CN-121393907-B - Method, medium and equipment for early warning risk of disease severity of enteritis patient

CN121393907BCN 121393907 BCN121393907 BCN 121393907BCN-121393907-B

Abstract

The invention discloses a method, medium and equipment for early warning of the disease severity of enteritis patients, which are characterized in that a sensing device is used for collecting an original signal of borborygmus and a clinical multidimensional physiological parameter sequence, an acoustic biomarker time sequence is generated by constructing a dynamic feature map of borygmus based on the original signal of borygmus, the acoustic biomarker time sequence and the clinical multidimensional physiological parameter sequence are input into a multimodal fusion early warning model to obtain an intestinal inflammation risk index, a signal quality self-evaluation flow is executed, a data quality warning code is generated when the intestinal inflammation risk index is abnormal, a multi-node collaborative monitoring mechanism is triggered based on the risk index to generate an intestinal state multidimensional situation map, an individualized risk baseline is established, a grading early warning instruction is generated, and finally a comprehensive early warning report is output. The invention realizes the multi-mode fusion analysis of the borborygmus signal and the clinical parameters, remarkably improves the accuracy and timeliness of the early warning of the illness state through dynamic risk assessment and signal quality monitoring, and provides a reliable basis for clinical decision.

Inventors

HUANG BIN
AN HONGLIN
Qiu Yiman
LI RUOFEI
ZHENG MINGXUAN
WU HUAPING
CHEN MINGJUN

Assignees

福建中医药大学

Dates

Publication Date: 20260508
Application Date: 20251225

Claims (9)

