Prostate Cancer Recurrence Risk Prediction System Based on Multimodal Data

Abstract

The present invention relates to the field of medical data mining technology and proposes a prostate cancer recurrence risk prediction system based on multimodal data. The system comprises the following steps: collecting physiological parameters and discomfort moments of prostate cancer patients after surgical treatment, marking target discomfort moments and obtaining discomfort moments adjacent to the target discomfort moment; obtaining the rise-fall ratio of physiological parameters of the same type at the target discomfort moment; dividing and updating abnormal data display intervals; determining the comprehensive convolution kernel scale of the prostate cancer patient at the target discomfort moment based on the correlation between any two different types of physiological parameters, the duration of the abnormal data display interval of the physiological parameters, and the rise-fall ratio of all types of physiological parameters at the target discomfort moment; and predicting the prostate cancer recurrence risk of the prostate cancer patient based on the comprehensive convolution kernel scale. The present invention can effectively manage the risk of prostate cancer recurrence.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data mining, and in particular to a prostate cancer recurrence risk prediction system based on multimodal data. Background Art

[0002] While surgical treatment can remove visible tumors in prostate cancer patients, cancer cells can spread through the bloodstream or lymphatic system to various locations, such as bones and lymph nodes, resulting in residual cancer cells called micrometastases. Furthermore, the postoperative recovery of prostate cancer patients is affected not only by clinical heterogeneity and physiology, but also by factors such as their diet, sleep, and medication, which can increase the risk of prostate cancer recurrence. Therefore, timely detection and early intervention can significantly delay clinical tissue recurrence and improve the patient's recovery. Therefore, constructing a prostate cancer recurrence risk prediction model can guide subsequent treatment decisions and improve patient prognosis, and is therefore of great clinical significance.

[0003] Traditional prediction models such as the MSKC model can be used to extract the characteristics of physiological parameters of prostate cancer patients after surgery and predict the risk of prostate cancer recurrence. However, the receptive fields of these traditional prediction models are limited, and there is a problem of a single recognition pattern. They cannot accurately identify adverse events that have complex impacts on the patient's recovery status during the recovery process, nor can they effectively manage the risk of prostate cancer recurrence. Summary of the Invention

[0004] The present invention provides a prostate cancer recurrence risk prediction system based on multimodal data to address the problem that traditional prediction models are affected by the limited receptive field and single recognition mode when predicting prostate cancer recurrence risk, and are unable to effectively manage prostate cancer recurrence risk. The technical solutions adopted are as follows:

[0005] One embodiment of the present invention provides a prostate cancer recurrence risk prediction system based on multimodal data, which includes the following modules:

[0006] A data acquisition module is used to collect physiological parameters and discomfort moments of prostate cancer patients after surgical treatment, record any discomfort moment as a target discomfort moment, and obtain the discomfort moment adjacent to the target discomfort moment based on the time interval between the discomfort moment of the prostate cancer patient and the target discomfort moment;

[0007] a physiological parameter analysis module, configured to divide the target discomfort moment into an initial detection phase and an end detection phase, and to obtain a rise-fall ratio of the same type of physiological parameter of the prostate cancer patient at the target discomfort moment based on the difference between the initial detection phase and the end detection phase;

[0008] An abnormal data performance interval division module is used to divide the abnormal data performance interval according to the values of all physiological parameters close to the target discomfort moment and the corresponding collection time, and update the abnormal data performance interval according to the maximum values of the same type of physiological parameters collected within the abnormal data performance interval;

[0009] A convolution kernel scale determination module is used to determine the comprehensive convolution kernel scale of a prostate cancer patient at a target discomfort moment based on the correlation between any two different types of physiological parameters, the duration of the abnormal data display interval of the physiological parameters, and the rise and fall ratio of all types of physiological parameters at the target discomfort moment;

[0010] The recurrence risk prediction module is used to predict the recurrence risk of prostate cancer patients based on the comprehensive convolution kernel scale.

[0011] Furthermore, the method for obtaining the discomfort moment adjacent to the target discomfort moment is:

[0012] Select the discomfort moment that is closest in time to the target discomfort moment, and record the time interval between the selected discomfort moment and the target discomfort moment as the short duration of the target discomfort moment;

[0013] Select each discomfort moment contained in the short duration of the target discomfort moment respectively, and take all discomfort moments whose time interval with the selected discomfort moment is less than or equal to the short duration as the selected discomfort moments. Repeat the selection of the selected discomfort moments according to the method of time interval being less than or equal to the short duration until there are no new discomfort moments to be selected, and record all the selected discomfort moments as the adjacent discomfort moments of the target discomfort moment.

