CN-121839006-B - Clinical auxiliary decision-making method and system based on integration time sequence multi-modal data

CN121839006BCN 121839006 BCN121839006 BCN 121839006BCN-121839006-B

Abstract

The invention discloses a clinical auxiliary decision-making method and a system based on integrated time sequence multi-modal data, which relate to the technical field of medical information and comprise the steps of collecting and time-aligning time sequence multi-modal clinical data of a patient, carrying out feature analysis and cross-modal fusion, and generating a unified multi-dimensional time sequence health state feature spectrum. Based on the characteristic spectrum and the medical knowledge graph, a clinical state evolution model is established, evolution distances between the current state of the patient and each key clinical event node are calculated, and then a dynamic risk assessment curved surface is established to identify a high risk evolution path. And for the high risk path, backtracking and positioning the key feature combination, retrieving a corresponding intervention measure evidence chain from the knowledge graph, and finally generating a clinical decision support report. The invention realizes dynamic quantitative risk assessment and automatic evidence-based decision recommendation for disease course evolution.

Inventors

WANG BOHAN
LU JIACHENG

Assignees

广州知汇云科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260313

Claims (8)

1. The clinical auxiliary decision-making method based on the integrated time sequence multi-mode data is characterized by comprising the following steps of: collecting patient time sequence multi-mode clinical data from different medical equipment and information systems; performing time alignment and missing value interpolation on the time sequence multi-modal clinical data to construct a synchronous multi-modal data stream taking a uniform time axis as a reference; Respectively carrying out characteristic analysis on each mode data in the synchronous multi-mode data stream, and extracting a characteristic set reflecting the physiological state of the disease; Performing cross-modal fusion on the feature sets extracted by different modalities to generate a unified multi-dimensional time sequence health state feature spectrum of the patient; based on the multidimensional time sequence health state characteristic spectrum and in combination with a preset medical knowledge graph, a clinical state evolution model for representing the dynamic evolution of the course of the patient is established, and the clinical state evolution model specifically comprises the following steps: Mapping the feature vector of each time point of the multi-dimensional time sequence health state feature spectrum into one state node in a clinical state evolution model, establishing a directional connection edge between the state nodes of adjacent time points according to the disease development path and state transition relation defined in the medical knowledge graph, assigning a weight to each directional connection edge, wherein the weight is calculated based on the feature vector change amplitude and a typical mode of state transition in the medical knowledge graph; Defining key clinical event nodes in the clinical state evolution model, and calculating the evolution distance between the current state of the patient and each key clinical event node; Constructing a dynamic risk assessment curved surface for the evolution of the patient to different key clinical events according to the evolution distance and a preset evolution threshold; For the high-risk evolution path identified in the dynamic risk assessment curved surface, backtracking the multi-dimensional time sequence health state characteristic spectrum, positioning key characteristic combinations and searching intervention measure evidence chains from the medical knowledge graph according to the key characteristic combinations, and generating a clinical decision support report, wherein the method comprises the following steps of: The method comprises the steps of identifying a high-risk evolution path in a dynamic risk assessment curved surface, backtracking the multi-dimensional time sequence health state characteristic spectrum, positioning key characteristic combinations which cause risk elevation, searching an associated intervention evidence chain from the medical knowledge graph according to the key characteristic combinations, synthesizing the dynamic risk assessment curved surface, the high-risk evolution path and the associated intervention evidence chain, and generating a structured clinical decision support report; according to the key feature combination, retrieving an associated intervention measure evidence chain from the medical knowledge graph, wherein the intervention measure evidence chain specifically comprises the following steps: The method comprises the steps of taking a key feature combination as a query condition, inputting the key feature combination into a medical knowledge graph, matching pathophysiological state nodes with direct or indirect causal and correlative relations with the key feature combination in the medical knowledge graph, searching knowledge paths connecting the pathophysiological state nodes with various clinical intervention measure nodes, carrying out evidence rank ordering and correlative scoring on all the searched knowledge paths, screening out a plurality of highest-scoring knowledge paths as recommended intervention measure evidence links, wherein the intervention measure evidence links describe logic reasoning links from the identified key features to specific intervention measures.
2. The method for clinical aid decision making based on integrated time series multi-modal data according to claim 1, wherein the collecting of patient time series multi-modal clinical data from different medical devices and information systems is specifically: The time sequence multi-mode clinical data at least comprises vital sign waveforms, medical image sequences and structured text records; continuously acquiring vital sign waveform data from a monitoring device interface, wherein the vital sign waveform data comprises an electrocardiogram, a blood oxygen saturation waveform and an arterial blood pressure waveform; Retrieving a medical image sequence of a patient from an image archiving and communication system according to an examination time sequence, wherein the medical image sequence comprises a computed tomography sequence and a magnetic resonance imaging sequence; Extracting structured text records corresponding to the patient from a hospital information system and a laboratory information system, wherein the structured text records comprise medical advice records, laboratory test reports and nursing records; And labeling the original acquisition time point and the data source mode type of each time sequence multi-mode clinical data.
