CN-122025130-A - Dynamic evaluation method for risk of prolapse recurrence of pelvic floor organ

CN122025130ACN 122025130 ACN122025130 ACN 122025130ACN-122025130-A

Abstract

The invention relates to the technical field of medical artificial intelligence and provides a dynamic evaluation method of the risk of prolapse recurrence of pelvic floor organs, which comprises the following steps of collecting diagnosis and treatment data, structurally modeling and estimating baseline risk; step two, extracting time sequence feature engineering and recurrence early warning indexes, step three, dynamic risk modeling and probability prediction, and step four, layering early warning response and personalized intervention. The invention solves the problem that the static model cannot perceive long-term dependence and individual difference by combining the improved CRF-LSTM layering model with the operation history feature and the dynamic time sequence index, and realizes dynamic accurate evaluation. At the modeling layer, the association relation between the CRF layer coding operation history and the time sequence characteristic, the LSTM layer learns the long-term time dependence characteristic, simultaneously introduces a time-to-event loss function to enhance the sensitivity to the survival time, and finally outputs a dynamic risk sequence evolving along with the follow-up time, thereby realizing the upgrading from static snapshot evaluation to dynamic continuous monitoring.

Inventors

YANG HUIXIA
SUN XIULI
GENG JING
LI YAQIN
WANG SHIYAN
XIE BING
YANG YINGCHAO

Assignees

北京大学人民医院

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (11)