[0014] Furthermore, the specific method of dividing the target discomfort moment into the initial detection stage and the final detection stage includes:

[0015] The earliest adjacent discomfort moment of all adjacent discomfort moments of the target discomfort moment is set before the earliest adjacent discomfort moment of the prostate cancer patient. The time period corresponding to minutes is recorded as the initial detection phase of the target discomfort moment, where represents the second time threshold;

[0016] The latest adjacent discomfort moment of all adjacent discomfort moments of the target discomfort moment will be used to calculate the The time period corresponding to the target discomfort moment is recorded as the end detection phase.

[0017] Furthermore, the method for obtaining the rise-fall ratio of the same type of physiological parameter of the prostate cancer patient at the target discomfort moment is:

[0018] Calculating the first-order differences of the same type of physiological parameters of the prostate cancer patient at the initial detection stage, recording the ratio of the mean of the first-order differences of the same type of physiological parameters of the prostate cancer patient at the initial detection stage to the mean of the same type of physiological parameters as the first incremental ratio of the same type of physiological parameters of the prostate cancer patient at the initial detection stage; recording the sum of the first incremental ratio of the same type of physiological parameters and a constant 1 as the first incremental sum ratio of the same type of physiological parameters; recording the product of the mean of the same type of physiological parameters of the prostate cancer patient at the initial detection stage and the first incremental sum ratio as the first product of the same type of physiological parameters of the prostate cancer patient at the initial detection stage;

[0019] Calculating the first-order differences of the same type of physiological parameters of the prostate cancer patient at the final detection stage, recording the ratio of the mean of the first-order differences of the same type of physiological parameters of the prostate cancer patient at the final detection stage to the mean of the same type of physiological parameters as the second incremental ratio of the same type of physiological parameters of the prostate cancer patient at the final detection stage; recording the sum of the second incremental ratio of the same type of physiological parameters and a constant 1 as the second incremental sum ratio of the same type of physiological parameters; recording the product of the mean of the same type of physiological parameters of the prostate cancer patient at the final detection stage and the second incremental sum ratio as the second product of the same type of physiological parameters of the prostate cancer patient at the final detection stage;

[0020] Based on the first product of the same type of physiological parameters of the prostate cancer patient in the initial detection stage and the second product of the same type of physiological parameters in the final detection stage, the rise and fall ratio of the same type of physiological parameters of the prostate cancer patient at the target discomfort moment is obtained.

[0021] Furthermore, the method for dividing the abnormal data performance interval is:

[0022] Obtain the first-order difference values of all physiological parameters adjacent to the target discomfort moment, and record the maximum and second-largest first-order difference values of the same type of physiological parameters, as well as the collection time corresponding to the maximum value of the same type of physiological parameters, as the prominent collection time;

[0023] The longest time interval divided by all prominent collection moments is recorded as the abnormal data performance interval.

[0024] Furthermore, the abnormal data performance interval is updated according to the maximum values of the same type of physiological parameters collected within the abnormal data performance interval, including the specific method of:

[0025] Record any maximum value of the same type of physiological parameter collected within the abnormal data performance interval as the target maximum value; divide the abnormal data performance interval into two abnormal data performance sub-intervals according to the collection time corresponding to the target maximum value; record the average value of the difference between adjacent extreme values of the same type of physiological parameter collected within the abnormal data performance sub-interval as the first average value of the same type of physiological parameter in the abnormal data performance sub-interval; for the same type of physiological parameter, record the absolute value of the difference between the first average values of the two abnormal data performance sub-intervals divided by the target maximum value as the first absolute value of the target maximum value;

[0026] The set of all physiological parameters of the same type collected within the abnormal data performance subinterval is recorded as the parameter set of the same type in the abnormal data performance subinterval, and the Jaccard correlation coefficient of the parameter set of the same type of physiological parameters corresponding to the two abnormal data performance subintervals divided by the target maximum value is recorded as the correlation value of the target maximum value;

[0027] The mean of the time intervals between all adjacent extreme values of the physiological parameters collected within the abnormal data performance interval is recorded as the mean time interval of the abnormal data performance interval;

[0028] Determining, based on the first absolute value and the correlation value of the target maximum value and the time interval mean of the abnormal data performance interval, the degree of classification of the target maximum value within the abnormal data performance interval for the physiological parameters of the same type;

[0029] Based on the classification criteria of all types of physiological parameters, the abnormal data performance interval is updated.

[0030] Furthermore, the abnormal data performance interval is updated according to the classification basis of all types of physiological parameters, including:

[0031] The maximum value corresponding to the maximum value of the division basis degree is used as the division position, the abnormal data performance interval is divided into two abnormal data performance sub-intervals, and the abnormal data performance sub-interval where the maximum value of the physiological parameter of the type corresponding to the division basis degree is located is used as the new abnormal data performance interval.

