In-situ post-neocystectomy urinary function recovery prediction method based on multi-modal data

Abstract

This invention relates to the field of postoperative functional recovery prediction technology, specifically a method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data. The method includes the following steps: acquiring volume, urine flow rate, and urethral pressure data and synchronizing them in time; organizing data according to the voiding cycle in conjunction with imaging information; dividing the voiding cycle into stages and assessing the recovery progress; and determining whether the recovery trend has deviated. In this invention, by synchronizing and integrating volume, urine flow rate, and urethral pressure data in time and organizing them uniformly with the imaging sequence according to the voiding cycle, multi-source information is continuously correlated within the same time axis. Based on the ratio of volume to flow rate and pressure abnormality markers, the dynamic changes in the voiding process are characterized. The recovery progress is assessed by combining stage division and trend fitting, and the stability and deviation of the continuous cycle trend are identified. This transforms the judgment of voiding function recovery from a static conclusion to a dynamic prediction with sustainable tracking, improving individual adaptability and clinical early warning value.

Description

Technical Field

[0001] This invention relates to the field of postoperative functional recovery prediction technology, and in particular to a method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data. Background Technology

[0002] Postoperative functional recovery prediction technology belongs to the field of medical information processing and clinical decision support technology. It involves data collection, data integration, information analysis, and prediction methods for assessing and judging the recovery of specific physiological functions in patients after surgical treatment. This technology is based on clinical diagnostic data, combined with imaging data, physiological test results, laboratory test indicators, and follow-up records. Through systematic processing of multi-source medical data, it supports doctors in judging and analyzing the patient's postoperative recovery process. Specifically, traditional orthotopic neobladder surgery voiding function recovery prediction refers to the process of assessing the recovery of a patient's voiding ability after cystectomy and orthotopic neobladder reconstruction. Based on various types of medical data generated before and after surgery, it predicts the state of voiding function recovery. Traditional methods rely on single or limited data sources such as urodynamic testing results, imaging results, clinical scoring scales, and medical records. Doctors make comprehensive judgments based on past clinical experience, statistical analysis results, and fixed evaluation indicators to predict the recovery of voiding function after orthotopic neobladder surgery.

[0003] Traditional methods for predicting urinary function recovery after orthotopic neobladder surgery rely heavily on limited information sources. The data often deviates from a continuous process representation, with urodynamic testing and imaging performed at different time points. Results are presented as static readings or report conclusions, failing to reflect the transitions in the coupling relationship between volume and flow rate during a single voiding cycle, and also struggling to capture short-term correspondences between pressure abnormalities and flow changes. Clinical scoring scales and medical records are primarily subjective descriptions and discrete grading, influenced by filling habits and observation conditions, resulting in weak cross-institutional comparability. Statistical analysis is often based on fixed evaluation indicators and population averages, limiting the scope of judgment when individual recovery paths diverge. Experience-based judgments depend on the operator's ability to identify abnormalities; when indicators conflict, there is a lack of traceable temporal evidence, leading to differing conclusions among different doctors for the same case. Follow-up is centered around outpatient visits, lacking continuous verification of trend consistency and stability. Recovery progress is easily misjudged as normal fluctuations in short-term fluctuations or sudden reversals, leading to delayed intervention. For example, when high-pressure compensation maintains a stable surface urine flow rate, a single test reading is close to the threshold range, and the report tends to be conservative. The actual potential risks are exposed in the following weeks, such as worsening nocturnal incontinence or decreased emptying efficiency, resulting in increased frequency of follow-up visits and repeated adjustments to the treatment plan. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data, comprising the following steps: S1: Acquire volume, urine flow rate and urethral pressure data generated by the monitoring device, integrate them in time synchronization, and establish a data structure with the imaging image sequence in units of voiding cycle to form voiding function cycle data group; S2: Based on the volume and urine flow rate change curves of the voiding function cycle data set, calculate the proportional relationship of continuous time segments, and mark abnormal segments in combination with pressure parameters to construct the volume-urine flow relationship and generate a volume-urine flow interaction structure. S3: Based on the sequence trend characteristics of the volume-flow interaction structure, and combined with the stage division rules of volume regulation capacity and pressure response performance in the reference index system, the voiding cycle is divided into stages, the time intervals and trend attributes corresponding to the stages are defined, and a voiding function recovery stage interval table is generated. S4: Call the trend feature sequence in the interval table of the urination function recovery stage, evaluate the fit with the standard trend sample, divide the patient's current recovery progress segment, and generate individual urination function recovery trend prediction results; S5: Continuously compare the predicted trend of individual urination function recovery with the current cycle prediction result to determine the stable state of trend change and whether there is a trend reversal or exceeding the judgment range, and generate a predicted trend deviation status.

[0005] As a further aspect of the present invention, the voiding function cycle data set includes a volume synchronization sequence, a urine flow rate synchronization sequence, a urethral pressure synchronization sequence, an imaging frame alignment index, a voiding cycle identifier, and cycle start and end timestamps; the volume-urine flow interaction structure includes a volume-urine flow ratio feature sequence, a pressure abnormality interval marker, an interactive continuous feature vector, a feature sequence smoothing parameter, and an interactive segment boundary marker; the voiding function recovery stage interval table includes a stage type label, a stage time segment, a stage trend direction attribute, a stage change amplitude attribute, and a stage rule matching level; the individual voiding function recovery trend prediction result includes a recovery progress segment label, a trend direction judgment value, a change amplitude prediction interval, a standard sample fit score, and a prediction confidence level; the prediction trend deviation status includes a trend coherence index, a trend stability index, a trend reversal indicator, a fluctuation exceeding limit indicator, and a deviation level label.

[0006] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire time-series data of volume, urine flow rate and urethral pressure recorded by the monitoring device worn by the patient, perform unified alignment processing based on time information, organize the three data sequences in time order, establish a data set under a unified time axis, and generate a synchronously arranged numerical sequence. S102: Based on the time axis of the synchronized numerical sequence, match the image frame sequence at the corresponding time point, filter the image frame data whose time error does not exceed the preset time alignment threshold, and pair the image frame number with the time data to generate an image frame pairing index table. S103: Based on the correspondence between images and numerical sequences in the image frame pairing index table, extract the data and image frame content within the interval from the start to the end of urination, and combine the volume, urine flow rate, urethral pressure and image sequence to form a unified structure to generate a urination function cycle data group.

[0007] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the volume and flow rate change curves recorded in the voiding function cycle data group, extract the volume and flow rate values ​​in the corresponding time series, call the time series positions of the volume and flow rate values, construct the correspondence matrix between volume and flow rate, and perform statistical processing based on the number of corresponding points of volume and flow rate in continuous time periods to obtain the volume-flow rate corresponding ratio sequence. S202: Based on the ratio change trend of continuous time periods in the volume flow rate corresponding ratio sequence, combined with the urethral pressure value sequence extracted from the voiding function cycle data group, the ratio value and urethral pressure value are compared on the same time axis. The corresponding time period position in the ratio sequence is selected according to the urethral pressure threshold, and the selected time interval is marked to obtain the volume flow rate abnormal interval label value. S203: Call the time period position recorded in the abnormal interval label value of the volume flow rate, extract the ratio change content of the corresponding interval in the volume flow rate ratio sequence, and combine it with the abnormal interval label into a unified structure to establish a sequence structure divided by time period and generate a volume flow rate interactive structure.

