CN-121983312-A - Dysphagia risk prediction method for ICU tracheal intubation patient in neurosurgery after extubation

CN121983312ACN 121983312 ACN121983312 ACN 121983312ACN-121983312-A

Abstract

The invention discloses a method for predicting risk of dysphagia of a patient with neurosurgery ICU trachea cannula after tube drawing, and belongs to the technical field of risk prediction of dysphagia. The invention relates to a method for predicting risk of dysphagia of a patient with neurosurgery ICU trachea cannula after tube drawing, which comprises the following steps of forming an original data set; obtaining a standardized data set, outputting a risk prediction value of dysphagia of a patient after tube drawing, obtaining an airway swallowing synergy evaluation result, and outputting a final accurate risk prediction result and an intervention suggestion. The invention solves the problems of low prediction accuracy and high missed judgment rate of neurosurgery patients in the prior art. According to the invention, the specialized risk factor set of neurosurgery is constructed, the pathogenic source of dysphagia is accurately anchored, the prediction pertinence is improved, the prediction efficiency is improved by constructing a dynamic time sequence fusion prediction model and carrying out multi-mode data weighted fusion, the hierarchical intervention closed-loop management is realized, and the medical resource allocation is optimized.

Inventors

Zhang Naqin
WANG NA
GUAN XIN
JI YUANYUAN
LI LONG
ZHANG MINGHUI

Assignees

北京市老年病医疗研究中心
首都医科大学宣武医院

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. The method for predicting the risk of swallowing disorder of the patient with the neurosurgery ICU trachea cannula after tube drawing is characterized by comprising the following steps of: s1, constructing a specialized risk factor set for neurosurgery, and simultaneously collecting related data of an ICU tracheal intubation patient for neurosurgery to form an original data set; s2, carrying out data cleaning, missing value filling, outlier rejection and data standardization on the original data set to obtain a standardized data set; s3, constructing a dynamic time sequence fusion prediction model based on the standardized data set, and outputting a risk prediction value of the dysphagia of the patient after tube drawing by using the dynamic time sequence fusion prediction model; s4, collecting multidimensional noninvasive physiological signals of the patient after tube drawing, and carrying out feature extraction and analysis on the signals to obtain an airway swallowing synergy function assessment result; And S5, constructing a multi-mode data fusion verification system, integrating a risk prediction value, an airway swallowing cooperative function evaluation result, a nerve electrophysiological signal characteristic and an image histology characteristic, performing cross verification and optimization on the prediction result, and outputting a final accurate risk prediction result and an intervention suggestion.
2. The method for predicting risk of dysphagia in a post-extubation patient with an ICU tracheal cannula according to claim 1, wherein the set of specialized risk factors for neurosurgery in S1 includes brain stem function injury classification, intracranial pressure fluctuation parameters, type of approach for neurosurgery and evaluation index of degree of injury of a swallowing-related neural pathway, and the related data of the ICU tracheal cannula patient for neurosurgery includes basic information, specialized risk factors, general clinical indexes and clinical outcome data.
3. The method for predicting risk of dysphagia after tube drawing of a patient with neurosurgical ICU tracheal intubation according to claim 2, wherein the process of constructing and training the dynamic time sequence fusion prediction model specifically comprises: extracting daily time sequence data during patient intubation from the standardized data set, and constructing a time sequence taking days as time units; dividing the time sequence by adopting a sliding window method, extracting the characteristics in each window, and forming a time sequence characteristic set; A dynamic time sequence fusion prediction model is constructed by combining a two-way long-short-term memory network with a time sequence attention mechanism; Dividing the standardized data set into a training set and a verification set according to the ratio of 7:3, training the dynamic time sequence fusion prediction model by adopting the training set, verifying the model by adopting the verification set, and optimizing model parameters by adopting a five-fold cross verification method.
4. A method for predicting risk of dysphagia in a patient after extubation of an ICU tracheal cannula according to claim 3, wherein the weight calculation of the time-series attention mechanism comprises: initializing a query vector adapting to a time sequence feature dimension based on a training target of a dynamic time sequence fusion prediction model; calculating the initial attention score of each time window feature, and obtaining the initial attention score of each time window by carrying out inner product operation on each mapped time window feature and the query vector; normalizing the initial attention scores of all the time windows by adopting a softmax function to ensure that the sum of the normalized attention scores is 1, and obtaining the basic attention weight of each time window; And multiplying the basic attention weight of the time window corresponding to the time window before tube drawing by a strengthening coefficient of 1.5 times to finish weight adjustment.