[0032] Furthermore, the method of determining the comprehensive convolution kernel scale of a prostate cancer patient at a target discomfort moment based on the correlation between any two different types of physiological parameters, the duration of the abnormal data display interval of the physiological parameters, and the rise and fall ratio of all types of physiological parameters at the target discomfort moment includes the following specific methods:

[0033] determining expansion coefficients of the two different types of physiological parameters according to a correlation between the two different types of physiological parameters;

[0034] Any two different types of physiological parameters are respectively recorded as the first type of physiological parameter and the second type of physiological parameter, and the product of the expansion coefficient of the first type of physiological parameter and the second type of physiological parameter and the time length of the abnormal data performance interval of the second type of physiological parameter is recorded as the convolution kernel scale of the second type of physiological parameter determined based on the first type of physiological parameter;

[0035] The comprehensive convolution kernel scale of the prostate cancer patient at the target discomfort moment is determined based on the rise and fall ratios of all types of physiological parameters at the target discomfort moment and the convolution kernel scale of the second type of physiological parameters determined by different types of physiological parameters.

[0036] Furthermore, determining the comprehensive convolution kernel scale of the prostate cancer patient at the target discomfort moment based on the rise-fall ratio of all types of physiological parameters at the target discomfort moment and the convolution kernel scale of the second type of physiological parameter determined by different types of physiological parameters includes:

[0037] The product of the convolution kernel scale of the second type of physiological parameter determined based on the first type of physiological parameter and the rise-fall ratio of the first type of physiological parameter of the prostate cancer patient at the target discomfort moment is recorded as the fourth product of the second type of physiological parameter determined by the first type of physiological parameter. The average of the fourth products of the second type of physiological parameter determined by all physiological parameters different from the second type of physiological parameter is recorded as the adjusted convolution kernel scale of the second type of physiological parameter.

[0038] The product of the normalized value of the rise and fall ratio of the second type of physiological parameter at the target discomfort moment and the adjusted convolution kernel scale of the second type of physiological parameter is recorded as the fifth product of the second type of physiological parameter at the target discomfort moment, the cumulative sum of the fifth products of all types of physiological parameters at the target discomfort moment is recorded as the first cumulative sum at the target discomfort moment, and the odd number closest to the first cumulative sum at the target discomfort moment is recorded as the comprehensive convolution kernel scale of the prostate cancer patient at the target discomfort moment.

[0039] Furthermore, the method for predicting the risk of prostate cancer recurrence in prostate cancer patients based on the comprehensive convolution kernel scale includes the following specific methods:

[0040] The value of the comprehensive convolution kernel scale at the discomfort moment is used as the value of the convolution kernel scale, and the MSKC model is used to predict the prostate cancer recurrence risk of prostate cancer patients, so as to obtain the predicted value of the prostate cancer recurrence risk of prostate cancer patients at the discomfort moment.

[0041] The beneficial effects of the present invention are:

[0042] This application collects physiological parameters and discomfort moments of prostate cancer patients after surgical treatment. First, based on the change trend of physiological parameters of prostate cancer patients and the distribution of patients' approaching discomfort moments, the significance of the fluctuation of physiological parameters of the same type when prostate cancer patients feel discomfort is evaluated, and the rise and fall ratio of physiological parameters of the same type is obtained to avoid the subjectivity of prostate cancer patients recording according to their own physical feelings; among the physiological parameters of prostate cancer patients after surgical treatment, the physiological parameters with the most obvious data change amplitude often reflect the characteristics of the patient's body being affected. According to the values of physiological parameters, the abnormal data performance interval is divided; because prostate cancer patients before and after surgery have different physiological parameters, the physiological parameters of prostate cancer patients before and after surgery have different physiological parameters. The risk of prostate cancer recurrence is affected by many factors such as daily diet, sleep, and medication. There are potential correlations between these different types of influencing factors. This application conducts correlation analysis based on the abnormal data performance intervals of different types of physiological parameters, and obtains the comprehensive convolution kernel scale of prostate cancer patients at any moment of discomfort based on the correlation analysis results; finally, the risk of prostate cancer recurrence in prostate cancer patients is predicted based on the comprehensive convolution kernel scale; this solves the problem that when traditional prediction models predict the risk of prostate cancer recurrence, they are affected by the limitations of the receptive field, the recognition mode is single, and effective management of the risk of prostate cancer recurrence cannot be achieved. Among them, the comprehensive convolution kernel scale is the size of the receptive field determined by this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 A schematic diagram of a process flow of a prostate cancer recurrence risk prediction system based on multimodal data provided by one embodiment of the present invention;

[0045] Figure 2 A flowchart for obtaining the rise-fall ratio is provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] See also Figure 1, which shows a flow chart of a prostate cancer recurrence risk prediction system based on multimodal data provided by an embodiment of the present invention. The system includes: a data acquisition module, a physiological parameter analysis module, an abnormal data performance interval division module, a convolution kernel scale determination module, and a recurrence risk prediction module.