[0008] As a further aspect of the present invention, in the process of obtaining the capacity-flow-rate ratio sequence, the time sequence positions of the capacity value and the flow rate value are segmented using a fixed time window. The time length of the fixed time window is set to be no less than five adjacent sampling points and no more than twenty adjacent sampling points on the continuous sampling time axis. Within each fixed time window, the ratio of the number of valid corresponding points in the correspondence matrix between capacity and flow rate to the total number of sampling points within that fixed time window is calculated, and the ratio is used as the capacity-flow-rate correspondence ratio value for a single time window. The capacity-flow-rate ratio values ​​formed by adjacent fixed time windows are arranged sequentially to form a continuous and non-overlapping sequence of capacity-flow-rate ratios. The urethral pressure threshold is a fixed value determined based on the statistical distribution of the urethral pressure value sequence. The fixed value is limited to a single threshold weighted by the median and standard deviation of the urethral pressure value sequence, and is used to filter the corresponding time period position in the volume flow rate corresponding ratio sequence.

[0009] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the sequence trend characteristics in the volume-flow interaction structure, extract the volume ratio change value and flow rate response rate value, call the continuous time segment content of the two indicators, classify their trend morphology according to the change of numerical trend, and generate a trend structure classification sequence. S302: Based on the trend category marked in the trend structure classification sequence, call the stage classification rules on capacity regulation capability and pressure response performance in the reference indicator system, perform matching operation according to the correspondence between trend type and the corresponding indicator interval, establish a mapping table between trend category and functional stage, and obtain the urination stage type mapping reference value. S303: Call the trend type number and time series position corresponding to the functional stage in the urination stage type mapping comparison value, divide the continuous time segment of the trend number under the same stage, organize the start time point and end time point of each stage, and supplement the trend attribute to which the stage belongs, and generate a urination function recovery stage interval table.

[0010] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the trend feature sequence recorded in the urination function recovery stage interval table, extract the trend direction value and change amplitude value corresponding to the stage, arrange and organize the trend features according to the stage time order, form a trend feature set under continuous time, perform consistency judgment on the trend features of adjacent stages in the set, and generate a stage trend feature sequence. S402: Based on the trend direction value and change amplitude value in the stage trend feature sequence, obtain the corresponding trend direction sequence and amplitude interval value in the standard trend sample, perform segment-by-segment comparison on the two sets of trend data under the same trend dimension, record the matching degree between each stage trend feature and the standard trend sample, and obtain the trend fit evaluation value. S403: Based on the matching results corresponding to the stages in the trend fit evaluation value, determine the trend segment number, and combine the stage time sequence to organize the segment number continuously, determine the recovery progress segment identifier corresponding to the current matching degree, and generate the individual urination function recovery trend prediction result.

[0011] As a further aspect of the present invention, the determination of the trend direction value is limited based on the sign of the numerical change of the corresponding indicator in adjacent time periods in the trend feature sequence, and the sign of change includes only three discrete values: rising, falling, and stable. The change range value is obtained by normalizing the numerical difference of corresponding indicators in adjacent time periods using a unified scale, and the change range value is limited to a preset range. The consistency judgment uses the condition that the trend direction values ​​of adjacent stages are the same and the difference in the change magnitude value does not exceed a preset magnitude threshold. The matching degree is recorded based on a set of matching identifiers formed by the phase trend feature sequence and the standard trend sample segment by segment under the same number of phases. The ratio of the number of phases that meet the consistency judgment condition to the total number of phases in the matching identifier set is limited to the trend fit evaluation value. The determination of the recovery progress segment identifier is based on the mapping of the numerical range to which the trend fit evaluation value belongs, and the numerical range is limited by the segment number range pre-divided in the standard trend sample.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call the trend direction value and change range value recorded in the individual urination function recovery trend prediction result and the current cycle prediction result, align and organize the two sets of trend data in chronological order, and perform segment-by-segment comparison processing under the same time index to calculate the consistency of trend direction and the degree of difference in change range between adjacent cycles, and obtain the trend continuity comparison value. S502: Based on the degree of difference in the time periods in the trend continuity comparison values, a judgment operation is performed on the trend direction change, and the change range is compared with the preset fluctuation judgment standard. The time period where the direction changes in the opposite direction is marked, and the time period where the change range exceeds the judgment standard is also marked, thus obtaining the trend abnormal section label. S503: Based on the distribution of anomaly types and time periods recorded in the trend anomaly segment annotation, the occurrence status of anomaly markers in the current period is summarized and organized, and the corresponding offset category identifiers are divided according to the combination results of anomaly statuses, and a single status output is formed to generate the predicted trend offset identifier status.

[0013] As a further aspect of the present invention, in the calculation of the trend continuity comparison value, the trend direction consistency is determined by the fact that the individual urination function recovery trend prediction result and the current cycle prediction result are completely consistent with the trend direction value under the same time index. The degree of difference in the range of change is obtained by comparing the range of change value in the individual urination function recovery trend prediction result with the range of change value in the current cycle prediction result under the same time index, and the numerical difference is limited to the form of absolute difference. The preset fluctuation judgment standard is a fixed fluctuation range obtained based on historical cycle trend data statistics. The fixed fluctuation range is used to limit the upper limit of the allowable difference in the range of change. In the trend abnormal segment labeling, the time period where the direction changes in the opposite direction is limited to the time index when the trend direction value changes from rising to falling or from falling to rising in adjacent cycles. The time period where the range of change exceeds the judgment standard is limited to the time index when the difference in the range of change exceeds the fixed fluctuation range. The predicted trend offset identifier status is formed based on a unique mapping determined by the combination of the occurrence frequency of different anomaly types in the current period in the trend anomaly segment annotation.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by synchronously integrating volume, urine flow rate, and urethral pressure data over time and organizing them uniformly with image sequences according to the voiding cycle, multi-source information is continuously correlated within the same time axis. Based on the ratio of volume to flow rate and abnormal pressure markers, the dynamic changes in the voiding process are characterized. The recovery progress is assessed by combining stage division and trend fitting, and the stability and deviation of the continuous cycle trend are identified. This transforms the judgment of voiding function recovery from a static conclusion to a dynamic prediction that can be continuously tracked, thereby improving individual adaptability and clinical early warning value. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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 schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data, comprising the following steps: S1: Acquire volume, urine flow rate and urethral pressure data generated by the monitoring device worn by the patient, synchronize the data in time, and combine it with the imaging image sequence to establish a data structure unit based on the voiding cycle, and generate voiding function cycle data group. S2: Based on the volume and urine flow rate change curves extracted from the voiding function cycle data set, the proportional relationship within a continuous time period is processed, and abnormal intervals are marked by pressure parameters to form a continuous feature sequence reflecting the relationship between volume and flow rate, generating a volume-urine flow interaction structure. S3: Based on the sequence trend characteristics in the volume-flow interaction structure, and combined with the stage classification rules of volume regulation capacity and pressure response performance in the reference index system, the voiding cycle is divided into different stage types, and the time segments and trend attributes of the stages are divided to generate a voiding function recovery stage interval table. S4: Call the trend feature sequence in the interval table of urination function recovery stage, compare the fit with the standard trend sample, and estimate the current recovery progress segment of the patient by evaluating the trend direction and the magnitude of change, and generate individual urination function recovery trend prediction results. S5: Call the individual urination function recovery trend prediction result and the current cycle prediction result, continuously compare the trend direction and range of change, identify the continuity and stability of the trend change, and determine whether there is a trend reversal or the fluctuation range exceeds the judgment standard, and generate the prediction trend deviation status. The current cycle prediction result refers to the current cycle trend prediction result generated based on the urination function cycle data set of the current urination cycle, according to the same trend determination rules as the individual urination function recovery trend prediction result.