5. The method for predicting risk of dysphagia after extubation of a patient with neurosurgical ICU tracheal intubation according to claim 1, wherein the evaluation of the cooperative function of airway swallowing in S4 specifically comprises: collecting throat vibration signals in the swallowing process of a patient after tube drawing by using a laser vibration sensor; Acquiring a respiration frequency and respiration depth change signal of a patient during swallowing by using a millimeter wave radar sensor; collecting vocal cord closing function parameters including vocal cord closing time and closing amplitude by using a portable glottal graph instrument; filtering, denoising and signal synchronization are carried out on the collected multidimensional signals, and feature extraction is carried out on the synchronized signals; Comparing the extracted characteristics with a preset normal reference range, and classifying the airway swallowing synergy function evaluation result into four grades of normal, mild abnormality, moderate abnormality and severe abnormality.
6. The method for predicting risk of dysphagia after tube drawing of a patient with neurosurgical ICU tracheal intubation according to claim 1, wherein the construction process of the multi-modal data fusion verification system specifically comprises the following steps: extracting the neuroelectrophysiology signal characteristics and the image histology characteristics of the patient respectively, adopting an L1 regularization method to reduce the dimension of the extracted image histology characteristics, and retaining important characteristics; integrating the risk prediction value, the airway swallowing coordination function evaluation result, and the neuroelectrophysiologic signal characteristic and the image histology characteristic to carry out weighted summation so as to obtain a multi-mode fusion comprehensive risk score; And designing a risk level dynamic updating and layering early warning mechanism according to the comprehensive risk score, outputting a final accurate risk prediction report according to the comprehensive risk grading result, and making personalized intervention suggestions for patients with different risk levels.
7. The method for predicting risk of dysphagia in a patient after extubation of an ICU tracheal cannula according to claim 6, wherein the process for obtaining the multimodal fusion composite risk score comprises: adopting a weighted fusion strategy to integrate a risk prediction value of a dynamic time sequence fusion prediction model, an airway swallowing cooperative function evaluation result, a neuroelectrophysiologic signal characteristic and an image histology characteristic; Calculating the weight coefficient of each mode data by adopting an analytic hierarchy process; And carrying out weighted summation on the weight coefficients of the modal data to obtain the comprehensive risk score after multi-modal fusion.
8. The method for predicting risk of dysphagia in a patient after extubation of an ICU tracheal cannula according to claim 7, wherein the specific implementation of the analytic hierarchy process comprises: Constructing a hierarchical structure model, determining a target layer by using multi-mode data fusion weights, determining a criterion layer by using the prediction accuracy, sensitivity and specificity of each mode data, and determining a scheme layer by using each mode data; constructing a judgment matrix, carrying out pairwise comparison scoring on each criterion layer index and scheme layer data, and respectively constructing a criterion layer judgment matrix and a scheme layer judgment matrix under each criterion according to the scoring result; Calculating the maximum eigenvalue and the corresponding eigenvector of the judgment matrix, and carrying out normalization processing on the eigenvector to obtain the weight coefficient of each mode data; And calculating a consistency index and a consistency proportion, and when the consistency proportion is smaller than 0.1, considering that the judgment matrix has satisfactory consistency, and the weight coefficient is effective.
9. The method for predicting risk of dysphagia in a patient after tube drawing in an ICU tracheal intubation of claim 8, wherein the calculating process of the weight coefficient of each mode data comprises Based on each constructed judgment matrix, calculating the maximum eigenvalue of the matrix and the corresponding eigenvector thereof by a sum-product method; Normalizing the feature vector to obtain the weight of each index of the criterion layer and the weight of each mode data of the scheme layer relative to each index of the criterion layer And carrying out weighted summation on the scheme layer weights by combining the criterion layer weights to obtain the comprehensive weight coefficient of each mode data.
10. The method for predicting risk of dysphagia of a patient with neurosurgical ICU tracheal intubation after tube drawing according to claim 6, wherein the construction of the risk level dynamic updating and layering early warning mechanism specifically comprises: Setting a swallowing disorder risk grade judgment standard, and classifying risk grades into three grades of high risk, medium risk and low risk; Constructing a risk level dynamic updating mechanism, automatically updating the risk level once according to a set time interval by combining multisource real-time fusion data based on a real-time prediction result output by a dynamic time sequence fusion prediction model, and immediately triggering emergency updating when a sudden risk factor occurs; designing a layered early warning mechanism, and setting corresponding early warning modes and pushing objects according to different risk levels; And establishing an early warning response tracking mechanism, recording response time of medical staff to early warning information, and taking intervention measures and intervention effects, and feeding the intervention effect data back to the dynamic time sequence fusion prediction model.