[0048] The data acquisition module collects physiological parameters and discomfort moments of prostate cancer patients after surgical treatment, records any discomfort moment as a target discomfort moment, and obtains the discomfort moment adjacent to the target discomfort moment based on the time interval between the discomfort moment of the prostate cancer patient and the target discomfort moment.

[0049] After prostate cancer patients undergo surgical treatment, their physiological parameters are continuously collected, including their heart rate, blood pressure, and emotional state scores. The collection interval for these parameters is 1 minute. The physiological parameters can be automatically extracted through a wristband worn by the prostate cancer patients.

[0050] In actual application, as other implementation methods, the implementer can decide the type of data included in the physiological parameters of prostate cancer patients and the value of the time interval for collecting the physiological parameters of prostate cancer patients based on actual conditions. This application does not impose any special restrictions.

[0051] It should be noted that, for ease of calculation, all physiological parameters of prostate cancer patients involved in the calculations in this embodiment have undergone data preprocessing to eliminate the effects of dimensioning. This embodiment uses the Z-Score standard normalization method to remove dimension from the same type of physiological parameters of prostate cancer patients. In actual applications, implementers may use other existing methods such as the maximum and minimum normalization method to remove dimension, and this is not limited here.

[0052] The subjective discomfort experienced by prostate cancer patients after surgery is potentially associated with cancer recurrence. After prostate cancer surgery, especially prostatectomy, patients may experience symptoms such as frequent urination, urgency, pain during urination, and pelvic pain. These discomforts may be related to recovery from prostate surgery, complications, or tumor recurrence. For example, pain or discomfort in the pelvic area may indicate local tumor recurrence or metastasis to the bones. Furthermore, postoperative symptoms such as difficulty urinating or blood in the urine may also indicate urinary system abnormalities, potentially indicating local cancer recurrence.

[0053] Therefore, in order to fully evaluate the physical condition of prostate cancer patients after surgery, when prostate cancer patients feel unwell, they use smart portable recorders to record the moments when they feel unwell, and all moments when they feel unwell are recorded as moments of discomfort.

[0054] The persistence of physical discomfort felt by prostate cancer patients was evaluated based on the temporal continuity and time intervals of the discomfort moments of prostate cancer patients.

[0055] Any discomfort moment of a prostate cancer patient is recorded as the target discomfort moment, and the time interval between the target discomfort moment and the target discomfort moment is calculated as The number of all discomfort moments in the time period consisting of the target discomfort moment is selected, and the discomfort moment closest in time to the target discomfort moment is selected. The time interval between the selected discomfort moment and the target discomfort moment is recorded as the short duration of the target discomfort moment. The discomfort moments included in the short duration of the target discomfort moment are selected. For each selected discomfort moment, all discomfort moments with a time interval less than or equal to the short duration are selected as the selected discomfort moments. The method of selecting the selected discomfort moments with a time interval less than or equal to the short duration is repeated until no new discomfort moments can be selected. All selected discomfort moments are recorded as the adjacent discomfort moments of the target discomfort moment.

[0056] in, Indicates the first time threshold. In this embodiment, the value of the first time threshold is 12 hours. For ease of understanding, when the target discomfort time is 8:00 on May 4, the time interval with the target discomfort time is The time is 20:00 on May 3 and 20:00 on May 4, and the composed time period is from 20:00 on May 3 to 20:00 on May 4.

[0057] It is understood that, generally, there is only one discomfort moment that is temporally closest to the target discomfort moment. In this case, only one discomfort moment is selected within the short duration of the target discomfort moment. However, there may be two discomfort moments that are temporally closest to the target discomfort moment, i.e., one discomfort moment each within a short duration before and after the target discomfort moment. Therefore, when selecting discomfort moments that are temporally closest to the target discomfort moment, it is necessary to analyze each selected discomfort moment separately. In other words, if there are two discomfort moments that are temporally closest to the target discomfort moment, both of these closest discomfort moments are analyzed separately.

[0058] It is understandable that the target discomfort moment may correspond to one adjacent discomfort moment or may correspond to multiple adjacent discomfort moments.

[0059] The same method can be used to obtain the time of discomfort adjacent to any discomfort time of a prostate cancer patient.

[0060] At this point, the physiological parameters of the prostate cancer patient and the times approaching the discomfort moments of all discomfort moments are obtained.

[0061] The physiological parameter analysis module divides the target discomfort moment into the initial detection stage and the final detection stage, and obtains the rise and fall ratio of the same type of physiological parameters of prostate cancer patients at the target discomfort moment based on the differences between the same type of physiological parameters of prostate cancer patients in the initial detection stage and the differences between the same type of physiological parameters in the final detection stage.