[0023] The voiding function cycle data set includes volume synchronization sequences, urine flow rate synchronization sequences, urethral pressure synchronization sequences, imaging frame alignment indexes, voiding cycle identifiers, and cycle start and end timestamps; the volume-urine flow interaction structure includes volume-urine flow ratio feature sequences, pressure abnormality interval markers, interactive continuous feature vectors, feature sequence smoothing parameters, and interactive segment boundary markers; the voiding function recovery stage interval table includes stage type labels, stage time segments, stage trend direction attributes, stage change amplitude attributes, and stage rule matching levels; the individual voiding function recovery trend prediction results include recovery progress segment labels, trend direction judgment values, change amplitude prediction intervals, standard sample fit scores, and prediction confidence levels; the prediction trend deviation status includes trend consistency indicators, trend stability indicators, trend reversal indicators, fluctuation exceeding limits indicators, and deviation level labels.

[0024] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire time-series data of volume, urine flow rate and urethral pressure recorded by the monitoring device worn by the patient, perform unified alignment processing based on time information, organize the three data sequences in time order, establish a data set under a unified time axis, and generate a synchronously arranged numerical sequence. The three time series data exported from the monitoring device—volume, urine flow rate, and urethral pressure—were homogenized. The exported fields were limited to sampling timestamp, measurement value, and sampling sequence number. Timestamps were uniformly converted to milliseconds and retained to the nearest integer. Then, each series was rearranged in ascending order based on its timestamp. If reversed or duplicate timestamps were found, deduplication and correction were performed. The correction rule was to retain only the record with the smallest sampling sequence number if multiple records appeared within the same millisecond; if a sampling sequence number was missing, it was padded according to the original file line number. Next, a unified time axis alignment operation was performed. First, a reference time zero point was determined, and the earliest timestamp among the three series was taken as the alignment starting point and recorded as zero milliseconds. Simultaneously, the timestamps of the other two series were subtracted from this starting time to obtain relative timestamps. Then, the alignment step size was determined. Based on the monitoring device's recording frequency, a 100-millisecond alignment point was used for volume and urethral pressure, and a 100-millisecond alignment point was used for urine flow rate, generating a unified time axis sequence. The time axis covered from zero milliseconds to an additional 3000 milliseconds after the end of urination. Subsequently, resampling and filling were performed on the three sets of sequences. The volume sequence was filled with a forward hold method, and when the alignment point was missing, the volume value earlier than that alignment point was taken. The urine flow rate sequence was filled with a nearest-neighbor matching method, and when the alignment point was missing, the urine flow rate value with the smallest absolute time difference was taken. The urethral pressure sequence was filled with median smoothing, and the smoothing window was taken as the median of the urethral pressure values ​​within a continuous 500 milliseconds. After completion, the three sets of alignment values ​​were merged into the same row of records and the synchronized numerical sequence was output.

[0025] Table 1. Synchronous sampling data during urination cycles Timestamp (milliseconds) Volume (ml) Urine flow rate (ml / s) Urethral pressure (cmH2O) Image frame number 0 120 0.0 28 1 100 122 0.2 29 4 200 125 0.6 31 7 300 128 1.4 34 10 400 132 3.2 38 13 500 136 6.8 45 16 600 140 10.5 52 19 700 145 12.8 58 22 800 149 11.7 61 25 900 153 9.9 57 28 As shown in Table 1, the example forms a uniform row record with a 100-millisecond alignment point and provides the same time index for subsequent image frame pairing and interval extraction.

[0026] S102: Based on the time axis of the synchronously arranged numerical sequence, match the image frame sequence at the corresponding time point, filter the image frame data whose time error does not exceed the preset time alignment threshold, and pair the image frame number with the time data to generate an image frame pairing index table. The image frame sequence undergoes timestamp extraction and frame numbering. The image frames are limited to video streams synchronously acquired during the same inspection process. The frame header time information is read frame by frame and converted to milliseconds. A mapping table is established between frame numbers and frame timestamps. Frames with missing timestamps are removed, and a removal list is recorded. Subsequently, based on the unified timeline of the synchronously arranged numerical sequence, a frame matching operation is performed on each alignment point. First, the frame with the closest timestamp to the alignment point is retrieved from the mapping table. The absolute time difference between the frame's timestamp and the alignment point is calculated and compared with a preset time alignment threshold. If the time difference is not greater than the threshold, it is recorded as a valid match; if the time difference is greater than the threshold, the alignment point is marked as having no valid frame and written into the missing frame marker column. The time alignment threshold was determined in this embodiment through an alignment verification experiment. The experiment selected 10 synchronously acquired recordings, each lasting 30 seconds. Identifiable event points were manually marked on the video, such as the moment of catheter connection or the moment an external indicator light flashed. Simultaneously, the pressure change locations of the same event point were marked on the numerical sequence. The time difference distribution between the two types of marked points was statistically analyzed, and the time difference covering more than 90% of the samples was taken as the threshold. The experimental results showed that the time difference was concentrated in the range of 0 to 120 milliseconds. 150 milliseconds was chosen as the time alignment threshold to cover extreme jitter samples. After setting the threshold, full pairing was performed to generate an image frame pairing index table. The index table fields were limited to alignment point timestamp, frame number, frame timestamp, time difference, and valid marker.