1. The method for early warning the severity risk of the enteritis patient is characterized by comprising the following steps: continuously acquiring an original abdominal borborygmus signal of a patient through a sensing device, and synchronously acquiring a clinical multidimensional physiological parameter sequence which at least comprises an inflammation marker index and a basic metabolism index; Based on the original borborygmus signal, constructing a borygmus dynamic characteristic map through an acoustic event detection and pattern recognition algorithm, and generating an acoustic biomarker time sequence reflecting the functional state of the intestinal tract; Inputting the acoustic biomarker time sequence and the clinical multidimensional physiological parameter sequence into a multimodal fusion early warning model, and realizing cross-modal semantic alignment through a characteristic interaction layer weighted by an attention mechanism to obtain an intestinal inflammation risk index; In the monitoring process, a self-evaluation flow of the bowel sound signal quality is periodically executed to obtain a self-evaluation result of the signal quality; when the signal quality self-evaluation result is abnormal, generating a data quality warning code based on the signal quality self-evaluation result, and pushing the data quality warning code to the monitoring terminal through a medical internet of things protocol; Based on the intestinal inflammation risk index, carrying out dynamic threshold judgment, triggering a multi-node cooperative monitoring mechanism when the risk index is detected to exceed the self-adaptive risk threshold, guiding the associated physiological parameter acquisition equipment to enter an enhanced sampling mode, and generating an intestinal state multidimensional situation map; establishing a patient individuation risk baseline, continuously monitoring risk evolution trend, and starting an intervention priority evaluation flow to generate a grading early warning instruction when detecting that the risk accumulation rate exceeds a trend early warning threshold; Based on the intestinal inflammation risk index, the data quality warning code, the intestinal state multidimensional situation map and the grading early warning instruction, generating and outputting an enteritis illness severity comprehensive early warning report; in the monitoring process, a self-evaluation flow of the bowel sound signal quality is periodically executed to obtain a signal quality self-evaluation result, which comprises the following steps: Performing quality evaluation on the original signal of the borborygmus in a preset evaluation period, and generating a signal purity index by calculating the ratio of the signal-to-noise ratio in a signal section to the power spectral density of background noise; analyzing the physiological rhythm characteristics of the borborygmus signal, calculating the periodic intensity of the signal through an autocorrelation function, and comparing the periodic intensity with a preset typical intestinal peristalsis rhythm range to generate a physiological consistency index; detecting the dynamic range of the signal amplitude, and identifying the saturation distortion and the motion artifact interference of the signal by counting the joint distribution characteristics of the zero crossing rate and the amplitude variance of the signal to generate a signal integrity index; Extracting signal frequency domain characteristics, separating effective borborygmus sound components from interference components on a mel frequency spectrum through a principal component analysis algorithm, calculating an effective frequency band energy duty ratio, and generating a frequency spectrum effectiveness index; Constructing a comprehensive evaluation model based on fuzzy logic, taking the signal purity index, the physiological consistency index, the signal integrity index and the frequency spectrum effectiveness index as input variables, and calculating a comprehensive quality score through membership function and rule reasoning; And generating a signal quality self-evaluation result comprising a quality grade and an abnormal type according to the comparison result of the comprehensive quality score and a preset quality threshold.
2. The method for early warning of the severity risk of a patient suffering from enteritis according to claim 1, wherein the step of constructing a dynamic feature map of borborygmus through an acoustic event detection and pattern recognition algorithm based on the original borygmus signal to generate an acoustic biomarker time sequence reflecting the functional state of the intestinal tract comprises the steps of: Preprocessing the original signal of the borborygmus in a preset signal analysis time window, and eliminating the interference of heart sounds, breathing sounds and environmental noise through spectral subtraction to obtain a pure borygmus signal; performing acoustic event detection on the pure borborygmus signal, and adopting a double-threshold end point detection algorithm based on an entropy ratio to locate the starting point and the ending point of a single borygmus event so as to generate a borygmus event sequence; Extracting features of the borygmus sound event sequence, and calculating event features of each event, wherein the event features comprise time domain features, frequency domain features and time-frequency domain joint features, the time domain features comprise short-time energy and zero crossing rate, the frequency domain features comprise mel frequency cepstrum coefficients and spectrum centroids, and the time-frequency domain joint features comprise wavelet packet energy entropy; based on the extracted event characteristics, dividing the borborygmus sound event into three pathological types of normal peristaltic sound, high Kang Kangjin sound and weak rare sound through a Gaussian mixture model clustering algorithm, and establishing a borborygmus sound event classification label; in a continuous monitoring period, counting the space-time distribution characteristics of each pathological type borborygmus event, calculating the cluster density of event burst, the duration of silence period and the entropy of rhythm samples, and forming a borygmus dynamic characteristic vector sequence; And inputting the borborygmus dynamic feature vector sequence into a long-short-period memory network for time sequence mode learning, and weighting and fusing the feature contribution degrees of different time steps through an attention mechanism to generate an acoustic biomarker time sequence reflecting the functional state of the intestinal tract.
3. The method for early warning of risk of severity of a patient suffering from enteritis according to claim 1, wherein inputting the acoustic biomarker sequence and the clinical multidimensional physiological parameter sequence into a multimodal fusion early warning model, realizing cross-modal semantic alignment through a feature interaction layer weighted by an attention mechanism, and obtaining an intestinal inflammation risk index comprises the following steps: performing feature coding on the acoustic biomarker time sequence, extracting a local dependence mode of acoustic features by using a time sequence convolution network, and generating acoustic advanced feature representation; feature coding is carried out on the clinical multidimensional physiological parameter sequence, nonlinear combination features of physiological parameters are extracted through a multi-layer perceptron, and clinical advanced feature representation is generated; inputting the acoustic advanced feature representation and the clinical advanced feature representation into a cross-modal attention interaction layer, and calculating an association weight matrix between the acoustic feature and the clinical feature through a bidirectional attention mechanism; Weighting and fusing the acoustic advanced feature representation and the clinical advanced feature representation based on the association weight matrix to generate a cross-modal joint feature representation; inputting the cross-modal joint characteristic representation into a risk prediction module, learning a time sequence evolution rule of the characteristic representation through a gating circulation unit, and focusing the characteristic state of a key time node by adopting a self-attention mechanism; Based on the learned time sequence evolution rule, calculating the inflammation risk probability of each time step through the full-connection layer and the activation function, and carrying out weighted aggregation on the risk probabilities in the continuous time window to generate the intestinal inflammation risk index.