[0062] When prostate cancer patients experience physical discomfort, the discomfort often corresponds to fluctuations in physiological parameters. For example, when prostate cancer patients experience pain, anxiety, fever, or other stress reactions, their heart rate, blood pressure, and other data will increase. As the prostate cancer patient's body recovers, their heart rate, blood pressure, and other data will gradually decrease. In particular, at the beginning and end of the prostate cancer patient's physical discomfort, physiological parameters often show an upward trend and a downward trend. Therefore, based on the changing trends of the prostate cancer patient's physiological parameters, the discomfort moments and the adjacent discomfort moments can be further divided to avoid the subjectivity of prostate cancer patients recording based on their own physical feelings.

[0063] The earliest adjacent discomfort moment of all adjacent discomfort moments of the target discomfort moment is set before the earliest adjacent discomfort moment of the prostate cancer patient. The time period corresponding to the target discomfort moment is recorded as the starting detection phase, and the latest discomfort moment of all the discomfort moments approaching the target discomfort moment is recorded as the starting detection phase of the target discomfort moment. The time period corresponding to the target discomfort moment is recorded as the end detection phase.

[0064] in, represents the second time threshold. In this embodiment, the value of the second time threshold is 30. For example, when the earliest time of discomfort among all the times close to the target discomfort time for a prostate cancer patient is 2:00, 1:30-2:00 is the initial detection phase. When the latest time of discomfort among all the times close to the target discomfort time for a prostate cancer patient is 13:00, 13:00-13:30 is the final detection phase.

[0065] Calculate the first-order differences of the same type of physiological parameters of the prostate cancer patient at the initial detection stage, and record the ratio of the mean of the first-order differences of the same type of physiological parameters of the prostate cancer patient at the initial detection stage to the mean of the same type of physiological parameters as the first incremental ratio of the same type of physiological parameters of the prostate cancer patient at the initial detection stage. The sum of the first incremental ratio of the same type of physiological parameters and a constant 1 is recorded as the first incremental sum ratio of the same type of physiological parameters; and the product of the mean of the same type of physiological parameters of the prostate cancer patient at the initial detection stage and the first incremental sum ratio is recorded as the first product of the same type of physiological parameters of the prostate cancer patient at the initial detection stage.

[0066] Calculate the first-order differences of the same type of physiological parameters of the prostate cancer patient at the final detection stage, and record the ratio of the mean of the first-order differences of the same type of physiological parameters of the prostate cancer patient at the final detection stage to the mean of the same type of physiological parameters as the second incremental ratio of the same type of physiological parameters of the prostate cancer patient at the final detection stage. The sum of the second incremental ratio of the same type of physiological parameters and a constant 1 is recorded as the second incremental sum ratio of the same type of physiological parameters; and the product of the mean of the same type of physiological parameters of the prostate cancer patient at the final detection stage and the second incremental sum ratio is recorded as the second product of the same type of physiological parameters of the prostate cancer patient at the final detection stage.

[0067] The calculation of the first-order difference is a well-known technique and will not be described in detail.

[0068] Based on the first product of the same type of physiological parameters of the prostate cancer patient in the initial detection stage and the second product of the same type of physiological parameters in the final detection stage, the rise and fall ratio of the same type of physiological parameters of the prostate cancer patient at the target discomfort moment is obtained.

[0069] The rise-fall ratio is positively correlated with the first product and negatively correlated with the second product.

[0070] It can be understood that the positive correlation and negative correlation in this application refer to the relationship between the independent variable and the dependent variable. The positive correlation is that the dependent variable increases (decreases) as the independent variable increases (decreases), which can be an additive relationship, a multiplicative relationship, etc.; the negative correlation is that the dependent variable decreases (increases) as the independent variable increases (decreases), which can be an inverse relationship, a subtractive relationship, etc.

[0071] Some other embodiments of the present application may be to record the ratio of the first product of the same type of physiological parameters of prostate cancer patients in the initial detection stage to the second product in the final detection stage as the rise and fall ratio of the same type of physiological parameters of prostate cancer patients at the target discomfort moment.

[0072] The rise-fall ratio reflects the significance of the fluctuation of physiological parameters corresponding to the rise-fall ratio when prostate cancer patients feel unwell. Figure 2 shown.

[0073] It is understandable that for any type of physiological parameter of a prostate cancer patient at a target discomfort moment, there is a corresponding rise-fall ratio. The rise-fall ratio corresponding to any type of physiological parameter of a prostate cancer patient at any discomfort moment can be obtained using the same method.

[0074] Thus, the rise-fall ratio corresponding to any type of physiological parameter of the prostate cancer patient at any discomfort moment is obtained.

[0075] The abnormal data performance interval division module divides the abnormal data performance interval according to the values of all physiological parameters close to the target discomfort moment and the corresponding collection time, and updates the abnormal data performance interval according to the maximum values of the same type of physiological parameters collected within the abnormal data performance interval.

[0076] The risk of prostate cancer recurrence after surgery is influenced by multiple factors, including diet, sleep, and medications, and there are potential correlations between these different factors. Among the physiological parameters of prostate cancer patients after surgery, those with the most significant changes often reflect the impact on the patient's body.