[0027] S103: Based on the correspondence between images and numerical sequences in the image frame pairing index table, extract the data and image frame content within the interval from the start to the end of urination, and combine the volume, urine flow rate, urethral pressure and image sequence to form a unified structure to generate a urination function cycle data set; The start and end intervals of urination are determined based on the image frame pairing index table. The start point is determined by the first alignment point where the urine flow rate increases from a continuous 0.0 ml / s to a continuous non-0.0 ml / s, requiring at least three alignment points with a urine flow rate greater than 0.5 ml / s within the subsequent 500 milliseconds. The end point is determined by the urine flow rate returning to a continuous 0.0 ml / s and remaining there for 2000 milliseconds, while the volume change tends to stabilize. Stability is determined by the volume change not exceeding 2 ml within 1000 milliseconds. After determining the start and end points, a set of aligned records of volume, urine flow rate, and urethral pressure from the start to the end point is extracted on a unified time axis, and the corresponding set of valid image frames is extracted simultaneously. When a missing frame is encountered, the frame content index is filled according to the time order of adjacent valid frames. The filling rule is that the missing frame alignment point is taken from the valid frame number earlier than that point; if the earlier direction is also missing, the valid frame number later than that point is taken, and the frame source column is marked as an alternative source. The extracted numerical records and frame indexes are then written into a unified structure. Fields are limited to alignment point timestamp, volume value, urine flow rate, urethral pressure, frame number, and frame validity marker. One-time metadata is also added, limited to subject number, collection date, sensor serial number, video file identifier, and sampling step size. For example, referring to Table 1, if the urine flow rate is 0.2 ml / s at 100 ms but does not meet the continuous intensity condition, and is 0.6 ml / s at 200 ms, and then greater than 0.5 ml / s for the next 300 to 700 ms, the starting point is determined to be 200 ms. At 900 ms, it remains at 9.9 ml / s. If, for the next 2000 ms starting at 1500 ms, it is 0.0 ml / s with a volume change within 2 ml, then the ending point is determined to be 1500 ms, corresponding to a truncation interval of 200 ms to 1500 ms. The extracted structure is output as a voiding function cycle data set. Each record in the data set has capacity, urine flow rate, urethral pressure and image frame location information under the same time index.

[0028] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the volume and flow rate change curves recorded in the voiding function cycle data group, extract the volume and flow rate values ​​in the corresponding time series, call the time series positions of the volume and flow rate values, construct the correspondence matrix between volume and flow rate, and perform statistical processing based on the number of corresponding points of volume and flow rate in continuous time periods to obtain the volume-flow rate corresponding ratio sequence. First, volume and urine flow rate sequences are extracted from the voiding function cycle data set in chronological order. During extraction, the timestamp and corresponding value of each record are retained, and a length check is performed on the two sequences. If a length discrepancy is found, gap filling is performed based on the time axis, following the previously mentioned alignment and filling rules. Then, a correspondence matrix between volume and urine flow rate is constructed. This matrix uses continuous time periods as row indices and volume bins and flow rate bins as column indices. In this embodiment, volume is divided into 10 ml intervals, and urine flow rate into 1 ml / second intervals. Each alignment point is traversed, ensuring the volume value falls into the corresponding volume interval and the urine flow rate value falls into the corresponding flow rate interval, and the count of the corresponding row and column positions in the matrix is ​​incremented by 1. Continuous time periods are divided using 1000 milliseconds as a statistical window, with no overlap between windows. Segments with a total duration of less than 1000 milliseconds are still counted as separate windows. After the counting is completed, the capacity-flow-rate ratio sequence is calculated for each statistical window. The ratio is calculated by the ratio between the number of matrix grid points recorded in the statistical window and the total number of alignment points in the window. The number of grid points is based on the combined count of the deduplicated capacity bins and flow-rate bins.

[0029] S202: Based on the ratio change trend of continuous time periods in the volume flow rate ratio sequence, combined with the urethral pressure value sequence extracted from the voiding function cycle data group, the ratio value and urethral pressure value are compared on the same time axis. The corresponding time period position in the ratio sequence is selected according to the urethral pressure threshold, and the selected time interval is marked to obtain the volume flow rate abnormal interval label value. Continuous time-period trend extraction was performed on the ratio sequence corresponding to volume flow rate. The ratio values ​​of adjacent statistical windows were arranged into a trend vector in chronological order, and the direction of change was determined for each adjacent window segment. The determination rule was that if the ratio value of the subsequent window was greater than that of the preceding window, it was considered an increase; if the ratio value of the subsequent window was less than that of the preceding window, it was considered a decrease; and if the absolute value of the difference did not exceed 0.05, it was considered stable. Subsequently, the urethral pressure value sequence was extracted from the voiding function cycle data group, and in-window aggregation was performed according to the statistical window boundaries consistent with the ratio sequence. The aggregation rule was to take the maximum urethral pressure and the mean urethral pressure within each statistical window. The maximum value was used for threshold screening, and the mean value was used to help confirm the persistence of pressure. In this embodiment, a two-level threshold was used for urethral pressure. The first-level threshold was 30 cmH2O, and the second-level threshold was 60 cmH2O. The thresholds were set based on publicly available descriptions of common urethral closure pressure boundaries. A threshold below 20 cmH2O is often used to indicate sphincter insufficiency. 30 cmH2O, as the lower limit of normal, is more conducive to reducing false alarms. 60 cmH2O serves as a high-pressure screening line to locate the stress stress zone. The threshold values ​​were calibrated through a validation experiment. The experiment selected 20 subjects for one cycle each. The visible sphincter contraction performance zones in the video frames were manually marked and compared with the maximum value within the urethral pressure window. The matching rate under different thresholds was statistically analyzed. The results showed that the matching rate was 0.82 when the threshold was 30 cmH2O, and 0.67 when the threshold was 60 cmH2O, but the localized zones were more concentrated. Therefore, a dual-threshold parallel screening was adopted. The screening operation compares the maximum urethral pressure with a threshold window by window. When the maximum value is greater than or equal to 60 cmH2O, the corresponding ratio time period is marked as a strong abnormality candidate. When the maximum value is between 30 cmH2O and 60 cmH2O and the ratio trend is decreasing, it is marked as a weak abnormality candidate. All others are marked as normal. For example, a window with a ratio of 0.3, decreasing from the previous window's 0.6, has a maximum urethral pressure of 61 cmH2O, and is marked as a strong abnormality candidate. Another window with a ratio of 0.55, increasing from the previous window's 0.52, has a maximum urethral pressure of 45 cmH2O, not meeting the decreasing condition, and is marked as normal. All candidate windows are merged in chronological order, with adjacent windows merging into the same abnormal interval if the interval between windows is no more than one window. The volume-flow-rate abnormal interval label is then output.

[0030] Table 2 Threshold and Judgment Criteria Setting Table Standard Name Numerical range or values Calibration test sample size example Calibration result values Time alignment threshold 150 milliseconds 10 synchronized recordings Coverage ratio 0.92 Urethral pressure level 1 threshold 30 cm water column 20 subjects Match rate 0.82 Secondary threshold of urethral pressure 60 cm water column 20 subjects Match rate 0.67 Ratio steady zone 0.05 20 subjects The average noise level between windows is 0.03. Fluctuation Judgment Standard Difference in range of variation: 2.0 20 subjects False alarm rate: 0.08 As shown in Table 2, the key thresholds are given with clear values, and quantitative results such as coverage ratio, matching rate or false alarm rate are obtained through calibration tests of the examples.

