risk of stone recurrence determined based on the patient's symptom descriptions and lifestyle habits during the follow-up period. Patients with a higher initial recurrence risk score have a higher risk of recurrence after upper urinary tract stone surgery. It can comprehensively assess the patient's static baseline risk and dynamic symptom risk, thereby achieving a more comprehensive and accurate quantitative assessment of the individual recurrence probability.

[0061] In this embodiment of the invention, when the risk of stone recurrence is greater than or equal to the preset recurrence threshold, the patient under test is highly likely to have stone recurrence and needs further diagnosis by a doctor; otherwise, the probability of stone recurrence in the patient under test is low, and other comprehensive medical examinations and diagnoses can be carried out according to the actual situation.

[0062] This plan is based on the patient's multimodal medical examination data, lifestyle habits and symptom descriptions. The assessment indicators are generated through neural network models and linear weighted regression. It does not provide a medical conclusion on whether the patient has experienced recurrence of upper urinary tract stones after surgery. Instead, it is provided to medical staff as an objective quantitative reference. The final medical conclusion still needs to be judged by a professional physician in combination with the patient's actual clinical signs, imaging re-examination results and other necessary medical test results.

[0063] This invention is now complete.

[0064] Example 2: Figure 3 This is a schematic diagram of a computer device for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data, as provided in one embodiment of the present invention. For example,... Figure 3 As shown, the computer device includes: a memory 201, a processor 202, and a computer program 203 stored in the memory 201 and running on the processor 202, wherein when the processor 202 executes the computer program 203, the computer device can execute any of the aforementioned multimodal data-based upper urinary tract stone postoperative recurrence risk assessment systems.

[0065] Furthermore, embodiments of this application also protect an apparatus that may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to execute the postoperative recurrence risk assessment system for upper urinary tract stones based on multimodal data provided in embodiments of this application.

[0066] This embodiment can divide the device into functional modules based on the above system example. For example, each module can correspond to a separate functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0067] It should be understood that the device provided in this embodiment is used to perform the above-described system for assessing the risk of recurrence after surgery for upper urinary tract stones based on multimodal data, and therefore can achieve the same effect as the system described above.

[0068] When using integrated units, the device may include a processing module and a storage module. When applied to a workpiece, the processing module can be used to control and manage the workpiece's operations. The storage module can be used to support the execution of relevant program code by the workpiece.

[0069] The processing module may be a processor or a controller, which can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor may also be a combination of functions that implement computing capabilities, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and microprocessors, etc., and the storage module may be a memory.

[0070] Example 3: This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to implement the above embodiment's system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data.

[0071] In this embodiment, the device and computer-readable storage medium are used to execute the corresponding system provided above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects of the corresponding system provided above, and will not be repeated here.

[0072] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0073] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data, characterized in that, The system includes: The data acquisition module is used to obtain the preliminary examination vectors of each patient with upper urinary tract stones, as well as the follow-up examination vectors of the patients to be tested and each patient with a history of recurrence. Both types of vectors are composed of medical examination data of various modalities. Patients with upper urinary tract stones are divided into training patients and validation patients. The modal contribution analysis module is used to train the neural network based on the patient's stone status and previous examination vectors; and to obtain the predicted contribution weight of each modality based on the expected comprehensive load index of the verification patient and the expected local load index under each modality determined by the trained neural network. The recurrence risk analysis module is used to adjust the expected local burden index of the patient under the follow-up period in all modalities using the predicted contribution weight, and to obtain the initial recurrence risk of the patient under the test; based on the similarity of the patient under the test to the similarity of lifestyle and symptom characteristics with the patient under the test and the initial recurrence risk, the risk of stone recurrence of the patient under the test is obtained. The recurrence risk assessment module is used to assess the risk of recurrence after surgery for upper urinary tract stones in the tested patients based on the stone recurrence risk level.

2. The system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data according to claim 1, characterized in that, The method for obtaining the preliminary examination vector and the follow-up examination vector includes: Acquire medical examination data of various modalities in patients with upper urinary tract stones before surgery, and arrange all the medical examination data of the preoperative modalities in a predetermined order to form a preoperative examination vector; Acquire medical examination data of several modalities of the patient to be tested during the follow-up period. The modalities during the follow-up period are a subset of the modalities before surgery. According to the predetermined order, place the medical examination data of all modalities during the follow-up period in the corresponding positions, and set the medical examination data of each remaining modality as the mean of the medical examination data of all training patients in that modality to generate a follow-up examination vector.

