Structured Data Processing System and Method for Diagnostic Images of Septic Shock

Abstract

The present invention relates to the technical field of medical information processing, and discloses a structured data processing system and method for infectious shock diagnosis and treatment image information. The system includes a processor, an image acquisition module, a multimodal data fusion module and other modules. The image acquisition module acquires CT images, ultrasound images and hemodynamic time series data. The multimodal data fusion module processes them to generate multimodal fusion data, which is then generated into dynamic feature vectors by the dynamic feature extraction module, and then a structured diagnosis and treatment record is generated by the structured processing engine. The anomaly detection module monitors the recorded parameters, the diagnosis and treatment decision-making module generates diagnosis and treatment suggestions according to the anomaly signals, the early warning feedback module provides visual warnings, and the performance monitoring module evaluates the system performance. The invention realizes multimodal data fusion processing, generates structured diagnosis and treatment records, provides accurate diagnosis and treatment decision-making support, has real-time monitoring and system optimization functions, and improves the diagnosis and treatment level and efficiency of infectious shock.

Description

Technical Field

[0001] The present invention relates to the technical field of medical information processing, and particularly to a structured data processing system and method for infectious shock diagnosis and treatment image information. Background Art

[0002] As a severe clinical syndrome, infectious shock has a rapid disease progression and a high fatality rate. Timely and accurate diagnosis and treatment are crucial for improving the prognosis of patients. However, in the current diagnosis and treatment process of infectious shock, many problems in data processing and decision support are faced.

[0003] In terms of medical image data processing, the diagnosis of infectious shock patients often relies on various medical image data, such as CT images, ultrasound images, and hemodynamic time-series data. These data have different modalities and characteristics, and traditional processing methods are difficult to effectively integrate and analyze them. CT images can clearly show the morphological structure of organs, but cannot directly reflect the dynamic changes of hemodynamics; ultrasound images can observe the blood flow of some tissues and organs in real time, but the information is relatively limited; although hemodynamic time-series data can reflect the changes of cardiac function and circulatory status over time, it lacks intuitive anatomical structure information. Aligning these multi-modal data in space-time and normalizing their features to comprehensively and accurately analyze the patient's condition has become a major challenge currently.

[0004] From the perspective of clinical data management, existing diagnosis and treatment records are mostly in unstructured or semi-structured forms, containing a large amount of text descriptions and scattered data. When doctors view and analyze these data, they need to spend a lot of time and effort to screen and integrate key information, with low efficiency and prone to missing important details. For example, when recording the patient's physiological parameters, hemodynamic indicators, and imaging features, there is no unified standard data format, and the recording methods of different doctors vary greatly, which brings great inconvenience to subsequent data analysis and clinical decision-making. In addition, unstructured data is not conducive to the informatization management and big data analysis of medical data, unable to fully explore the potential value behind the data, and difficult to achieve precision medicine based on big data.

[0005] In terms of diagnosis and treatment decision support, the condition of infectious shock is complex and changeable, and each patient's situation is different, which poses a high requirement for doctors' diagnosis and treatment decision-making ability. Currently, doctors mainly rely on their own experience and limited clinical guidelines to formulate diagnosis and treatment plans, lacking personalized and precise decision support tools. Due to the lack of full consideration of individual differences of patients, traditional diagnosis and treatment plans may not achieve the best treatment effect and may even delay the condition. Moreover, when facing multiple possible diagnosis and treatment plans, how to quickly and accurately select the optimal plan and coordinate the resource conflicts and time-series conflicts between different plans is also an urgent problem to be solved in the clinical diagnosis and treatment process.

[0006] In addition, during the entire diagnosis and treatment process, there is a lack of an effective real-time monitoring and early warning mechanism. The subtle changes in the patient's condition cannot be detected in a timely manner, resulting in the inability to take corresponding intervention measures at the early stage of the deterioration of the condition. At the same time, the performance evaluation of the diagnosis and treatment system is not perfect enough, and the problems existing in aspects such as data processing and decision support of the system cannot be detected in a timely manner, making it difficult to optimize and improve the system targeted, which restricts the improvement of the diagnosis and treatment level of septic shock. To sum up, it is extremely urgent to develop a system and method that can effectively process the diagnostic and treatment image information of septic shock, realize structured data management, provide accurate diagnostic and treatment decision support, and have real-time monitoring and system optimization functions. Summary of the Invention