1. The dynamic evaluation method for the risk of prolapse recurrence of pelvic floor organ is characterized by comprising the following steps: collecting diagnosis and treatment data, structural modeling and baseline risk estimation, wherein the diagnosis and treatment data of a target patient group are collected, and the data comprise core parameters of an operation history and follow-up dynamic data streams and are stored in a time sequence structure mode to be a multi-dimensional time sequence; step two, extracting weighted time sequence characteristics based on the structured data set constructed in the step one; Thirdly, based on the weighted time sequence characteristics extracted in the second step, adopting an improved CRF-LSTM layering model to carry out dynamic risk modeling and outputting a dynamic risk sequence; And step four, based on the dynamic risk sequence output in the step three, executing individual dynamic early warning by combining the time sequence characteristics.
2. The method according to claim 1, wherein the first step comprises the steps of performing surgical history on core parameters including surgical codes, patch types, and surgical bleeding amount, and performing follow-up dynamic data flows including POP-Q stage scores, PFDI-20 scores, new disease states, and lifestyle indicators; recording for each patient when establishing the unified time-event data format in step one Wherein In order to follow-up event expiration time, For recurrence indication, and obtaining time dependent risk estimate by adopting survival type depth model for baseline risk estimate The survivor depth model is DeepSurv or a Cox-Time model.
3. The method for dynamically assessing the risk of prolapse recurrence of pelvic floor organ according to claim 1, wherein the first step further comprises generating two types of auxiliary quantities and corresponding smooth sequences, uncertainty estimates, the first type of auxiliary quantity being: Calculating to obtain a smooth sequence The second auxiliary quantity is: calculating to obtain a smooth sequence Simultaneous generation of uncertainty estimates To support subsequent threshold robustness designs, wherein uncertainty estimates Obtained by the method of MonteCarloDropout and, For the following period The month's POP-Q stage score, For post-operative 6 month POP-Q stage scores, The rate of change was scored for PFDI-20, Is the time variation.
4. The method for dynamically assessing the risk of recurrence of pelvic floor organ prolapse according to claim 3, wherein the structural dynamic index in the second step is specifically The method is used for reflecting the morphological shift degree of the pelvic floor support structure after operation along with time, and the symptom change rate index of the rate based on the time gradient is as follows Wherein For continuous follow-up interval time and employing exponentially weighted smoothing EWMA pairs Processing to obtain a smooth sequence 。
5. The method for dynamically assessing the risk of prolapse recurrence of pelvic floor organ according to claim 4, wherein the feature matrix formed by extracting the high-order statistical features in the second step is: Wherein the method comprises the steps of Is that An order derivative for capturing acceleration trends of symptom exacerbations, Is that To the point of The variance of POP-Q stage scores over the period, Is that To the point of The variance of PFDI-20 scores over the period of time, For a set time window length.
6. The method according to claim 1, wherein the step two, a feature weighting mechanism of uncertainty perception is introduced in the step two, at each time point The feature weighting coefficients are defined as Wherein Predicting uncertainty for the model in step one, and finally obtaining a calibrated weighted time-series input sequence And (3) the characteristic matrix formed in the second step.
7. The method for dynamically assessing the risk of prolapse recurrence of pelvic floor organ according to claim 6, wherein the modified CRF-LSTM layered model in step three works with the CRF layer with the core parameters of the surgical history And timing characteristics For input, joint probabilities are defined Wherein the method comprises the steps of As a function of the state characteristics, As a function of the state transition characteristics, Is normalized factor, and the CRF layer coded hidden state sequence Input to LSTM layer, LSTM layer hidden state is updated as And obtaining recurrence probability through output layer The function is activated for Sigmoid, In order to output the weight vector(s), Is a bias term.
8. The method according to claim 1, wherein the third step is to introduce a Time-to-Event loss function from Time to Event to enhance the survival Time sensitivity of the model, the loss function being Wherein the method comprises the steps of Is shown in The probability of survival without prior recurrence is that, For the indication of recurrence, For patients At the time of Probability of recurrence.
9. The method for dynamically assessing the risk of prolapse recurrence of pelvic floor organ according to claim 1, wherein the personalized dynamic early warning in the fourth step is achieved by: defining a static optimal threshold Wherein the method comprises the steps of The clinical benefit of preventing recurrence is that, Representing the cost of excessive intervention, combining historical data with decision curve analysis to determine a threshold interval where net benefit is greatest ; Constructing an individualized dynamic threshold adjustment mechanism, namely, a patient-specific threshold function , Patient baseline covariates including surgical type, age, BMI, complications, second uncertainty weighted dynamic threshold correction: and thirdly, estimating uncertainty of model output in the third step, namely, continuously triggering the condition: , in order to indicate the function, A set continuous super-threshold period number; adjusting the risk level by combining dynamic index change , The risk level is automatically raised when the risk level is increased, Is that Is set at the risk threshold value of (c), Is that Risk threshold of (2); By reinforcement learning strategies Optimizing intervention strategies, wherein states Action of Including "maintenance follow-up/increase frequency/initiate intervention" to expect a return over a long period of time Maximizing the balance between recurrence prevention benefit and intervention cost, and realizing online Bayesian updating of threshold parameters by using an intervention result back feeding model.
10. An evaluation system for application to the method of any one of claims 1-9, comprising: the data acquisition and structuring modeling module is used for executing the first step; The time sequence feature extraction module is used for executing the second step; the dynamic risk modeling module is used for executing the third step; The layering early warning and intervention decision-making module is used for executing the fourth step; all the modules work cooperatively in sequence to realize dynamic evaluation and personalized intervention of the risk of prolapse recurrence of pelvic floor organs.
11. A structured time series database for dynamic assessment of risk of prolapse recurrence of pelvic floor organs, comprising: Core data set of surgical history Including surgical coding, patch type, amount of intraoperative bleeding; follow-up dynamic data set The method comprises the steps of structurally storing the data into a multi-dimensional time sequence according to time sequence, wherein the multi-dimensional time sequence comprises POP-Q stage scores, PFDI-20 scores, new disease states and life habit indexes; time-event recording format, recording for each patient Wherein In order to follow-up event expiration time, Is an indication of recurrence; An auxiliary quantity sequence comprising 、 Is a smooth sequence of (2) And (3) with ; Uncertainty estimation sequence Obtained by the MonteCarloDropout method; Baseline risk estimation sequence Generating by a survivor type depth model; The database is used for supporting dynamic modeling, early warning and intervention decision-making of the risk of prolapse recurrence of pelvic floor organs.