[0031] S203: Call the time period position recorded in the volume flow rate abnormal interval label value, extract the ratio change content of the corresponding interval in the volume flow rate corresponding ratio sequence, and combine it with the abnormal interval label into a unified structure to establish a sequence structure divided by time period and generate a volume flow rate interactive structure. First, the start and end window positions of each abnormal interval recorded in the volume-flow-rate abnormal interval labeling values ​​are retrieved. Based on this, a ratio segment of the corresponding interval is extracted from the volume-flow-rate corresponding ratio sequence. During extraction, the start and end timestamps, ratio values, trend directions, and corresponding maximum urethral pressure values ​​of each statistical window within the segment are retained. Then, the abnormal interval labels and ratio segments are structurally combined. The combination rule forms a sequence structure record for each abnormal interval. The record fields are limited to interval number, interval start timestamp, interval end timestamp, interval type label, interval ratio sequence array, interval trend direction array, and interval maximum urethral pressure array. The interval type label is derived from the aforementioned candidate level and merging results. If any statistical window within the interval is a strong abnormal candidate, the interval type is labeled as a strong abnormal interval; otherwise, it is labeled as a weak abnormal interval. To ensure that the sequence structure divided by time period is verifiable, index verification information is added to each interval. The verification information is limited to the consistency between the number of statistical windows within the interval and the interval duration. If the difference between the number of statistical windows multiplied by 1000 milliseconds and the interval duration exceeds 500 milliseconds, it is marked as a boundary that needs to be verified, and a boundary verification prompt column is recorded.

[0032] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the sequence trend characteristics in the volume-flow interaction structure, extract the volume ratio change value and flow rate response rate value, call the continuous time segment content of the two indicators, classify their trend patterns according to the changes in the numerical trend, and generate a trend structure classification sequence. First, the volume-flow interaction structure is traversed interval by interval, and two types of indicators are extracted from the ratio sequence array of each interval: volume ratio change value and flow rate response rate value. The volume ratio change value is obtained by using the difference between the first window ratio and the last window ratio within the interval as the interval change value, and the difference between the maximum ratio and the minimum ratio within the interval is recorded as the interval fluctuation amplitude. The flow rate response rate value is obtained by using the time difference between the peak urine flow rate within the interval and the start time of the interval to infer the response speed. Specifically, the urine flow rate alignment sequence is extracted from the voiding function cycle data group according to the timestamp of the abnormal interval, the timestamp of the alignment point where the maximum urine flow rate is located within the time period is found, and the time difference is calculated with the start timestamp of the interval. A time difference of less than 500 milliseconds is recorded as a fast response, a time difference between 500 and 1500 milliseconds is recorded as a medium-speed response, and a time difference greater than 1500 milliseconds is recorded as a slow response. Subsequently, the continuous time intervals of the two indicators are retrieved, and their trend patterns are classified. The classification is based on a combination rule of numerical trend and response speed, which is limited to four categories: the first category is an increase in ratio with a fast response; the second category is an increase in ratio with a medium or slow response; the third category is a decrease in ratio with a fast response; and the fourth category is a decrease in ratio with a medium or slow response. When the ratio change value falls into the stable zone, it is further judged according to the fluctuation amplitude. If the fluctuation amplitude does not exceed 0.10, it is classified into the stable category and the previous interval category is used. If the fluctuation amplitude is greater than 0.10, it is classified according to the most recent non-stationary judgment in the direction of change.

[0033] S302: Based on the trend category marked in the trend structure classification sequence, call the stage classification rules on capacity regulation capability and pressure response performance in the reference indicator system, perform matching operation according to the correspondence between trend type and the corresponding indicator interval, establish a mapping table between trend category and functional stage, and obtain the urination stage type mapping reference value. First, the stage classification rules in the reference indicator system are prepared. The rules are given in the form of segmented intervals for volume regulation capacity and pressure response performance. Volume regulation capacity is determined by a combination of volume ratio change value and fluctuation amplitude, while pressure response performance is determined by a combination of maximum urethral pressure and urine flow rate response rate. Then, a matching operation is performed on each trend structure classification sequence. First, the category number, volume ratio change value, fluctuation amplitude, urine flow rate response rate category, and the corresponding interval maximum urethral pressure value of the record are read. The interval is then determined according to the rule table. The determination operation adopts the principle of item-by-item comparison and unique hit. First, the maximum urethral pressure is compared with the 30 cmH2O and 60 cmH2O thresholds in Table 2 to determine whether the pressure response performance is in the low-pressure, medium-pressure, or high-pressure segment. Then, the volume ratio change value is compared with the steady band of 0.05 to determine whether the volume regulation capacity is in the rising segment, falling segment, or steady segment. Finally, the urine flow rate response rate category is used to further subdivide the stage and form a unique stage identifier. In this embodiment, three stage types are used for stage identification. The first stage corresponds to an increase in capacity regulation and a low or medium pressure range. The second stage corresponds to a stable capacity regulation and a medium pressure range. The third stage corresponds to a decrease in capacity regulation and a high pressure range. If a capacity regulation increases but the pressure is in the high pressure range, it is processed as the third stage and marked as an increasing high pressure conflict in the conflict marker column.

[0034] S303: Call the trend type number and time series position corresponding to the functional stage in the voiding stage type mapping comparison value, divide the continuous time segment of the trend number under the same stage, organize the start time point and end time point of each stage, and supplement the trend attribute to which the stage belongs, and generate a voiding function recovery stage interval table. First, the urination stage type mapping reference value is called to attach a stage type and corresponding trend type number to each record in the trend structure classification sequence. Then, the records are divided into continuous segments according to their time series positions. The segmentation operation adopts the same stage merging rule. If adjacent records have the same stage type and the interval does not exceed 1000 milliseconds, they are merged into the same continuous segment, with the earliest start time stamp as the stage start point and the latest end time stamp as the stage end point. If the stage types are different or the interval exceeds 1000 milliseconds, a new stage segment is opened. For each stage segment, the start and end times are sorted, and the trend attribute to which the stage belongs is added. The trend attribute is the trend type number that appears most frequently in the stage segment as the primary trend number, while the trend number that appears the second most frequently is retained as the secondary trend number. If the difference between the primary and secondary trends does not exceed 1, the stage is marked as a mixed trend stage. Subsequently, a boundary consistency check is performed on the stage segment. The check is based on whether the original alignment point range covered by the stage segment is consistent with the start and end points of the voiding function cycle data group. If the start point of the stage segment is earlier than the voiding start point, the stage start point is clipped to the voiding start point. If the end point of the stage segment is later than the voiding end point, the stage end point is clipped to the voiding end point. The clipping direction and clipping milliseconds are recorded in the clipping mark column.

[0035] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the urination function to restore the trend feature sequence recorded in the stage interval table, extract the trend direction value and change amplitude value corresponding to the stage, arrange and organize the trend features according to the stage time order, form a trend feature set under continuous time, perform consistency judgment on the trend features of adjacent stages in the set, and generate a stage trend feature sequence. The table of urination function recovery stages is read, and the corresponding trend feature sequence is extracted for each stage number in chronological order. The trend direction value is extracted by mapping the main trend number to a direction label. The mapping rule is: for a main trend number corresponding to an increasing ratio, the direction label is upward; for a main trend number corresponding to a decreasing ratio, the direction label is downward; and for a main trend number corresponding to a stable ratio, the direction label is horizontal. The magnitude value is extracted based on the ratio segment covered by the stage. The difference between the maximum and minimum ratios within the stage is taken as the magnitude, and the magnitude is divided into three levels: magnitudes not exceeding 0.10 are considered small, magnitudes between 0.10 and 0.25 are considered medium, and magnitudes greater than 0.25 are considered large. The stages are then arranged in chronological order to form a continuous trend feature set. Consistency judgment is performed on the trend features of adjacent stages in the set. Consistency judgment is divided into directional consistency and magnitude consistency. Directional consistency is determined by the adjacent stages having the same direction label; otherwise, they are inconsistent. Amplitude consistency is determined by the adjacent stages having the same magnitude level or differing by one level; otherwise, they are inconsistent. For inconsistencies, a marking process is performed. The marking fields are limited to direction jump markers and amplitude jump markers. The stage boundary timestamps where the jumps occur are recorded, and the stage trend feature sequence is output.