[0007] The purpose of the present invention is to provide a structured data processing system and method for diagnostic and treatment image information of septic shock to solve the problems mentioned in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solutions: A structured data processing system for diagnostic and treatment image information of septic shock, the system includes: a processor, an image acquisition module, a multi-modal data fusion module, a dynamic feature extraction module, a structured processing engine, a diagnostic and treatment decision module, an anomaly detection module, a warning feedback module, and a performance monitoring module;

[0009] The image acquisition module is used to obtain the medical image data of the patient, including CT images, ultrasound images, and hemodynamic time-series data; the multi-modal data fusion module performs spatio-temporal alignment and feature normalization processing on the medical image data to generate multi-modal fusion data; the dynamic feature extraction module performs time-series analysis and spatial feature extraction on the multi-modal fusion data through a preset deep learning model to generate dynamic feature vectors; the structured processing engine maps the dynamic feature vectors into a predefined clinical data template to generate a structured diagnostic record;

[0010] The anomaly detection module monitors the physiological parameters, hemodynamic indicators, and image features in the structured diagnostic record in real time. If it detects that the parameters deviate from the preset threshold range, it generates an anomaly signal and sends it to the processor; the diagnostic and treatment decision module calls a preset clinical pathway algorithm according to the anomaly signal to generate an adaptive diagnostic and treatment suggestion; the warning feedback module receives the anomaly signal and triggers a visual warning interface; the performance monitoring module records the response time of the anomaly signal and the number of corrections of the diagnostic and treatment decision to generate a system performance log.

[0011] Preferably, the specific operation process of the dynamic feature extraction module is as follows:

[0012] Perform time-dependent modeling on hemodynamic time-series data using a long short-term memory network to output time-series feature vectors; perform multi-scale feature extraction on CT images and ultrasound images respectively through a three-dimensional convolutional neural network to output spatial feature vectors; splice the time-series feature vectors and spatial feature vectors tensorially, and generate dynamic feature vectors after dimensionality reduction through a fully connected layer.

[0013] Preferably, the structured processing engine includes a template matching sub-module and a semantic parsing sub-module;

[0014] The template matching sub-module maps the physiological parameters in the dynamic feature vectors to cardiac function indicators, microcirculation status, and organ perfusion labels based on the data fields defined in the clinical guidelines; the semantic parsing sub-module performs entity recognition and relationship extraction on the unstructured text report through natural language processing technology to generate standardized semantic data;

[0015] The structured diagnosis and treatment record is generated by merging the standardized semantic data and the mapped data fields.

[0016] Preferably, the specific operation process of the anomaly detection module is as follows:

[0017] Set dynamic threshold intervals for cardiac function indicators, microcirculation status, and organ perfusion labels respectively, and the dynamic threshold intervals are adaptively adjusted based on the patient's historical data and population statistical data;

[0018] Calculate the deviation degree between the current parameter value and the dynamic threshold interval in real time. If the deviation degree of any parameter exceeds the preset tolerance coefficient, an anomaly signal is generated and the anomaly type is marked.

[0019] Preferably, the diagnosis and treatment decision-making module includes a path optimization sub-module and a conflict resolution sub-module;

[0020] The path optimization sub-module uses a reinforcement learning algorithm to generate a priority ranking of diagnosis and treatment plans according to the anomaly type and the patient's individual differences; the conflict resolution sub-module coordinates the resource conflicts and time-series conflicts between multiple plans through a fuzzy logic algorithm and outputs the final diagnosis and treatment recommendations.

[0021] Preferably, the specific analysis process of the performance monitoring module is as follows:

[0022] Record the response time from the generation of the anomaly signal to the output of the diagnosis and treatment recommendation. If the response time exceeds the preset response threshold, it is marked as a delayed event;

[0023] Count the number of times the diagnosis and treatment recommendation is clinically corrected. If the number of corrections exceeds the preset correction threshold, it is marked as an inefficient decision-making event;

[0024] Calculate the occurrence frequencies of the delayed event and the inefficient decision-making event weighted to generate a system performance score.

[0025] Preferably, the processor is communicatively connected to the data traceability module, and the data traceability module is configured to:

[0026] Label the source tags for each item of data in the structured diagnosis and treatment record, including the imaging device model, acquisition time, and operator information;

[0027] When the system performance score is lower than the preset qualified threshold, trigger data traceability analysis to locate the abnormal nodes of the data sources of latency events or inefficient decision-making events.

[0028] Preferably, the early warning feedback module includes multi-level early warning sub-modules;

[0029] The multi-level early warning sub-modules divide the early warning levels according to the severity of the abnormal signals. A level-1 early warning triggers a pop-up prompt on the interface, a level-2 early warning is synchronously sent to the mobile terminal, and a level-3 early warning directly connects to the emergency system.