3. The system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data according to claim 2, characterized in that, The training of the neural network includes: Obtain the volume of each stone in the body of the patient with upper urinary tract stones; normalize the sum of the volumes of all stones in the patient with upper urinary tract stones, and use the normalized result as the stone burden index. The preliminary examination vector of the patient with upper urinary tract stones and the stone burden index are used as input data and output data of a training sample, respectively; the training samples of all training patients constitute the training set of the neural network; and the neural network is trained using the training set.

4. The system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data according to claim 1, characterized in that, The method for obtaining the expected comprehensive burden index and the expected local burden index of the patient under each modality includes: The patient's preliminary examination vectors are input into the trained neural network, and the neural network outputs the expected comprehensive workload index. Choose any modality and denote it as the example modality. Set the remaining elements in the early examination vector of the verification patient, except for the elements corresponding to the example modality, to the mean of the medical examination data of all training patients in that modality, to obtain the local early vector under the example modality. Input the local early vector into the trained neural network, and the neural network outputs the expected local load index of the verification patient under the example modality.

5. The system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data according to claim 3, characterized in that, The step of obtaining the prediction contribution weight for each modality includes: Patients whose absolute difference between the stone burden index and the expected comprehensive burden index is less than a preset difference threshold are selected as the analysis patients; A risk matrix is ​​constructed from the expected local burden indices of all analyzed patients across all modalities. The number of rows and columns of the risk matrix are the number of analyzed patients and the number of modalities in the preoperative period, respectively. A column matrix is ​​constructed from the expected comprehensive burden indices of all analyzed patients, denoted as the comprehensive result matrix. Set up a contribution weight column matrix, where the number of rows in the column matrix is ​​equal to the number of modalities in the preoperative period; Multiply the risk matrix by the contribution weight column matrix to obtain a new matrix. Calculate the sum of squared differences between the corresponding elements of the new matrix and the comprehensive result matrix, and use this sum as the error function. Use each element in the contribution weight column matrix corresponding to the minimum value of the error function as the prediction contribution weight of the corresponding mode.

6. The system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data according to claim 4, characterized in that, The process of obtaining the initial relapse risk of the patient to be tested includes: Obtain the expected local burden index for each modality of the patient under test during the follow-up period; sum the expected local burden index of the patient under test in all modalities during the follow-up period with the predicted contribution weight, and use the ratio of the sum to the sum of the predicted contribution weights of all modalities during the follow-up period as the initial recurrence risk.

7. The system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data according to claim 1, characterized in that, The process of obtaining the risk of stone recurrence in the patient under test includes: When the number of modalities of the patient during the follow-up period is equal to the number of modalities before the operation, the initial recurrence risk is taken as the stone recurrence risk. When the number of modalities of the patient under test during the follow-up period is less than the number of modalities before surgery, it is determined whether the initial recurrence risk is greater than or equal to the preset recurrence threshold. If so, the initial recurrence risk is used as the stone recurrence risk. If not, a matching patient is selected from the patients with historical recurrence. Based on the similarity of the lifestyle and symptom characteristics between the patient under test and each matching patient, the similarity of the symptoms between the two patients is obtained. The product of the difference between the preset recurrence threshold and the initial recurrence risk of each matching patient of the patient under test and the similarity of the symptoms is averaged to obtain the recurrence risk of the patient under test. The sum of the initial recurrence risk and the disease recurrence risk of the patient to be tested is taken as the recurrence risk of the patient's stones.

8. The system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data according to claim 7, characterized in that, The process of obtaining the similarity of disease presentation between two corresponding patients includes: Obtain the lifestyle vectors and symptom vectors of the patient to be tested and their matching patients, and denote them as analysis vectors; Calculate the cosine similarity of each analysis vector between the test patient and each of their matched patients, and use the mean of all cosine similarities as the similarity of disease manifestations between the test patient and each of their matched patients.

9. A system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data, as described in claim 7, is characterized in that... The absolute difference between the elements of each dimension in the follow-up examination vector of the matched patient and the patient to be tested is less than the preset element difference threshold of the corresponding dimension, and the initial recurrence risk of the matched patient is less than the preset recurrence threshold.

10. A system for assessing the risk of recurrence after upper urinary tract stone surgery based on multimodal data, as described in claim 5, is characterized in that... The number of patients analyzed was greater than the number of modalities in the preoperative period.

Images