4. The method for early warning of the severity of a patient suffering from enteritis according to claim 1, wherein the step of performing dynamic threshold judgment based on the risk index of intestinal inflammation, and triggering a multi-node collaborative monitoring mechanism to guide the associated physiological parameter acquisition equipment to enter an enhanced sampling mode when the risk index is detected to exceed an adaptive risk threshold, and generating an intestinal state multidimensional situation map comprises the steps of: establishing a dynamic threshold model based on patient historical baseline data, and calculating an adaptive risk threshold of the intestinal inflammation risk index through an index weighted moving average algorithm; Comparing the current risk index of the intestinal inflammation with the self-adaptive risk threshold in real time, and triggering a cooperative monitoring activation signal when the risk index of the intestinal inflammation continuously exceeds the self-adaptive risk threshold and reaches a preset risk duration; According to the collaborative monitoring activation signal, a control instruction is sent to an associated body temperature monitoring device, a heart rate variability acquisition module and a body surface electrophysiological sensor so as to switch to an enhanced sampling mode; Under the enhanced sampling mode, synchronously acquiring high-frequency physiological parameter data, including a core body temperature fluctuation sequence, heart rate variability frequency domain indexes and abdomen impedance change tracks; Carrying out space-time alignment processing on physiological parameter data acquired synchronously by multiple sources, eliminating time offset among signals through a dynamic time warping algorithm, and constructing a space-time aligned multi-mode data cube; extracting space-time joint characteristics from the multi-modal data cube by adopting a tensor decomposition algorithm, and identifying a collaborative variation mode of different physiological parameters in space-time dimension by adopting a non-negative matrix decomposition algorithm; Based on the extracted space-time joint characteristics, an intestinal state multidimensional situation map is generated through a self-organizing mapping neural network.
5. The method for pre-warning the severity of a patient suffering from enteritis according to claim 4, wherein the method for extracting the spatio-temporal joint features from the multi-modal data cube by using a tensor decomposition algorithm and identifying the collaborative variation pattern of different physiological parameters in the spatio-temporal dimension by using a non-negative matrix decomposition algorithm comprises: Constructing the multi-modal data cube as a data structure comprising third-order tensors of a temporal dimension, a spatial dimension, and a modal dimension; performing Tucker decomposition on the constructed third-order tensor, extracting a tensor core matrix as a space-time feature base, and acquiring a factor matrix corresponding to each dimension; Calculating coupling weights of different physiological parameters in space-time dimensions based on the factor matrix, and identifying a dominant change mode through eigenvalue decomposition; performing non-negative matrix decomposition on the tensor core matrix, and decomposing a space-time feature matrix into a mode dictionary matrix and an activation coefficient matrix; Screening feature modes with obvious physiological significance in the mode dictionary matrix through a sparse coding algorithm to generate space-time joint features; calculating the mutual information association degree between each characteristic mode in the space-time combined characteristics, and constructing a physiological parameter cooperative change network; and identifying the cooperative change modes of different physiological parameters in the space-time dimension based on the topological structure characteristics of the physiological parameter cooperative change network.
6. The method of claim 1, wherein establishing a patient personalized risk baseline, continuously monitoring risk evolution trend, and initiating an intervention priority assessment procedure when a risk accumulation rate exceeding a trend early warning threshold is detected, generating a hierarchical early warning instruction, comprises: Collecting an intestinal inflammation risk index historical sequence of a patient in a preset baseline learning period, fitting an individualized risk baseline model through a robust regression algorithm, and generating a patient-specific risk reference curve; Continuously monitoring a real-time intestinal inflammation risk index, identifying mutation points in a risk sequence by adopting a mutation point detection algorithm, and calculating a risk accumulation rate between adjacent mutation points; Based on a comparison result of the risk accumulation rate and a preset trend early warning threshold, starting an intervention priority evaluation flow when detecting that the risk accumulation rate exceeds the trend early warning threshold; Extracting a multi-dimensional risk feature vector at the current moment, wherein the multi-dimensional risk feature vector comprises a risk index gradient, a risk duration, a multi-mode feature deviation degree and a history response mode matching degree; Inputting the multidimensional risk feature vector into a pre-trained priority evaluation model, and calculating priority scores of different intervention levels through a random forest multi-classification algorithm; and generating a grading early warning instruction comprising three grades of emergency intervention, priority treatment and conventional observation according to a comparison result of the priority grade and a preset grading threshold.
7. The method for pre-warning the severity of a patient suffering from enteritis according to claim 6, wherein extracting the multi-dimensional risk feature vector at the current moment, including risk index gradient, risk duration, multi-modal feature deviation and historical response pattern matching, comprises: Calculating a first derivative sequence of an intestinal inflammation risk index in a current time window, smoothing by Savitzky-Golay filtering, and taking a maximum value as a risk index gradient; counting the duration from the risk index exceeding the personalized risk baseline to the current moment for the first time, and calculating the risk duration by combining the integral area of the risk index; Extracting the time sequence characteristics of the acoustic biomarker and the clinical multidimensional physiological parameter characteristics at the current moment, respectively calculating the mahalanobis distance between the acoustic biomarker and the patient in the history normal state, and obtaining the multi-modal characteristic deviation degree after weighted fusion; Constructing a historical response mode matching library based on dynamic time warping, performing similarity calculation on the current risk evolution track and typical response modes in the historical library, and selecting the highest similarity as the matching degree of the historical response modes; And combining the risk index gradient, the risk duration quantization index, the multi-mode feature deviation degree and the historical response mode matching degree into a multi-dimensional risk feature vector.
8. A computer readable storage medium, on which computer program instructions are stored, which computer program instructions, when executed by a processor, implement the method of any one of claims 1 to 7.
9. An electronic device comprising a memory and a processor, wherein the memory is configured to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method of any of claims 1-7.