[0030] Preferably, the system further includes a model iteration module, and the model iteration module is configured to:

[0031] Regularly collect system performance logs and clinical feedback data, update the deep learning model and reinforcement learning strategy through an online learning algorithm, and optimize the accuracy of dynamic feature extraction and diagnosis and treatment decision-making.

[0032] Preferably, the present invention further includes a method for processing structured data of infectious shock diagnosis and treatment image information, and the method includes the following steps:

[0033] Obtain multi-modal medical image data through an image acquisition module;

[0034] Perform spatio-temporal alignment and feature normalization processing on the multi-modal data to generate fused data;

[0035] Use a deep learning model to extract spatio-temporal dynamic features of the fused data to generate feature vectors;

[0036] Map the feature vectors to a clinical data template to generate a structured diagnosis and treatment record;

[0037] Real-time monitor the parameter deviation of the diagnosis and treatment record, and trigger abnormal signals and adaptive diagnosis and treatment suggestions;

[0038] Record system response events and generate performance logs, and locate abnormal nodes through data traceability;

[0039] Trigger a multi-level feedback mechanism according to the early warning level, and optimize the algorithm performance through model iteration.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] The present invention obtains various medical imaging data through an image acquisition module, and uses a multi-modal data fusion module to perform spatio-temporal alignment and feature normalization processing to generate multi-modal fusion data. This process realizes the complementary advantages of different modal data, enabling the data to more comprehensively reflect the patient's condition. For example, by combining the anatomical structure information of CT images, the real-time blood flow information of ultrasound images, and the dynamic change information of hemodynamic time-series data, it provides a rich and accurate data basis for subsequent analysis. The dynamic feature extraction module further performs time-series analysis and spatial feature extraction on the fusion data to generate dynamic feature vectors. This deep data mining method can more accurately capture the trend of disease changes. Compared with traditional single-data processing methods, it greatly improves the accuracy and effectiveness of data processing, and helps doctors understand the patient's condition more comprehensively and deeply.

[0042] The structured processing engine maps the dynamic feature vectors to a predefined clinical data template to generate a structured diagnosis and treatment record. This makes the originally complex and disordered medical data become standardized and orderly, facilitating doctors to quickly access and analyze. Doctors no longer need to spend a lot of time looking for key information in unstructured text records, and can directly obtain various physiological parameters, hemodynamic indicators, imaging features, and diagnostic results of patients from the structured diagnosis and treatment record, significantly improving the diagnosis and treatment efficiency. At the same time, structured data is more convenient for the informatization management and big data analysis of medical data, providing strong support for medical research and clinical decision-making, and is conducive to realizing precision medicine based on big data. Precise anomaly detection and timely warning: The anomaly detection module monitors various parameters in the structured diagnosis and treatment record in real time. Through a dynamically adjusted threshold range based on the patient's historical data and population statistical data, it can more accurately determine whether the parameters are abnormal. When it detects that the parameter deviates from the preset threshold range, it immediately generates an anomaly signal and marks the anomaly type. This function helps doctors detect subtle changes in the patient's condition in a timely manner. The warning feedback module divides multi-level warnings according to the severity of the anomaly signal, from interface pop-up prompts to connecting to the emergency system, ensuring that doctors can take corresponding intervention measures in the first time, effectively avoiding the deterioration of the condition and improving the success rate of patient treatment.

[0043] The diagnosis and treatment decision-making module calls the preset clinical pathway algorithm according to the anomaly signal and generates adaptive diagnosis and treatment suggestions in combination with the patient's individual differences. The path optimization sub-module uses a reinforcement learning algorithm to prioritize the diagnosis and treatment plans, fully considering the disease characteristics and individual differences of different patients, making the diagnosis and treatment plan more targeted. The conflict resolution sub-module coordinates the resource conflicts and timing conflicts between multiple plans through a fuzzy logic algorithm to ensure that the finally output diagnosis and treatment suggestions are practical. This personalized and precise diagnosis and treatment decision-making support method can help doctors formulate more scientific and reasonable diagnosis and treatment plans, improve the treatment effect, and improve the patient's prognosis.