Description

Dynamic evaluation method for risk of prolapse recurrence of pelvic floor organ Technical Field The invention relates to the technical field of medical artificial intelligence, in particular to a dynamic evaluation method for the risk of prolapse recurrence of pelvic floor organs. Background Pelvic floor organ prolapse is a common pelvic floor dysfunction disease of women, mainly caused by abnormal positions of pelvic organs such as uterus, vagina and the like due to damage or degeneration of a pelvic floor supporting structure, and is clinically repaired and treated by surgery. The postoperative recurrence rate is higher, and early symptoms of recurrence are hidden, if the early symptoms of recurrence are not recognized and intervened in time, the disease progress, the life quality of a patient is reduced, and even secondary operation is needed, so the accurate assessment and dynamic monitoring of postoperative recurrence risk become key links of clinical diagnosis and treatment. Currently, the traditional technique of risk assessment of prolapse recurrence of pelvic floor organs mainly adopts a mode of "static time point assessment+empirical threshold judgment". On the data acquisition level, the off-line follow-up data of fixed time points after operation (such as 3 months and 6 months after operation) are relied on, the core acquisition index is POP-Q stage score and PFDI-20 score, part of techniques can supplement operation history information, but a multi-dimensional time sequence continuously stored in time sequence is not formed, and the data presents isolated and fragmented characteristics. Vallabh-Patel et al propose 3 composite criteria for success of the surgery including subjective, objective and whether to need a re-surgery or treatment, that is, (1) the most distal prolapse of anterior and posterior walls of the vagina is less than or equal to 0cm from the hymen, and the tip descent distance is less than or equal to 1/2 of the total vaginal length. (2) The disappearance of the associated POP symptoms was judged according to the PFDI-20 3 rd problem ("frequently seeing or feeling vaginal tumor removal"). (3) no further surgery or pessary treatment due to prolapse. At the same time, the patients meeting the above 3 criteria are successful in operation, otherwise, the patients are recurrent or failed. The core defects of the traditional technology are derived from the static and homogeneous evaluation logic of the traditional technology, and the evaluation logic is seriously mismatched with the dynamic evolution of the risk of the prolapse recurrence of pelvic floor organs and the individual heterogeneous clinical characteristics, so that the risk prediction precision is low, the early warning is delayed and the intervention strategy is stiff. On the one hand, the traditional technology relies on isolated time point data, cannot capture fine dynamic changes in early recurrence, such as early signals of slow rising trend of POP-Q scores between 6 months and 12 months after operation, accelerated deterioration of PFDI-20 scores and the like, and usually needs to wait until the scores are obviously out of standard to identify risks and miss early intervention opportunities, and meanwhile, a simple statistical model cannot establish association between operation history and postoperative dynamic data, so that critical influencing factors are ignored in risk prediction, and accuracy is insufficient. On the other hand, the unified threshold decision mechanism of the traditional technology does not consider individual difference and prediction uncertainty, for example, the same risk threshold is adopted for young patients without complications and elderly patients with multiple complications, excessive intervention is easy to be caused for patients with low risk, and intervention is not timely caused by threshold rigidification for patients with high risk. Disclosure of Invention The invention provides a dynamic evaluation method for the risk of prolapse recurrence of pelvic floor organs, which solves the problems that in the related technology, the evaluation of the risk of prolapse recurrence of pelvic floor organs is mainly static and isolated time point judgment, the dynamic change trend of illness state along with time cannot be captured, and the individual difference is not targeted, so that the risk prediction precision is insufficient, early warning is delayed or excessive intervention is caused, and personalized and dynamic risk management and control are difficult to realize. The technical scheme of the invention is as follows: The dynamic evaluation method for the risk of prolapse recurrence of pelvic floor organ is executed by a computer system and comprises the following steps: and step one, diagnosis and treatment data acquisition, structural modeling and baseline risk estimation. The computer system collects diagnosis and treatment data of target patient group, including core parameters