3. The method for clinical decision assistance based on integrated time-series multi-modal data according to claim 2, wherein the time-series multi-modal clinical data is subjected to time alignment and missing value interpolation to construct a synchronized multi-modal data stream based on a unified time axis, specifically: Converting the time stamps of all data in the time sequence multi-mode clinical data to the same standard time coordinate system; For the converted time sequence multi-mode clinical data, establishing a unified time axis index by taking a preset time granularity as a reference; checking the integrity of the modal data at each point in time of the timeline index; For the time points with data missing, performing numerical interpolation by adopting an interpolation method based on the same-mode data of adjacent time points or an extrapolation method based on cross-mode data correlation; And reorganizing the interpolated modal data according to the uniform time axis index to generate synchronous multi-modal data streams with one-to-one correspondence of time points.
4. The clinical assistant decision-making method based on integrated time sequence multi-modal data according to claim 3, wherein the feature analysis is performed on each modal data in the synchronized multi-modal data stream, and the feature set reflecting the physiological state of the disease is extracted, specifically: analyzing the vital sign waveform data in the synchronous multi-mode data stream in a time domain, a frequency domain and a time-frequency domain to extract waveform morphological characteristics, rhythm characteristics and energy distribution characteristics; Performing focus region segmentation and quantification based on deep learning on the medical image sequences in the synchronous multi-mode data stream, and extracting image histology characteristics including texture characteristics, shape characteristics and intensity statistical characteristics; carrying out medical entity identification and relation extraction on the structured text records in the synchronous multi-mode data stream, and constructing clinical event characteristics including diagnosis events, medication events and symptom events; Waveform morphological features, rhythm features, energy distribution features, image histology features and clinical event features extracted for each mode data are respectively summarized into feature sets of each corresponding mode.
5. The method for clinical decision-making assistance based on integrated time-series multi-modal data according to claim 4, wherein the cross-modal fusion is performed on the feature sets extracted by different modalities to generate a unified multi-dimensional time-series health status feature spectrum of the patient, specifically: mapping the feature sets of different modalities to the same high-dimensional potential feature space; calculating correlation weights among different modal features in the high-dimensional potential feature space; according to the correlation weight, carrying out weighted fusion on the characteristics which come from different modes and reflect the same physiological or pathological dimension to form a fused dimension characteristic; All the fused dimension features are arranged and combined according to the unified time axis index; the patient-unified multi-dimensional time-sequential health status profile consists of all the fused dimensional features at each time point.
6. The method for clinical aid decision making based on integrated time series multi-modal data according to claim 5, wherein the step of defining critical clinical event nodes in the clinical state evolution model and calculating the evolution distance between the current state of the patient and each critical clinical event node is as follows: Extracting predefined critical clinical event nodes with important clinical significance from the medical knowledge graph, wherein the critical clinical event nodes comprise disease deterioration turning points, complication occurrence points and treatment response evaluation points; Identifying, in the clinical state evolution model, a state node that is closest in feature space to the critical clinical event node; Taking a state node corresponding to the latest time point of the patient as a current state node; Calculating the shortest path length from the current state node to the state node corresponding to each key clinical event node along the directed connecting edge in the clinical state evolution model; The shortest path length is the evolution distance between the current state of the patient and each critical clinical event node.
7. The method for clinical decision assistance based on integrated time series multi-modal data according to claim 6, wherein the step of constructing a dynamic risk assessment curved surface for patient evolution to different critical clinical events according to the evolution distance and a preset evolution threshold is specifically as follows: setting a corresponding evolution threshold value for each key clinical event node; dividing the evolution distance between the current state of the patient and each key clinical event node by the corresponding evolution threshold value to obtain normalized relative evolution urgency; Constructing a multidimensional risk vector by taking different key clinical events as dimensions and relative evolution urgency as a numerical value; Mapping the multidimensional risk vector into a continuous two-dimensional or three-dimensional space through a nonlinear function to form a dynamic risk assessment curved surface; each point on the dynamic risk assessment surface represents a particular combination of risk states, the height or color of which represents the overall risk level.
8. A clinical aid decision making system based on integrated time series multimodal data comprising a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of a clinical aid decision making method based on integrated time series multimodal data as claimed in any one of claims 1 to 7.