[0044] The performance monitoring module records the response time of abnormal signals and the number of corrections to the diagnosis and treatment decisions, and generates a system performance log. Through the analysis of these data, the performance of the system can be comprehensively evaluated, and problems existing in the operation process of the system can be discovered in time. When the system performance score is lower than the preset qualified threshold, the data tracing module performs a tracing analysis on the data to locate the abnormal nodes of the data sources of delay events or inefficient decision-making events, providing a clear direction for system optimization. The model iteration module regularly collects system performance logs and clinical feedback data, updates the deep learning model and reinforcement learning strategy through an online learning algorithm, continuously optimizes the accuracy of dynamic feature extraction and diagnosis and treatment decisions, realizes the continuous optimization and self-improvement of the system, and ensures that the system always maintains an efficient and reliable operating state. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is the working principle diagram of the structured data processing system for infectious shock diagnosis and treatment image information of the present invention;

[0046] Figure 2 is the flowchart of conflict resolution and solution optimization for diagnosis and treatment decisions;

[0047] Figure 3 is the flowchart of system model iteration optimization;

[0048] Figure 4 is the flowchart of the method for processing infectious shock diagnosis and treatment data. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0050] Please refer to Figures 1-4 , the present invention provides a technical solution: a structured data processing system for infectious shock diagnosis and treatment image information, the system includes:

[0051] An image acquisition module: responsible for acquiring medical image data of patients, including CT images, ultrasound images, and hemodynamic time-series data. These data are collected from the hospital's imaging equipment and monitoring instruments, providing basic materials for subsequent analysis.

[0052] Multi-modal Data Fusion Module: Performs spatio-temporal alignment and feature normalization on the collected medical image data. Spatio-temporal alignment is to ensure the consistency of different modal data in time and space, and feature normalization is to unify the value ranges of different features for subsequent analysis and processing, and finally generates multi-modal fusion data.

[0053] Dynamic Feature Extraction Module: Performs time series analysis and spatial feature extraction on the multi-modal fusion data through a preset deep learning model. Time series analysis is used to mine the changing patterns of data over time, and spatial feature extraction focuses on the spatial information in the images, and finally generates dynamic feature vectors.

[0054] Structured Processing Engine: Maps the dynamic feature vectors into a predefined clinical data template to generate structured diagnosis and treatment records. This clinical data template is designed based on clinical guidelines and actual diagnosis and treatment needs, and can present complex medical data in a structured form for doctors to view and analyze conveniently.

[0055] Abnormality Detection Module: Monitors the physiological parameters, hemodynamic indicators and imaging features in the structured diagnosis and treatment records in real time. Once a parameter deviation from the preset threshold range is detected, an abnormality signal is generated and sent to the processor to promptly detect the patient's abnormal conditions.

[0056] Diagnosis and Treatment Decision-making Module: Invokes the preset clinical pathway algorithms according to the abnormality signals to generate adaptive diagnosis and treatment suggestions. These algorithms are formulated based on a large amount of clinical experience and research results and can provide scientific and reasonable diagnosis and treatment references for doctors.

[0057] Early Warning Feedback Module: Receives the abnormality signals and triggers a visual early warning interface. Through an intuitive visual way, doctors can quickly understand the patient's abnormal conditions and take corresponding measures in a timely manner.

[0058] Performance Monitoring Module: Records the response time of the abnormality signals and the number of corrections of the diagnosis and treatment decisions, and generates a system performance log. By recording and analyzing these data, the performance of the system can be evaluated to provide a basis for system optimization.

[0059] The present invention will be further described below in conjunction with Embodiments 1 to 5:

[0060] Embodiment 1:

[0061] The function of this embodiment is to elaborate in detail how the dynamic feature extraction module extracts features from different modal data and generates dynamic feature vectors, providing accurate data support for subsequent structured processing and diagnosis and treatment decision-making.

[0062] The operation process of the dynamic feature extraction module is as follows: The hemodynamic time-series data is modeled for time dependence using a Long Short-Term Memory network (LSTM). The hemodynamic time-series data contains important information such as the patient's cardiac function and vascular resistance that changes over time. The Long Short-Term Memory network can effectively capture the long-term dependence relationships in the data and overcome the problems of vanishing gradients and exploding gradients that are prone to occur in traditional neural networks when processing time-series data. Assuming the hemodynamic time-series data is the sequence , the input to the LSTM network is . Through a series of gating mechanisms, such as the forget gate , the input gate , and the output gate , the cell state is updated, and finally the time-series feature vector is output. Its core formula is:

[0063]

[0064] where is the Sigmoid function, represents element-wise multiplication, and and are the weight matrix and bias vector respectively.