Description

Method, medium and equipment for early warning risk of disease severity of enteritis patient Technical Field The invention relates to the technical field of medical monitoring, in particular to a method, medium and equipment for early warning of the severity risk of enteritis patients. Background Enteritis is a common digestive system disease in clinic, the disease development often presents dynamic change characteristics, and accurate evaluation of the severity of the disease is of great significance to treatment decision and prognosis improvement. At present, the clinical monitoring of enteritis patients mainly depends on laboratory examination, imaging evaluation and clinical symptom observation, and the methods are mostly carried out discontinuously, so that continuous and dynamic monitoring of intestinal function states is difficult to realize. In particular, for the important index of intestinal tract movement function, namely borborygmus, the prior art is mostly stopped at qualitative auscultation or short-time recording, and lacks a systematic signal analysis and multi-parameter fusion mechanism. Meanwhile, the clinical acquired physiological parameters and the borborygmus signals have modal differences, and effective information integration and collaborative analysis are difficult to realize. In addition, the existing monitoring schemes have insufficient control over data quality and lack of dynamic risk assessment capability for individual differences, so that timeliness and accuracy of disease early warning are limited. Disclosure of Invention In view of the problems, the invention provides a method, medium and equipment for early warning the severity risk of enteritis patients, which are used for realizing continuous and dynamic evaluation of intestinal inflammation risks and solving the problem that the severity of enteritis patients is difficult to early warn timely and accurately by fusing the dynamic characteristics of borborygmus and clinical physiological parameters and establishing a multi-mode risk early warning model. In order to achieve the above object, in a first aspect, the present application provides a method for early warning of risk of severity of an enteritis patient, comprising: Continuously acquiring an original abdominal borborygmus signal of a patient through a sensing device, and synchronously acquiring a clinical multidimensional physiological parameter sequence which at least comprises an inflammation marker index and a basic metabolism index; based on the original signal of the borborygmus, constructing a dynamic feature map of the borborygmus through an acoustic event detection and pattern recognition algorithm, and generating an acoustic biomarker time sequence reflecting the functional state of the intestinal tract; inputting the acoustic biomarker time sequence and the clinical multidimensional physiological parameter sequence into a multimodal fusion early warning model, and realizing the cross-modal semantic alignment through a characteristic interaction layer weighted by an attention mechanism to obtain an intestinal inflammation risk index; In the monitoring process, a self-evaluation flow of the bowel sound signal quality is periodically executed to obtain a self-evaluation result of the signal quality; when the signal quality self-evaluation result is abnormal, generating a data quality warning code based on the signal quality self-evaluation result, and pushing the data quality warning code to the monitoring terminal through a medical internet of things protocol; Based on the risk index of the intestinal inflammation, carrying out dynamic threshold judgment, triggering a multi-node cooperative monitoring mechanism when the risk index is detected to exceed the self-adaptive risk threshold, guiding the associated physiological parameter acquisition equipment to enter an enhanced sampling mode, and generating an intestinal state multidimensional situation map; establishing a patient individuation risk baseline, continuously monitoring risk evolution trend, and starting an intervention priority evaluation flow to generate a grading early warning instruction when detecting that the risk accumulation rate exceeds a trend early warning threshold; based on the intestinal inflammation risk index, the data quality warning code, the intestinal state multidimensional situation map and the grading early warning instruction, a comprehensive early warning report of the enteritis severity is generated and output. Further, based on the original signal of the borborygmus, the dynamic feature map of the borborygmus is constructed through an acoustic event detection and pattern recognition algorithm, and an acoustic biomarker time sequence reflecting the functional state of the intestinal tract is generated, and the method comprises the following steps: Preprocessing an original signal of the borborygmus in a preset signal analysis time window, and eliminating the interference o