[0036] S402: Based on the trend direction value and change amplitude value in the stage trend feature sequence, obtain the corresponding trend direction sequence and amplitude interval value in the standard trend sample, perform segment-by-segment comparison on the two sets of trend data under the same trend dimension, record the matching degree between each stage trend feature and the standard trend sample, and obtain the trend fit evaluation value. First, prepare a standard trend sample. The standard trend sample comes from the set of subjects with stable recovery trajectories selected in the verification test of the example. The selection criteria are that the direction jump markers and amplitude jump markers of the same subject are all negative for three consecutive cycles, and the maximum urine flow rate falls within the common range of 10 ml / s to 21 ml / s. For each selected subject, the direction sequence and amplitude level sequence are extracted according to the stage number, and a majority vote is performed on the stage with the same number to obtain the standard direction sequence and standard amplitude interval value. The amplitude interval value is expressed as the percentile interval of the amplitude of the stage, and 25% to 75% is taken as the standard interval. Subsequently, under the same trend dimension, the phase trend feature sequence and the standard trend sample are compared segment by segment. The comparison is performed independently for each phase. First, the direction labels are compared. If the directions are consistent, it is recorded as a direction hit; if the directions are inconsistent, it is recorded as a direction miss. Then, the amplitude is compared. If the amplitude of the phase falls within the corresponding standard amplitude range, it is recorded as an amplitude hit; otherwise, it is recorded as an amplitude miss. Finally, the two types of hit results are combined into a matching degree. The matching degree uses a discrete score: 1.0 for both direction hit and amplitude hit, 0.7 for both direction hit and amplitude miss, 0.4 for both direction miss and amplitude hit, and 0.0 for both direction miss and amplitude miss. The trend fit evaluation value is output. The evaluation value is indexed by the phase number, and the matching degree value and whether it is qualified are recorded. The deviation direction information when the direction miss is retained.

[0037] S403: Based on the matching results corresponding to the stage in the trend fit evaluation value, determine the trend segment number, and combine the stage time sequence to organize the segment number continuously, determine the recovery progress segment identifier corresponding to the current matching degree, and generate individual urination function recovery trend prediction results. Read the trend fit evaluation value, and organize the qualified markers of each stage in chronological order. Consecutive qualified stages are considered candidates for the same recovery progress segment, and the location of an unqualified stage is considered the segment boundary. Segment numbers are generated starting from 1 and incrementing. The first qualified stage is assigned segment 1. If qualified stages are consecutive, the segment number remains unchanged. If two consecutive unqualified stages occur, the segment number is incremented by 1, and the new number is used in subsequent qualified stages. If one consecutive unqualified stage occurs, boundary buffering is performed. The buffering rule is to observe the matching degree of the unqualified stage. If the matching degree is 0.4, it is merged into the previous qualified segment and marked as weakly unqualified and merged. If the matching degree is 0.0, it is treated as an independent breakpoint and the segment number increment is triggered. Subsequently, the recovery progress segment identifier corresponding to the current matching degree is determined. The identifier fields are limited to the current segment number, segment start and end timestamps, number of stages within the segment, and segment stability flag. The stability flag is determined based on the number of times the directional jump flag appears within the segment. A frequency of 0 occurrences is considered stable, a frequency of 1 occurrence is considered mild fluctuation, and a frequency of 2 or more occurrences is considered significant fluctuation. The segment number and segment attributes are written into the individual voiding function recovery trend prediction result, and the segment boundary timestamps are written back to the boundary column of the voiding function recovery stage interval table to form a contextually traceable association. The advantage of this operation logic is that by having two types of features, directional hit and amplitude hit, participate in the stage scoring and introducing a buffering and incorporation rule, short-term amplitude deviation will not directly cut off the segment continuity, while directional reversal will trigger segment segmentation.

[0038] Please see Figure 6 The specific steps of S5 are as follows: S501: Call the individual urination function to restore the trend prediction results and the trend direction and change range values ​​recorded in the current cycle prediction results, align and organize the two sets of trend data in chronological order, and perform segment-by-segment comparison processing under the same time index to calculate the consistency of trend direction and the degree of difference in change range between adjacent cycles, and obtain the trend continuity comparison value. First, the individual's urination function recovery trend prediction result and the current cycle prediction result are read. Both include the segment number, direction label, amplitude level, and segment start and end timestamps in chronological order. The alignment and sorting operation is constrained by both the stage number and the timestamp. First, the stage boundary timestamps of the two sets of results are projected onto the same global time index. The global time index is based on the current cycle and generates index points in 1000-millisecond statistical windows starting from the start timestamp of the current cycle. Then, the individual prediction results are mapped according to the corresponding statistical window positions. The mapping rule is that if the individual prediction segment covers the statistical window, the window is assigned the direction label and amplitude level of that segment. After mapping is completed, segment-by-segment comparison is performed under the same time index. The comparison is divided into two parts: trend direction consistency and degree of difference in range of change. Trend direction consistency is compared by window level. If the two groups have the same direction label in the window, they are recorded as consistent; otherwise, they are recorded as inconsistent. The ratio of the number of consistent windows in the whole period to the total number of windows is calculated to obtain the direction consistency ratio. The degree of difference in range of change is converted into numerical difference according to the difference in amplitude level. The difference is mapped as follows: the difference is 0 if the level is the same, the difference is 1 if the level differs by 1, and the difference is 2 if the level differs by 2. Then, the average of the difference values ​​in the whole period is taken as the degree of difference in range of change. For example, there are 12 statistical windows in a certain period. If 9 of the windows have the same direction, the direction consistency ratio is 0.75. The amplitude level difference value sequence is 0, 1, 1, 0, 2, etc., with an average value of 0.83. The direction consistency ratio and the degree of difference in range of change are combined into a trend continuity comparison value, and a list of timestamps of inconsistent windows and the timestamp position of the window with the largest difference are attached. The direction consistency ratio is used to determine whether there is a reverse change in direction, and the degree of difference in range of change is used to determine whether it exceeds the preset fluctuation judgment standard.