[0065] Multi-scale feature extraction is performed on CT images and ultrasound images respectively through a three-dimensional convolutional neural network (3D-CNN). CT images and ultrasound images contain rich information about human organ structures and lesions. The three-dimensional convolutional neural network can perform convolutional operations on images in three-dimensional space to extract spatial features at different scales. Taking CT images as an example, after being processed by multiple convolutional layers and pooling layers, the spatial feature vector is finally output. Assuming the CT image is , after the operations of the convolutional layer and the pooling layer , the spatial feature vector is output.

[0066] The time-series feature vector and the spatial feature vector are concatenated tensorially, and a dynamic feature vector is generated after dimensionality reduction through a fully connected layer. Specifically, the time-series feature vector obtained from the hemodynamic time-series data and the spatial feature vectors , obtained from CT images and ultrasound images are concatenated tensorially to obtain a high-dimensional vector. Then, it is processed by a fully connected layer for dimensionality reduction to remove redundant information and generate the final dynamic feature vector .

[0067] Example 2:

[0068] The function of this embodiment is to elaborate in detail how the structured processing engine converts dynamic feature vectors into structured diagnosis and treatment records, making medical data more convenient for doctors to view and analyze, and improving the efficiency of diagnosis and treatment.

[0069] The structured processing engine includes a template matching sub-module and a semantic parsing sub-module.

[0070] Based on the data fields defined in the clinical guidelines, the template matching sub-module classifies and maps the physiological parameters in the dynamic feature vectors to the cardiac function index, microcirculation status, and organ perfusion labels. Clinical guidelines are the norms and standards summarized through long-term practice and research in the medical field, defining a series of data fields related to the diagnosis and treatment of septic shock. The template matching sub-module classifies the physiological parameters in the dynamic feature vectors according to these data fields. For example, the parameters reflecting the systolic and diastolic functions of the heart are mapped to the cardiac function index label, the parameters related to the microcirculation blood flow status are mapped to the microcirculation status label, and the parameters reflecting the organ blood perfusion are mapped to the organ perfusion label.

[0071] The semantic parsing sub-module performs entity recognition and relationship extraction on the unstructured text reports through natural language processing technology to generate standardized semantic data. In medical records, there are a large number of unstructured text reports, such as doctors' diagnostic descriptions, examination result explanations, etc. The semantic parsing sub-module uses natural language processing technology to analyze these texts. First, it performs entity recognition to identify the key entities in the text, such as disease names, symptoms, examination items, etc.; then it performs relationship extraction to determine the relationships between these entities, such as causal relationships, concomitant relationships, etc. Through these operations, the unstructured text is converted into standardized semantic data. For example, for the text "The patient has fever, hypotension, considering septic shock, and echocardiogram shows weakened left ventricular systolic function", the semantic parsing sub-module can identify entities such as "fever", "hypotension", "septic shock", "weakened left ventricular systolic function", etc., and extract the relationships between them to generate standardized semantic data.

[0072] The structured diagnosis and treatment record is generated by combining the standardized semantic data and the mapped data fields. The standardized semantic data generated by the semantic parsing sub-module is integrated with the mapped data fields of the template matching sub-module to form a complete structured diagnosis and treatment record. In this way, doctors can quickly obtain the patient's condition information from the structured diagnosis and treatment record, including various physiological parameters, diagnosis results, and the relationships between them.

[0073] Example 3:

[0074] The operation process of the anomaly detection module is as follows: Dynamically set threshold intervals for cardiac function indicators, microcirculation status, and organ perfusion labels respectively. These dynamic threshold intervals are not fixed, but are adaptively adjusted based on the patient's historical data and population statistical data. The patient's condition changes dynamically, and there are also individual differences among different patients. Therefore, fixed thresholds cannot accurately reflect the true situation of the patient. By collecting the patient's historical data, including cardiac function indicators, microcirculation status, and organ perfusion label data at different time points, and combining population statistical data, such as the average value and standard deviation of various indicators of patients with the same condition, to determine the dynamic threshold interval for each patient. For example, for the ejection fraction in the cardiac function indicators of a certain patient, a dynamic threshold interval that changes over time is determined based on its historical ejection fraction data and the ejection fraction statistical data of patients with the same type of septic shock.