[0077] Obtain the first-order difference values of all physiological parameters adjacent to the target discomfort moment, record the maximum value and the second largest value of the first-order difference values of the same type of physiological parameters, and the collection time corresponding to the maximum value of the same type of physiological parameters as prominent collection times, and record the time interval divided from the earliest moment to the latest moment of all prominent collection times as the abnormal data performance interval.

[0078] It can be understood that the first-order difference value of the physiological parameter near the moment of discomfort is the difference between the physiological parameter near the moment of discomfort and the physiological parameter of the same type collected at the previous moment of discomfort.

[0079] Any maximum value of the same type of physiological parameter collected within the abnormal data performance interval is recorded as the target maximum value, the collection time corresponding to the target maximum value is used as the dividing position, and the collection time corresponding to the target maximum value is used to divide the abnormal data performance interval into two abnormal data performance sub-intervals. The average value of the difference between adjacent extreme values of the same type of physiological parameter collected within the abnormal data performance sub-interval is recorded as the first average value of the same type of physiological parameter in the abnormal data performance sub-interval. For the same type of physiological parameter, the absolute value of the difference between the first average values of the two abnormal data performance sub-intervals divided by the target maximum value is recorded as the first absolute value of the target maximum value; the set consisting of all the same type of physiological parameters collected within the abnormal data performance sub-interval is recorded as the same type parameter set of the abnormal data performance sub-interval, and the Jaccard correlation coefficient of the same type parameter set corresponding to the same type of physiological parameter of the two abnormal data performance sub-intervals divided by the target maximum value is recorded as the correlation value of the target maximum value; the mean value of the time intervals between all adjacent extreme values of the physiological parameter collected within the abnormal data performance interval is recorded as the time interval mean of the abnormal data performance interval.

[0080] The calculation of the Jaccard correlation coefficient is a well-known technique and will not be described in detail.

[0081] According to the first absolute value and the correlation value of the target maximum value and the time interval mean of the abnormal data performance interval, the basis degree for dividing the target maximum value in the abnormal data performance interval into the physiological parameters of the same type is determined.

[0082] It can be understood that the division basis degree is positively correlated with the first absolute value and the time interval mean, and negatively correlated with the correlation value of the target maximum value.

[0083] Preferably, as an embodiment of the present application, the product of the first absolute value of the target maximum value and the mean time interval of the abnormal data performance interval is recorded as the third product of the target maximum value for the same type of physiological parameters within the abnormal data performance interval, and the ratio of the third product of the target maximum value for the same type of physiological parameters within the abnormal data performance interval to the correlation value of the target maximum value is recorded as the basis for dividing the target maximum value for the same type of physiological parameters within the abnormal data performance interval.

[0084] The same method can be used to obtain the division basis degree of any maximum value of the same type of physiological parameters collected within the abnormal data performance interval.

[0085] The maximum value corresponding to the maximum value of the division basis degree is used as the division position, and the abnormal data performance interval is divided into two abnormal data performance sub-intervals. The abnormal data performance sub-interval where the maximum value of the physiological parameter of the type corresponding to the division basis degree is located is used as the new abnormal data performance interval to achieve the update of the abnormal data performance interval.

[0086] At this point, the abnormal data performance intervals corresponding to all types of physiological parameters are obtained and updated.

[0087] The convolution kernel scale determination module determines the comprehensive convolution kernel scale of prostate cancer patients at the target discomfort moment based on the correlation between any two different types of physiological parameters, the time length of the abnormal data performance interval of the physiological parameters, and the rise and fall ratio of all types of physiological parameters at the target discomfort moment.

[0088] Furthermore, correlation analysis is performed based on the abnormal data performance intervals of different types of physiological parameters.

[0089] Physiological parameters of the same type for the same prostate cancer patient are arranged in the order of their collection time to obtain a physiological parameter sequence of the same type of physiological parameters for the same prostate cancer patient. The Pearson correlation coefficient between the physiological parameter sequences of two different types of physiological parameters is recorded as the overall correlation of the two different types of physiological parameters. The Jaccard correlation coefficient of two sets of two different types of physiological parameters is recorded as the interval intersection of the two different types of physiological parameters.

[0090] When the interval intersection of two different types of physiological parameters is smaller, the values of these two different types of physiological parameters deviate from the changing trend of all physiological parameters of prostate cancer patients. Therefore, the interval expansion of the physiological parameters corresponding to the smaller interval intersection should be carried out to capture more potential human discomfort data information.

[0091] The ratio of the interval intersection degree and the overall correlation degree of two different types of physiological parameters is recorded as the first ratio of the two different types of physiological parameters; when the first ratio of the two different types of physiological parameters is greater than or equal to When the expansion coefficient of the two different types of physiological parameters is assigned to 0, when the first ratio of the two different types of physiological parameters is less than When , the expansion coefficients of the two different types of physiological parameters are assigned to the first ratio of the two different types of physiological parameters.