Description

Dysphagia risk prediction method for ICU tracheal intubation patient in neurosurgery after extubation Technical Field The invention relates to the technical field of dysphagia risk prediction, in particular to a method for predicting dysphagia risk of a neurosurgery ICU tracheal intubation patient after tube drawing. Background Dysphagia is one of the common complications of neurosurgical ICU tracheal intubation patients after tube drawing, with a incidence rate of up to 30% -60%. Once a patient suffers from dysphagia, a series of secondary problems such as aspiration pneumonia, malnutrition, dehydration and the like are extremely easy to cause, so that not only can the hospitalization time be prolonged and the medical cost be increased, but also the disability rate and the death rate of the patient can be obviously improved, and the prognosis of the patient is seriously and negatively influenced. Chinese patent publication No. CN117224084a discloses a method and apparatus for predicting risk of acquired dysphagia in ICU patients, the method comprising measuring tongue pressure and oral cavity stereo sense and entering into a data collection module; the method comprises the steps of inputting conventional influencing factors into a data collection module, importing measured parameter data into a data cleaning module, and inputting cleaned data into a prediction module to predict the occurrence risk of PED. According to the method, oral cavity stereo sense and tongue pressure are listed in the construction of the PED prediction model, so that ICU medical staff can be helped to screen patients with high risk of dysphagia in an early stage, and early recognition of the dysphagia is realized, and therefore, the rehabilitation of swallowing can be carried out in an early stage. The patent above-mentioned neglects the dysphagia of neurosurgery ICU patient in the actual use to lead to by brain stem injury, intracranial pressure unusual, swallowing relevant neural route impaired, the influence of special operation access totally, the general factor can't cover above-mentioned core pathogenic root, and then lead to the prediction rate of accuracy is low and the rate of missed judgement to neurosurgery patient is high, consequently, does not satisfy current demand, has provided the dysphagia risk prediction method after the trachea cannula patient of neurosurgery ICU pulls out to this we. Disclosure of Invention The invention aims to provide a method for predicting dysphagia risk of a patient with an ICU trachea cannula in neurosurgery, which is used for accurately anchoring a dedicated pathogenic source of dysphagia by constructing a specialized risk factor set in neurosurgery, improving prediction pertinence, constructing a dynamic time sequence fusion prediction model based on cannula time sequence data, integrating multi-mode data hierarchy weighted fusion verification, remarkably improving prediction efficiency, outputting personalized intervention in a grading manner, realizing full-flow closed loop, optimizing medical resource configuration and solving the problems in the background art. The technical scheme of the invention for realizing the aim is that the method for predicting the risk of dysphagia of a patient with the trachea cannula of the ICU in neurosurgery after tube drawing comprises the following steps: s1, constructing a specialized risk factor set for neurosurgery, and simultaneously collecting related data of an ICU tracheal intubation patient for neurosurgery to form an original data set; s2, carrying out data cleaning, missing value filling, outlier rejection and data standardization on the original data set to obtain a standardized data set; s3, constructing a dynamic time sequence fusion prediction model based on the standardized data set, and outputting a risk prediction value of the dysphagia of the patient after tube drawing by using the dynamic time sequence fusion prediction model; s4, collecting multidimensional noninvasive physiological signals of the patient after tube drawing, and carrying out feature extraction and analysis on the signals to obtain an airway swallowing synergy function assessment result; And S5, constructing a multi-mode data fusion verification system, integrating a risk prediction value, an airway swallowing cooperative function evaluation result, a nerve electrophysiological signal characteristic and an image histology characteristic, performing cross verification and optimization on the prediction result, and outputting a final accurate risk prediction result and an intervention suggestion. Preferably, the set of specialized risk factors for neurosurgery in S1 includes brain stem function injury classification, intracranial pressure fluctuation parameters, neurosurgery access type and swallowing related nerve access injury degree evaluation index, and the related data of the neurosurgery ICU tracheal intubation patient includes basic information, specialized ris