[0075] Calculate the deviation degree between the current parameter value and the dynamic threshold interval in real time. If the deviation degree of any parameter exceeds the preset tolerance coefficient, an anomaly signal is generated and the anomaly type is marked. The calculation method of the deviation degree can adopt various ways, such as calculating the ratio of the difference between the current parameter value and the boundary value of the dynamic threshold interval to the width of the threshold interval. Assume that the current parameter value is ,the dynamic threshold interval is ,the preset tolerance coefficient is ,the deviation degree calculation formula is:

[0076]

[0077] When the deviation degree is greater than ,it indicates that the current parameter is abnormal. The anomaly detection module generates an anomaly signal and marks the anomaly type according to the label category to which the parameter belongs, such as "abnormal cardiac function indicator", "abnormal microcirculation status", or "abnormal organ perfusion", etc.

[0078] Example 4:

[0079] The function of this example is to elaborate in detail how the diagnosis and treatment decision-making module generates a reasonable diagnosis and treatment plan according to the anomaly type and patient individual differences, and coordinates the conflicts among multiple plans, provides the final diagnosis and treatment suggestions for doctors, and assists doctors in making scientific diagnosis and treatment decisions.

[0080] The diagnosis and treatment decision-making module includes a path optimization sub-module and a conflict resolution sub-module.

[0081] The path optimization sub-module uses a reinforcement learning algorithm to generate a priority ranking of treatment plans according to the type of abnormality and the individual differences of patients. Reinforcement learning is a method of learning the optimal strategy by an agent interacting with the environment and obtaining rewards. In the present invention, the path optimization sub-module takes the type of abnormality and the individual differences of patients as inputs, explores different treatment plans through the reinforcement learning algorithm, and obtains rewards according to the effects of these plans on the improvement of the patient's condition. For example, for patients with different types of abnormalities (such as cardiac function abnormalities, microcirculation abnormalities, etc.) and different physical conditions (age, underlying diseases, etc.), different drug treatment plans, fluid resuscitation plans, etc. are tried, and corresponding rewards are given according to the changes in the patient's condition indicators (such as blood pressure recovery, improvement of organ perfusion, etc.). Through continuous learning and optimization, a priority ranking of treatment plans for different situations is generated to provide reference for doctors.

[0082] The conflict resolution sub-module coordinates the resource conflicts and timing conflicts between multiple plans through a fuzzy logic algorithm and outputs the final treatment advice. In the actual treatment process, there may be situations where multiple treatment plans are applicable at the same time but there are resource conflicts (such as insufficient drug supply, limited medical equipment, etc.) or timing conflicts (such as contradictions in the implementation order of different treatment measures). The fuzzy logic algorithm can handle fuzzy and uncertain information, fuzzify the conflict factors of different plans, and make inferences and decisions according to expert experience and preset rules. For example, for the conflicts between drug treatment plans and surgical treatment plans in terms of resources and time, the fuzzy logic algorithm comprehensively considers various factors, such as the urgency of the patient's condition, the availability of resources, etc., to determine the final treatment advice and ensure the feasibility and effectiveness of the treatment plan.

[0083] Example 5:

[0084] The analysis process of the performance monitoring module is as follows: Record the response time from the generation of the abnormal signal to the output of the treatment advice. If the response time exceeds the preset response threshold, it is marked as a delayed event. The preset response threshold is set according to clinical actual needs and experience. Generally speaking, patients with septic shock are in critical condition and rapid treatment decisions need to be made. If the system takes too long to give treatment advice after detecting an abnormal signal, it may affect the treatment effect of the patient. For example, if the preset response threshold is set to 5 minutes, when the abnormal signal is generated and the system takes more than 5 minutes to output the treatment advice, this event is marked as a delayed event.

[0085] The number of times the diagnosis and treatment recommendations are clinically revised is counted. If the number of revisions exceeds the preset revision threshold, it is marked as an inefficient decision-making event. The diagnosis and treatment recommendations are generated by the system based on the algorithm, but in actual clinical applications, doctors may revise the diagnosis and treatment recommendations based on their own experience and further observation of the patient. If the number of revisions is too many, it means that the diagnosis and treatment recommendations generated by the system may be inaccurate or unreasonable. The preset revision threshold can be adjusted according to the actual situation and clinical data of different hospitals. For example, after statistical analysis, in a certain hospital, if the number of revisions of the diagnosis and treatment recommendations exceeds 10 times within a month, it is marked as an inefficient decision-making event.

