AI and big data-based early warning system for acute and critical diseases of digestive system

Abstract

The invention relates to the field of medical data processing, and discloses an AI and big data-based early warning system for acute and critical diseases of a digestive system. Comprising the steps that a multi-modal data acquisition and alignment module synchronizes image and sign data, tissue biomechanical characteristics are quantified through a focus topological stiffness extraction module, a corrected blood flow transfer function is generated in combination with a blood flow transfer function reconstruction module, a visceral perfusion waveform is deduced through a visceral blood flow reverse mapping module, and a visceral perfusion result is obtained. The intervention response closed-loop analysis module is used for evaluating the physiological response difference, and the output module is used for issuing early warning according to the resistance and the coupling loss index. The method comprises the following steps: dynamically correcting a fluid dynamic model by extracting a lesion topological stiffness characteristic in a medical image, constructing a patient-specific peripheral visceral blood flow transfer function, reversely mapping peripheral signs into a visceral microcirculation perfusion waveform by using the function, and analyzing and calculating a structural function decoupling index in combination with an intervention response, so as to obtain a visceral microcirculation perfusion result. And non-invasive and early warning of organ ischemia risk and functional failure is realized.

Description

Technical Field

[0001] This invention relates to the field of medical data processing, specifically to an early warning system for acute and critical illnesses of the digestive system based on AI and big data. Background Technology

[0002] Severe acute pancreatitis, decompensated cirrhosis, acute gastrointestinal bleeding, and sepsis are common acute and critical illnesses of the digestive system, often accompanied by severe hemodynamic disturbances. In the clinical monitoring of these diseases, real-time assessment of visceral microcirculation perfusion is crucial for halting the progression of multiple organ failure.

[0003] Currently, clinical monitoring of patients' hemodynamic status mainly relies on peripheral vital signs (such as heart rate and mean arterial pressure) or systemic indicators such as central venous pressure. However, the human body has compensatory mechanisms in the early stages of shock or under stress, maintaining the perfusion pressure of vital organs such as the heart and brain by constricting the visceral vascular bed. As a result, systemic indicators often remain within the normal range in the early stages of critical illness, while severe occult hypoperfusion has already occurred in the digestive system. Existing monitoring methods, if they want to directly obtain visceral perfusion data, usually require invasive catheter monitoring techniques or intermittent imaging examinations. Invasive procedures carry risks of infection and bleeding and are difficult to maintain long-term; intermittent imaging examinations cannot achieve continuous time-axis monitoring and are difficult to capture turning points in the disease's progression. Therefore, current technology lacks an effective means to non-invasively and continuously infer the visceral microcirculation status using easily obtainable peripheral vital sign data, making it difficult for clinicians to identify organ ischemia risks in a timely manner during the compensatory phase.

[0004] Furthermore, with the development of computational medicine, estimating hemodynamic status based on physiological parameters using mathematical models has become a trend. However, existing hemodynamic monitoring models are mostly constructed based on general human parameters, assuming that the vascular bed has standardized compliance and resistance characteristics. In actual clinical scenarios, patients with acute and critical digestive system diseases often have complex anatomical changes, such as arteriosclerosis caused by cirrhosis, tissue edema caused by inflammatory exudation, or compression by space-occupying lesions. These pathological changes affect the physical and mechanical properties of the vascular bed, causing the patient's physiological response characteristics to deviate from the standard population model. Existing monitoring systems usually process physiological monitoring data separately from anatomical imaging data, failing to utilize patient-specific anatomical features (such as tissue texture and density) to correct the structural parameters of the fluid dynamics model. This results in significant deviations in the model's calculation results when facing patients with underlying structural lesions, reducing the accuracy of the assessment.