[0092] in, Indicates the third time threshold. In this embodiment, the value of the third time threshold is 1.

[0093] Any two different types of physiological parameters are respectively recorded as the first type of physiological parameters and the second type of physiological parameters, and the product of the expansion coefficient of the first type of physiological parameters and the second type of physiological parameters and the time length of the abnormal data performance interval of the second type of physiological parameters is recorded as the convolution kernel scale of the second type of physiological parameters determined based on the first type of physiological parameters.

[0094] At this point, the convolution kernel scales of other different types of physiological parameters determined by any one type of physiological parameter are obtained.

[0095] The comprehensive convolution kernel scale of the prostate cancer patient at the target discomfort moment is determined based on the rise and fall ratios of all types of physiological parameters at the target discomfort moment and the convolution kernel scale of the second type of physiological parameters determined by different types of physiological parameters.

[0096] The product of the convolution kernel scale of the second type of physiological parameter determined based on the first type of physiological parameter and the rise and fall ratio of the first type of physiological parameter of the prostate cancer patient at the target discomfort moment is recorded as the fourth product of the second type of physiological parameter determined by the first type of physiological parameter, and the average of the fourth products of the second type of physiological parameter determined by all physiological parameters different from the second type of physiological parameter is recorded as the adjusted convolution kernel scale of the second type of physiological parameter.

[0097] The product of the normalized value of the rise and fall ratio of the second type of physiological parameter at the target discomfort moment and the adjusted convolution kernel scale of the second type of physiological parameter is recorded as the fifth product of the second type of physiological parameter at the target discomfort moment, the cumulative sum of the fifth products of all types of physiological parameters at the target discomfort moment is recorded as the first cumulative sum at the target discomfort moment, and the odd number closest to the first cumulative sum at the target discomfort moment is recorded as the comprehensive convolution kernel scale of the prostate cancer patient at the target discomfort moment.

[0098] The same method can be used to obtain the comprehensive convolution kernel scale of prostate cancer patients at any discomfort moment.

[0099] The recurrence risk prediction module predicts the recurrence risk of prostate cancer patients based on the comprehensive convolution kernel scale.

[0100] The value of the comprehensive convolution kernel scale at the discomfort moment is used as the value of the convolution kernel scale, and the MSKC model is used to predict the prostate cancer recurrence risk of prostate cancer patients, so as to obtain the predicted value of the prostate cancer recurrence risk of prostate cancer patients at the discomfort moment.

[0101] Among them, the MSKC model is a well-known prediction model. Using the MSKC model to predict the risk of prostate cancer recurrence in prostate cancer patients is a well-known technology and will not be described in detail.

[0102] At this point, the risk of prostate cancer recurrence in prostate cancer patients can be predicted.