[0086] The frequency of delay events and inefficient decision-making events is weighted to generate a system performance score. Different types of events have different impacts on system performance, so it is necessary to weight the frequency of delay events and inefficient decision-making events. For example, if the weight of delay events is set to 0.6 and the weight of inefficient decision-making events is set to 0.4, the system performance score calculation formula is:

[0087]

[0088] The processor is communicatively connected to the data traceability module, which is used to label the source of each data item in the structured diagnosis and treatment record, including the imaging device model, acquisition time, and operator information. When the system performance score is lower than the preset qualified threshold, the data traceability analysis is triggered to locate the abnormal node of the data source of the delayed event or inefficient decision-making event. By marking the data source label, the data collection process and related information can be traced. When there is a problem with system performance, the data traceability module can help find the root cause of the problem. For example, if it is found that a delay event is caused by a specific imaging device collecting data too slowly, or an inefficient decision-making event is caused by an operator entering data incorrectly, improvements can be made to these problems.

[0089] The early warning feedback module includes a multi-level early warning sub-module. The multi-level early warning sub-module divides the early warning level according to the severity of the abnormal signal. The first-level early warning triggers a pop-up window prompt on the interface, the second-level early warning is sent to the mobile terminal simultaneously, and the third-level early warning is directly connected to the emergency system. The severity of the abnormal signal can be evaluated based on factors such as the size of the deviation and the importance of the abnormal parameters. The first-level early warning is suitable for some relatively minor abnormal situations, and prompts the doctor through a pop-up window on the system interface; the second-level early warning is for more serious abnormalities. In addition to the prompt on the interface, the early warning information will also be sent to the doctor's mobile terminal to ensure that the doctor can know it in time; the third-level early warning is for emergency situations that endanger the patient's life, directly connecting to the emergency system and starting the emergency rescue procedure.

[0090] The system also includes a model iteration module, which is used to regularly collect system performance logs and clinical feedback data, update the deep learning model and reinforcement learning strategy through online learning algorithms, and optimize the accuracy of dynamic feature extraction and diagnosis and treatment decisions. As clinical data accumulates continuously and the system runs, it is necessary to continuously optimize the model and algorithm to improve the performance of the system. The model iteration module updates the deep learning model and reinforcement learning strategy by collecting data in the system performance logs, such as information on latency events, inefficient decision-making events, etc., and clinical feedback data, such as doctors' evaluations of diagnosis and treatment suggestions, patients' treatment effects, etc., and using online learning algorithms. For example, online learning algorithms such as stochastic gradient descent are used to adjust the parameters of the deep learning model so that the model can extract dynamic features more accurately; the reinforcement learning strategy is optimized to improve the accuracy and rationality of diagnosis and treatment decisions.

[0091] The present invention also includes a method for processing structured data of infectious shock diagnosis and treatment image information, and the method includes:

[0092] Obtain multi-modal medical image data through an image acquisition module;

[0093] Perform spatio-temporal alignment and feature normalization processing on the multi-modal data to generate fused data;

[0094] Use a deep learning model to extract spatio-temporal dynamic features of the fused data to generate feature vectors;

[0095] Map the feature vectors to a clinical data template to generate a structured diagnosis and treatment record;

[0096] Real-time monitor the parameter deviation of the diagnosis and treatment record, and trigger abnormal signals and adaptive diagnosis and treatment suggestions;

[0097] Record system response events and generate performance logs, and locate abnormal nodes through data traceability;

[0098] Trigger a multi-level feedback mechanism according to the warning level, and optimize the algorithm performance through model iteration.

[0099] The implementation manner of this method refers to the above-mentioned embodiments and will not be elaborated in the specification.