[0039] S502: Based on the degree of difference in the time period in the trend continuity comparison value, perform a judgment operation on the change in trend direction, compare the change range with the preset fluctuation judgment standard, mark the time period where the direction changes in the opposite direction, and mark the time period where the change range exceeds the judgment standard to obtain the trend abnormal section label. First, the inconsistency window timestamp list in the trend continuity comparison values ​​is read. For each inconsistency window, a direction change judgment is performed. The judgment rule is that if the current period direction label is opposite to the individual's predicted direction label, it is marked as a direction reversal; if they are not opposite but not the same, it is marked as a direction deviation. The determination of opposite directions uses upward and downward as opposites, and parallel and any direction are not considered opposite. Then, the range of change is compared. The preset fluctuation judgment standard is 2.0 in this embodiment. See Table 2, which is the judgment line for the average difference of amplitude levels. When the average value is greater than or equal to 2.0, it is marked as the range exceeding the standard; when the average value is between 1.0 and 2.0, it is marked as the range being too large; when the average value is less than 1.0, it is marked as the range being normal. For example, if the difference in the range of change in the aforementioned S501 example is 0.83, then the range is normal; if the average difference of another subject in the comparison is 2.1, then the range exceeds the standard and the range abnormality mark is triggered. Further, at the window level, abnormal locations are located. The location rules are as follows: when there is a range exceeding the limit or a range that is too large, windows with an amplitude difference value equal to 2 are retrieved, and their timestamp positions are marked as range abnormal windows; when there is a reverse change in direction, the corresponding window is marked as a direction abnormal window. The two types of markings are merged. If the same window satisfies both direction and range abnormality, it is marked as a composite abnormal window. Adjacent abnormal windows are merged into an abnormal time period in chronological order, with a merging interval threshold of 1 statistical window. The output is a trend abnormal segment label, with fields limited to abnormal segment number, start timestamp, end timestamp, abnormal type, and composite label, and retaining the number of windows and the maximum amplitude difference value contained in each abnormal segment.

[0040] S503: Based on the distribution of anomaly types and time periods recorded in the trend anomaly segment annotation, the occurrence status of anomaly markers in the current period is summarized and organized. The corresponding offset category identifiers are divided according to the combination results of anomaly statuses, and a single status output is generated to produce the predicted trend offset identifier status. The system reads the labels of trend anomaly segments and summarizes the anomaly occurrence status according to anomaly type and time period distribution. The summary dimensions are limited to the number of occurrences of directional anomalies, the number of occurrences of range anomalies, the number of occurrences of compound anomalies, the total duration of anomalies, and the earliest occurrence timestamp of anomalies. Based on the combination results of anomaly status, the system classifies the offset category. In this embodiment, the category is set to 3 types: the first type is stable direction and normal range, and the judgment condition is that the number of occurrences of directional anomalies is 0 and the number of occurrences of range anomalies is 0; the second type is range offset, and the judgment condition is that the number of occurrences of directional anomalies is 0 and the number of occurrences of range anomalies is greater than or equal to 1; the third type is directional offset, and the judgment condition is that the number of occurrences of directional anomalies is greater than or equal to 1, without distinguishing whether range anomalies exist. For example, if the number of directional abnormalities in a certain cycle is 0 but the number of range abnormalities is 2, it is classified as the second category; if the number of directional abnormalities is 1 and the number of compound abnormalities is 1, it is classified as the third category, forming a single state output. The output fields are limited to the offset category identifier, the evidence time period list, and the key evidence window timestamp. The key evidence window is the earliest directional abnormality window or the window with the largest amplitude difference. The trial uses the cycle data of the same batch of 20 subjects. The offset category output by the process is compared with the consistency of the conclusion of manual review, and compared with the control process that only uses the urine flow rate threshold rule.

[0041] Table 3. Comparison of Predicted Misalignment Judgment Test Results Experimental group Sample size Number of samples consistent with human conclusions Consistency rate Average processing time in seconds This embodiment's process 20 17 0.85 18.4 Comparison process 20 13 0.65 11.2 As shown in Table 3, the consistency rate of the process in this embodiment is 0.85, while the consistency rate of the control process is 0.65, representing an improvement of 20 percentage points compared to the control process, while maintaining the processing time at the level of 18.4 seconds. These experimental results demonstrate that the generation of the offset category identifier and the labeling of trend anomaly segments, through the participation of multiple parameters including time alignment, pressure screening, and trend fit evaluation, result in a more consistent single-state output. Furthermore, this offset category identifier can be directly derived from the summary results of anomaly types and time period distributions and traced back to specific timestamp evidence.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data, characterized in that, Includes the following steps: S1: Acquire volume, urine flow rate and urethral pressure data generated by the monitoring device, integrate them in time synchronization, and align them with the imaging image sequence in time. Establish a data structure based on the voiding cycle to form a voiding function cycle data group. S2: Based on the volume and urine flow rate change curves of the voiding function cycle data set, calculate the proportional relationship of continuous time segments, and mark abnormal segments in combination with pressure parameters to construct the volume-urine flow relationship and generate a volume-urine flow interaction structure. S3: Based on the sequence trend characteristics of the volume-flow interaction structure, and combined with the stage division rules of volume regulation capacity and pressure response performance in the reference index system, the voiding cycle is divided into stages, the time intervals and trend attributes corresponding to the stages are defined, and a voiding function recovery stage interval table is generated. S4: Call the trend feature sequence in the interval table of the urination function recovery stage, evaluate the fit with the standard trend sample, divide the patient's current recovery progress segment, and generate individual urination function recovery trend prediction results; S5: Continuously compare the predicted trend of individual urination function recovery with the current cycle prediction result to determine the stable state of trend change and whether there is a trend reversal or exceeding the judgment range, and generate a predicted trend deviation status.

2. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 1, characterized in that, The voiding function cycle data set includes a volume synchronization sequence, a urine flow rate synchronization sequence, a urethral pressure synchronization sequence, an imaging frame alignment index, a voiding cycle identifier, and cycle start and end timestamps; the volume-urine flow interaction structure includes a volume-urine flow ratio feature sequence, pressure abnormality interval markers, interactive continuous feature vectors, feature sequence smoothing parameters, and interactive segment boundary markers; the voiding function recovery stage interval table includes stage type labels, stage time segments, stage trend direction attributes, stage change amplitude attributes, and stage rule matching levels; the individual voiding function recovery trend prediction results include recovery progress segment labels, trend direction judgment values, change amplitude prediction intervals, standard sample fit scores, and prediction confidence levels; the prediction trend deviation status includes trend coherence indicators, trend stability indicators, trend reversal indicators, fluctuation exceeding limits indicators, and deviation level labels.

3. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire time-series data of volume, urine flow rate and urethral pressure recorded by the monitoring device worn by the patient, perform unified alignment processing based on time information, organize the three data sequences in time order, establish a data set under a unified time axis, and generate a synchronously arranged numerical sequence. S102: Based on the time axis of the synchronized numerical sequence, match the image frame sequence at the corresponding time point, filter the image frame data whose time error does not exceed the preset time alignment threshold, and pair the image frame number with the time data to generate an image frame pairing index table. S103: Based on the correspondence between images and numerical sequences in the image frame pairing index table, extract the data and image frame content within the interval from the start to the end of urination, and combine the volume, urine flow rate, urethral pressure and image sequence to form a unified structure to generate a urination function cycle data group.

4. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Based on the volume and flow rate change curves recorded in the voiding function cycle data group, extract the volume and flow rate values ​​in the corresponding time series, call the time series positions of the volume and flow rate values, construct the correspondence matrix between volume and flow rate, and perform statistical processing based on the number of corresponding points of volume and flow rate in continuous time periods to obtain the volume-flow rate corresponding ratio sequence. S202: Based on the ratio change trend of the continuous time period in the volume flow rate corresponding ratio sequence, combined with the urethral pressure value sequence extracted from the voiding function cycle data group, the ratio value sequence and the urethral pressure value sequence are time-aligned on the same time axis, and the ratio value segment at the corresponding time position is filtered and marked according to whether the urethral pressure value meets the preset threshold to obtain the volume flow rate abnormal interval label value. S203: Call the time period position recorded in the abnormal interval label value of the volume flow rate, extract the ratio change content of the corresponding interval in the volume flow rate ratio sequence, and combine it with the abnormal interval label into a unified structure to establish a sequence structure divided by time period and generate a volume flow rate interactive structure.

5. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 4, characterized in that, In the process of obtaining the capacity-flow-rate ratio sequence, the time sequence positions of the capacity value and the flow rate value are segmented using a fixed time window. The time length of the fixed time window is set to be no less than five adjacent sampling points and no more than twenty adjacent sampling points on the continuous sampling time axis. Within each fixed time window, the ratio of the number of valid corresponding points in the correspondence matrix between capacity and flow rate to the total number of sampling points within that fixed time window is calculated, and the ratio is used as the capacity-flow-rate correspondence ratio value for a single time window. The capacity-flow-rate ratio values ​​formed by adjacent fixed time windows are arranged sequentially to form a continuous and non-overlapping sequence of capacity-flow-rate ratios. The urethral pressure threshold is a fixed value determined based on the statistical distribution of the urethral pressure value sequence. The fixed value is limited to the median and standard deviation of the urethral pressure value sequence being linearly weighted according to a preset weight determined based on historical samples or empirical statistics to form a single threshold, which is used to filter the corresponding time period position in the volume flow rate corresponding ratio sequence.

6. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the sequence trend characteristics in the volume-flow interaction structure, extract the volume ratio change value and flow rate response rate value, call the continuous time segment content of the two indicators, classify their trend morphology according to the change of numerical trend, and generate a trend structure classification sequence. S302: Based on the trend category marked in the trend structure classification sequence, call the stage classification rules on capacity regulation capability and pressure response performance in the reference indicator system, perform matching operation according to the correspondence between trend type and the corresponding indicator interval, establish a mapping table between trend category and functional stage, and obtain the urination stage type mapping reference value. S303: Call the trend type number and time series position corresponding to the functional stage in the urination stage type mapping comparison value, divide the continuous time segment of the trend number under the same stage, organize the start time point and end time point of each stage, and supplement the trend attribute to which the stage belongs, and generate a urination function recovery stage interval table.

7. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 6, characterized in that, The specific steps of S4 are as follows: S401: Call the trend feature sequence recorded in the urination function recovery stage interval table, extract the trend direction value and change amplitude value corresponding to the stage, arrange and organize the trend features according to the stage time order, form a trend feature set under continuous time, perform consistency judgment on the trend features of adjacent stages in the set, and generate a stage trend feature sequence. S402: Based on the trend direction value and change amplitude value in the stage trend feature sequence, obtain the corresponding trend direction sequence and amplitude interval value in the standard trend sample, perform segment-by-segment comparison on the two sets of trend data under the same trend dimension, record the matching degree between each stage trend feature and the standard trend sample, and obtain the trend fit evaluation value. S403: Based on the matching results corresponding to the stages in the trend fit evaluation value, determine the corresponding recovery progress segment number according to the falling situation of the matching results in the preset matching degree interval, and sort the segment numbers continuously in combination with the stage time sequence, determine the recovery progress segment identifier corresponding to the current matching degree, and generate the individual urination function recovery trend prediction result.

8. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 7, characterized in that, The determination of the trend direction value is based on the sign of the numerical change of the corresponding indicator in adjacent time periods in the trend feature sequence, and the sign of change only includes three discrete values: rising, falling, and stable. The change range value is obtained by normalizing the numerical difference of corresponding indicators in adjacent time periods using a unified scale, and the change range value is limited to a preset range. The consistency judgment uses the condition that the trend direction values ​​of adjacent stages are the same and the difference in the change magnitude value does not exceed a preset magnitude threshold. The matching degree is recorded based on a set of matching identifiers formed by the phase trend feature sequence and the standard trend sample segment by segment under the same number of phases. The ratio of the number of phases that meet the consistency judgment condition to the total number of phases in the matching identifier set is limited to the trend fit evaluation value. The determination of the recovery progress segment identifier is based on the mapping of the numerical range to which the trend fit evaluation value belongs, and the numerical range is limited by the segment number range pre-divided in the standard trend sample.

9. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 7, characterized in that, The specific steps of S5 are as follows: S501: Call the trend direction value and change range value recorded in the individual urination function recovery trend prediction result and the current cycle prediction result, align and organize the two sets of trend data in chronological order, and perform segment-by-segment comparison processing under the same time index to calculate the consistency of trend direction and the degree of difference in change range between adjacent cycles, and obtain the trend continuity comparison value. S502: Based on the degree of difference in the time periods in the trend continuity comparison values, a judgment operation is performed on the trend direction change, and the change range is compared with the preset fluctuation judgment standard. The time period where the direction changes in the opposite direction is marked, and the time period where the change range exceeds the judgment standard is also marked, thus obtaining the trend abnormal section label. S503: Based on the distribution of anomaly types and time periods recorded in the trend anomaly segment annotation, the occurrence status of anomaly markers in the current period is summarized and organized, and the corresponding offset category identifiers are divided according to the combination results of anomaly statuses, and a single status output is formed to generate the predicted trend offset identifier status.

10. The method for predicting the recovery of voiding function after orthotopic neobladder surgery based on multimodal data according to claim 9, characterized in that, In the calculation of the trend continuity comparison value, the trend direction consistency is determined by the fact that the trend direction value of the individual urination function recovery trend prediction result and the current cycle prediction result are completely consistent under the same time index. The degree of difference in the range of change is obtained by comparing the range of change value in the individual urination function recovery trend prediction result with the range of change value in the current cycle prediction result under the same time index, and the numerical difference is limited to the form of absolute difference. The preset fluctuation judgment standard is a fixed fluctuation range obtained based on historical cycle trend data statistics. The fixed fluctuation range is used to limit the upper limit of the allowable difference in the range of change. In the labeling of the abnormal trend segment, the time period in which the direction changes in the opposite direction is defined as the time index when the trend direction value changes from rising to falling or from falling to rising in adjacent periods, and the time period in which the change range exceeds the judgment standard is defined as the time index when the degree of difference in the change range exceeds the fixed fluctuation range. The predicted trend offset identifier status is formed based on a unique mapping determined by the combination of the occurrence frequency of different anomaly types in the current period in the trend anomaly segment annotation.

Images