Description

Clinical auxiliary decision-making method and system based on integration time sequence multi-modal data Technical Field The invention belongs to the technical field of medical information, and particularly relates to a clinical auxiliary decision-making method and system based on integration time sequence multi-mode data. Background In clinical practice, the health status of a patient is continuously generated with time-sequential multimodal data by monitoring devices, laboratory systems, imaging devices, etc. In the prior art, a data fusion and machine learning method is generally adopted to integrate and analyze the multi-mode data so as to realize disease risk prediction or auxiliary diagnosis. However, these approaches have focused on static risk assessment at a single point in time or short term window, or on prediction only for a particular clinical endpoint event. They are difficult to continuously and quantitatively characterize the trend and real-time risk of a patient's dynamic progression of their health status to a variety of potentially critical clinical events throughout the course of the disease. Existing clinical auxiliary decision-making systems often stop sending out early warning after identifying high risk. The system lacks a mechanism to automatically correlate the identified high risk path to a specific, interpretable combination of time-sequential physiological characteristics that results in that risk. At the same time, the ability to automatically match and retrieve evidence of corresponding interventions from structured medical knowledge based on these key features is lacking. This results in faults between risk early warning and specific, operational clinical decision advice, and insufficient automation and interpretability of the decision support process. Disclosure of Invention The present invention aims to solve at least one of the technical problems existing in the prior art; therefore, the invention provides a clinical auxiliary decision-making method based on integration time sequence multi-modal data, which comprises the following steps: collecting patient time sequence multi-mode clinical data from different medical equipment and information systems; performing time alignment and missing value interpolation on the time sequence multi-modal clinical data to construct a synchronous multi-modal data stream taking a uniform time axis as a reference; Respectively carrying out characteristic analysis on each mode data in the synchronous multi-mode data stream, and extracting a characteristic set reflecting the physiological state of the disease; Performing cross-modal fusion on the feature sets extracted by different modalities to generate a unified multi-dimensional time sequence health state feature spectrum of the patient; Based on the multi-dimensional time sequence health state characteristic spectrum, a clinical state evolution model for representing the dynamic evolution of the course of the patient is established by combining a preset medical knowledge graph; Defining key clinical event nodes in the clinical state evolution model, and calculating the evolution distance between the current state of the patient and each key clinical event node; Constructing a dynamic risk assessment curved surface for the evolution of the patient to different key clinical events according to the evolution distance and a preset evolution threshold; And backtracking the multi-dimensional time sequence health state characteristic spectrum aiming at the high risk evolution path identified in the dynamic risk assessment curved surface, positioning key characteristic combinations and searching an intervention measure evidence chain from the medical knowledge graph according to the key characteristic combinations to generate a clinical decision support report. Further, the acquisition of patient time sequence multi-mode clinical data from different medical equipment and information systems is specifically as follows: The time sequence multi-mode clinical data at least comprises vital sign waveforms, medical image sequences and structured text records; continuously acquiring vital sign waveform data from a monitoring device interface, wherein the vital sign waveform data comprises an electrocardiogram, a blood oxygen saturation waveform and an arterial blood pressure waveform; Retrieving a medical image sequence of a patient from an image archiving and communication system according to an examination time sequence, wherein the medical image sequence comprises a computed tomography sequence and a magnetic resonance imaging sequence; Extracting structured text records corresponding to the patient from a hospital information system and a laboratory information system, wherein the structured text records comprise medical advice records, laboratory test reports and nursing records; And labeling the original acquisition time point and the data source mode type of each time sequence multi-mode clinical