[0100] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0101] Although embodiments of the present invention have been shown and described, those of ordinary skill in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A septic shock diagnosis and treatment image information structured data processing system, characterized in that: It includes a processor, an image acquisition module, a multimodal data fusion module, a dynamic feature extraction module, a structured processing engine, a diagnosis and treatment decision module, an anomaly detection module, a warning feedback module and a performance monitoring module; The image acquisition module is used to obtain the patient's medical imaging data, including CT images, ultrasound images and hemodynamic time series data; the multimodal data fusion module performs time-space alignment and feature normalization processing on the medical imaging data to generate multimodal fusion data; the dynamic feature extraction module performs time series analysis and spatial feature extraction on the multimodal fusion data through a preset deep learning model to generate a dynamic feature vector; the structured processing engine maps the dynamic feature vector to a predefined clinical data template to generate a structured diagnosis and treatment record; The abnormality detection module monitors the physiological parameters, hemodynamic indicators and imaging features in the structured diagnosis and treatment records in real time, and generates an abnormality signal and sends it to the processor if it detects that the parameters deviate from the preset threshold range; The diagnosis and treatment decision module calls a preset clinical pathway algorithm according to the abnormal signal to generate adaptive diagnosis and treatment suggestions; The early warning feedback module receives the abnormal signal and triggers the visual early warning interface; the performance monitoring module records the response time of the abnormal signal and the number of revisions of the diagnosis and treatment decision, and generates a system performance log; The specific operation process of the dynamic feature extraction module is as follows: Long short-term memory network is used to model the time dependency of hemodynamic time series data, and the time series feature vector is output; multi-scale feature extraction is performed on CT images and ultrasound images through three-dimensional convolutional neural network, and spatial feature vector is output; the time series feature vector and the spatial feature vector are tensor spliced, and the dynamic feature vector is generated after dimensionality reduction through the fully connected layer; The structured processing engine includes a template matching submodule and a semantic parsing submodule; The template matching submodule maps the physiological parameters in the dynamic feature vector to cardiac function indicators, microcirculation status and organ perfusion labels based on the data fields defined in the clinical guidelines; the semantic parsing submodule performs entity recognition and relationship extraction on the unstructured text report through natural language processing technology to generate standardized semantic data; The structured medical record is generated by merging the standardized semantic data with the mapped data fields; The specific analysis process of the performance monitoring module is as follows: Record the response time from the generation of abnormal signals to the output of diagnosis and treatment recommendations. If the response time exceeds the preset response threshold, it is marked as a delay event. Count the number of times the diagnosis and treatment recommendations are clinically revised. If the number of revisions exceeds the preset revision threshold, it will be marked as an inefficient decision-making event. The frequencies of delay events and inefficient decision-making events are weighted and calculated to generate a system performance score.

2. The septic shock diagnosis and treatment image information structured data processing system according to claim 1, characterized in that: The specific operation process of the anomaly detection module is as follows: Dynamic threshold intervals are set for cardiac function indicators, microcirculation status and organ perfusion labels respectively, and the dynamic threshold intervals are adaptively adjusted based on patient historical data and group statistical data; The deviation between the current parameter value and the dynamic threshold interval is calculated in real time. If the deviation of any parameter exceeds the preset tolerance coefficient, an abnormal signal is generated and the abnormal type is marked.

3. The septic shock diagnosis and treatment image information structured data processing system according to claim 2, characterized in that: The diagnosis and treatment decision module includes a path optimization submodule and a conflict resolution submodule; The path optimization submodule adopts a reinforcement learning algorithm to generate a priority ranking of diagnosis and treatment plans according to the abnormality type and individual differences of patients; the conflict resolution submodule coordinates the resource conflicts and timing conflicts among multiple plans through a fuzzy logic algorithm and outputs the final diagnosis and treatment recommendations.

4. The septic shock diagnosis and treatment image information structured data processing system according to claim 3, characterized in that: The processor is in communication with a data tracing module, and the data tracing module is used to: Label each data item in the structured medical record with its source, including imaging equipment model, acquisition time, and operator information; When the system performance score is lower than the preset qualified threshold, data traceability analysis is triggered to locate abnormal nodes of the data source of delayed events or inefficient decision-making events.

5. The septic shock diagnosis and treatment image information structured data processing system according to claim 4, characterized in that: The early warning feedback module includes a multi-level early warning submodule; The multi-level warning submodule divides the warning level according to the severity of the abnormal signal. The first-level warning triggers a pop-up window prompt on the interface, the second-level warning is sent to the mobile terminal simultaneously, and the third-level warning is directly connected to the emergency system.

6. The septic shock diagnosis and treatment image information structured data processing system according to claim 5, characterized in that: The system further comprises a model iteration module, wherein the model iteration module is configured to: Regularly collect system performance logs and clinical feedback data, update deep learning models and reinforcement learning strategies through online learning algorithms, and optimize the accuracy of dynamic feature extraction and diagnosis and treatment decisions.

7. A method for processing structured data of septic shock diagnosis and treatment image information based on the system for processing structured data of septic shock diagnosis and treatment image information according to any one of claims 1 to 6, characterized in that: The following steps are involved: Acquire multimodal medical imaging data through an image acquisition module; Perform spatiotemporal alignment and feature normalization on multimodal data to generate fused data; Use deep learning models to extract spatiotemporal dynamic features of fused data and generate feature vectors; Map feature vectors to clinical data templates to generate structured diagnosis and treatment records; Real-time monitoring of parameter deviations in diagnosis and treatment records, triggering abnormal signals and adaptive diagnosis and treatment recommendations; Record system response events and generate performance logs, and locate abnormal nodes through data tracing; A multi-level feedback mechanism is triggered according to the warning level, and the algorithm performance is optimized through model iteration.

Images