[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A prostate cancer recurrence risk prediction system based on multimodal data, characterized by: The system includes the following modules: A data acquisition module is used to collect physiological parameters and discomfort moments of prostate cancer patients after surgical treatment, record any discomfort moment as a target discomfort moment, select the discomfort moment closest in time to the target discomfort moment, and record the time interval between the selected discomfort moment and the target discomfort moment as the short duration of the target discomfort moment; Select each discomfort moment within the short duration of the target discomfort moment, and take all discomfort moments whose time interval with the selected discomfort moment is less than or equal to the short duration as the selected discomfort moments. Repeat the selection process for the selected discomfort moments according to the method of time interval less than or equal to the short duration until there are no new discomfort moments to be selected. All selected discomfort moments are recorded as the adjacent discomfort moments of the target discomfort moment. The physiological parameter analysis module is used to analyze the earliest discomfort moment of all the discomfort moments approaching the target discomfort moment in the prostate cancer patient. The time period corresponding to minutes is recorded as the initial detection phase of the target discomfort moment, where Represents the second time threshold; the prostate cancer patient is taken to be after the latest of all the adjacent discomfort moments of the target discomfort moment The time period corresponding to the minutes was recorded as the end detection phase of the target discomfort moment; Calculate the first-order difference of the same type of physiological parameters of the prostate cancer patient in the initial detection stage, and record the ratio of the mean of the first-order difference of the same type of physiological parameters of the prostate cancer patient in the initial detection stage to the mean of the same type of physiological parameters as the first incremental ratio of the same type of physiological parameters of the prostate cancer patient in the initial detection stage; record the sum of the first incremental ratio of the same type of physiological parameters and a constant 1 as the first incremental sum ratio of the same type of physiological parameters; record the product of the mean of the same type of physiological parameters of the prostate cancer patient in the initial detection stage and the first incremental sum ratio as the first product of the same type of physiological parameters of the prostate cancer patient in the initial detection stage; calculate the same type of physiological parameters of the prostate cancer patient in the final detection stage. First-order differences of parameters, recording the ratio of the mean of the first-order differences of the same type of physiological parameters of the prostate cancer patient at the end of the detection stage to the mean of the same type of physiological parameters as the second incremental ratio of the same type of physiological parameters of the prostate cancer patient at the end of the detection stage; recording the sum of the second incremental ratio of the same type of physiological parameters and a constant 1 as the second incremental sum ratio of the same type of physiological parameters; recording the product of the mean of the same type of physiological parameters of the prostate cancer patient at the end of the detection stage and the second incremental sum ratio as the second product of the same type of physiological parameters of the prostate cancer patient at the end of the detection stage; obtaining the rise and fall ratio of the same type of physiological parameters of the prostate cancer patient at the target discomfort moment based on the first product and the second product; The abnormal data manifestation interval division module is used to obtain the first-order difference values of all physiological parameters adjacent to the target discomfort moment, and record the maximum value and the second largest value of the first-order difference values of the same type of physiological parameters, as well as the collection time corresponding to the maximum value of the same type of physiological parameters, as prominent collection times; the longest time interval divided by all prominent collection times is recorded as the abnormal data manifestation interval, and any maximum value of the same type of physiological parameters collected within the abnormal data manifestation interval is recorded as the target maximum value; according to the collection time corresponding to the target maximum value, the abnormal data manifestation interval is divided into two abnormal data manifestation sub-intervals, and the average value of the difference between adjacent extreme values of the same type of physiological parameters collected within the abnormal data manifestation sub-interval is recorded as the first average value of the same type of physiological parameters in the abnormal data manifestation sub-interval; for the same type of physiological parameters, the absolute value of the difference between the first average values of the two abnormal data manifestation sub-intervals divided by the target maximum value is recorded as the first average value of the target maximum value. an absolute value; a set consisting of all physiological parameters of the same type collected within the abnormal data performance subinterval is recorded as the set of parameters of the same type of the abnormal data performance subinterval, and the Jaccard correlation coefficient of the set of parameters of the same type of physiological parameters corresponding to the two abnormal data performance subintervals divided by the target maximum value is recorded as the correlation value of the target maximum value; the mean of the time intervals between all adjacent extreme values of the physiological parameters collected within the abnormal data performance interval is recorded as the mean time interval of the abnormal data performance interval; according to the first absolute value and the correlation value of the target maximum value and the mean time interval of the abnormal data performance interval, the division basis degree of the target maximum value within the abnormal data performance interval for the physiological parameters of the same type is determined; the maximum value corresponding to the maximum value of the division basis degree is used as the division position, the abnormal data performance interval is divided into two abnormal data performance subintervals, and the abnormal data performance subinterval where the maximum value of the physiological parameter of the type corresponding to the division basis degree is located is used as the new abnormal data performance interval; The convolution kernel scale determination module is used to determine the expansion coefficients of two different types of physiological parameters according to the correlation between the two different types of physiological parameters; record any two different types of physiological parameters as the first type of physiological parameter and the second type of physiological parameter, respectively; record the product of the expansion coefficients of the first type of physiological parameter and the second type of physiological parameter and the time length of the abnormal data performance interval of the second type of physiological parameter as the convolution kernel scale of the second type of physiological parameter determined based on the first type of physiological parameter; record the product of the convolution kernel scale of the second type of physiological parameter determined based on the first type of physiological parameter and the rise and fall ratio of the first type of physiological parameter of the prostate cancer patient at the target discomfort moment as the first type of physiological parameter. The fourth product of the second type of physiological parameter determined by the parameter, and the mean of the fourth products of the second type of physiological parameters determined by all physiological parameters different from the second type of physiological parameters, are recorded as the adjusted convolution kernel scale of the second type of physiological parameter; the product of the normalized value of the rise and fall ratio of the second type of physiological parameter at the target discomfort moment and the adjusted convolution kernel scale of the second type of physiological parameter are recorded as the fifth product of the second type of physiological parameter at the target discomfort moment, the cumulative sum of the fifth products of all types of physiological parameters at the target discomfort moment is recorded as the first cumulative sum at the target discomfort moment, and the odd number closest to the first cumulative sum at the target discomfort moment is recorded as the comprehensive convolution kernel scale of the prostate cancer patient at the target discomfort moment; The recurrence risk prediction module is used to predict the recurrence risk of prostate cancer patients based on the comprehensive convolution kernel scale.

2. The prostate cancer recurrence risk prediction system based on multimodal data according to claim 1, characterized in that: The method for predicting the risk of prostate cancer recurrence in prostate cancer patients based on the comprehensive convolution kernel scale includes the following specific methods: The value of the comprehensive convolution kernel scale at the discomfort moment is used as the value of the convolution kernel scale, and the MSKC model is used to predict the prostate cancer recurrence risk of prostate cancer patients, so as to obtain the predicted value of the prostate cancer recurrence risk of prostate cancer patients at the discomfort moment.

Images