[0005] Finally, during intensive care, medical staff often use interventions such as fluid resuscitation or vasoactive drugs to correct shock and observe the patient's physiological responses. Existing monitoring equipment can only display numerical changes after intervention (such as whether blood pressure has increased), but it is difficult to deeply analyze the physiological system's response mechanism to intervention. In particular, when patients do not respond to treatment or respond weakly, existing analytical techniques cannot quantitatively distinguish whether this lack of response is due to structural rigidity of blood vessels (such as irreversible resistance caused by fibrosis) or functional dysregulation (such as severe vasospasm). The lack of quantitative analysis tools for the structural-functional coupling relationship makes it difficult for clinicians to clarify the pathophysiological causes when facing complex refractory shock, thus affecting the precise adjustment of treatment plans. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an early warning system for acute and critical illnesses of the digestive system based on AI and big data. This system solves the problem that existing technologies lack quantitative means to quantify the nonlinear mapping relationship between peripheral signs and visceral microcirculation, which makes it impossible to identify functional ischemia of organs in a timely manner during the compensatory period.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an early warning system for acute and critical illnesses of the digestive system based on AI and big data, comprising: The multimodal data acquisition and alignment module is configured to acquire medical image data, high-frequency time-series vital sign data, and medical intervention event data, and based on the timestamp of the medical intervention event data, perform time axis synchronization and correlation operations to generate multimodal input vectors. The lesion topology stiffness extraction module is configured to receive the medical image data, segment the anatomical regions of the digestive system organs and identify the lesion regions, and calculate the equivalent stiffness matrix reflecting the biomechanical characteristics of the digestive system vascular bed based on the texture density features and geometric morphology features of the region of interest, and use the equivalent stiffness matrix to aggregate and generate global equivalent topology stiffness coefficients. The blood flow transfer function reconstruction module is configured to receive the global equivalent topological stiffness coefficient and use the global equivalent topological stiffness coefficient to dynamically correct the structural parameters of the preset basic fluid dynamics model to generate a patient-specific peripheral visceral blood flow transfer function. The visceral blood flow inverse mapping module is configured to receive the high-frequency time-series vital sign data and the patient-specific peripheral visceral blood flow transfer function, deduce the perfusion waveform of visceral microcirculation through frequency domain inverse operation, and calculate the visceral vascular resistance index. The intervention response closed-loop analysis module is configured to monitor the medical intervention event data. When an excitation event is detected, the expected physiological response is calculated using the patient-specific peripheral visceral blood flow transfer function, and the structural-functional decoupling index is calculated in combination with the actual observed physiological response. The output control module is configured to generate graded early warning signals based on the visceral vascular resistance index and the structural-functional decoupling index, and output the attribution analysis results through an interactive terminal.

[0008] Preferably, the time axis synchronization and association operations performed by the multimodal data acquisition and alignment module specifically include: based on the current moment, backtracking a time window of a set length to extract high-frequency time-series vital sign data, while retrieving the medical image data that is closest to the current moment and within the validity period, and mapping the medical intervention event data to the corresponding time axis position to ensure the correspondence between physiological parameters and anatomical parameters in time sequence logic.

[0009] Preferably, the specific method for calculating the global equivalent topological stiffness coefficient by the lesion topological stiffness extraction module is as follows: Medical image data is divided into a set of regions of interest (ROIs) including the liver, pancreas, spleen, and mesenteric vascular network using a semantic segmentation algorithm; local texture feature vectors of each voxel point within each ROI are extracted, and a pre-defined tissue elastoplastic mapping function is used to convert these local texture feature vectors into estimated local stiffness values; based on the contribution weight of each ROI to blood flow resistance, the estimated local stiffness values ​​within each ROI are integrally and weightedly summed to obtain the global equivalent topological stiffness coefficient; thereby physicalizing the pathological changes of organs into impedance parameters.

[0010] Preferably, the pre-set basic hydrodynamic model in the blood flow transfer function reconstruction module is a second-order linear time-invariant system. The structural parameters of the basic hydrodynamic model include the basic natural frequency and the basic damping ratio. The specific process of dynamic correction includes: using the global equivalent topological stiffness coefficient, calculating the reconstructed damping ratio based on the logarithmic growth law, and calculating the reconstructed natural frequency based on the exponential decay law; the blood flow transfer function reconstruction module substitutes the reconstructed damping ratio and the reconstructed natural frequency into the basic hydrodynamic model to replace the basic damping ratio and the basic natural frequency, thereby constructing a patient-specific peripheral visceral blood flow transfer function. This step realizes the transformation from a general physiological model to an individual pathological model.

[0011] Preferably, the process of the visceral blood flow inverse mapping module to deduce the perfusion waveform of visceral microcirculation includes: applying a window function to high-frequency time-series vital sign data to eliminate spectral leakage, followed by a Fast Fourier Transform to obtain the input spectrum; replacing the complex frequency domain variables in the patient-specific peripheral visceral blood flow transfer function with frequency domain variables to obtain the system frequency response function; calculating the product of the input spectrum and the system frequency response function to obtain the frequency domain estimate of the visceral perfusion waveform; and performing an inverse Fast Fourier Transform on the frequency domain estimate of the visceral perfusion waveform to reconstruct the perfusion waveform of visceral microcirculation in the time domain. This process utilizes frequency domain deconvolution technology to achieve non-invasive deduction of deep blood flow status.

[0012] Preferably, the visceral blood flow reverse mapping module calculates the visceral vascular resistance index as follows: based on the equivalent principle of Ohm's law in fluid mechanics, the average systemic mean arterial pressure or equivalent perfusion pressure within the calculation window is calculated based on high-frequency time-series vital sign data as the driving pressure potential energy; the perfusion waveform of the reconstructed visceral microcirculation in the time domain is integrated to obtain the effective visceral perfusion intensity within the calculation window as the effective flow flux; the ratio of the driving pressure potential energy to the effective flow flux is calculated, and the ratio is multiplied by a preset unit conversion coefficient to obtain the visceral vascular resistance index; this index is used to quantify the contraction state of the visceral vascular bed.

[0013] Preferably, the stimulus events identified by the intervention response closed-loop analysis module include fluid resuscitation events and vasoactive drug infusion events; the intervention response closed-loop analysis module marks the stimulus events as test input signals and extracts physiological data within a preset time window after the event occurs as the actual observed physiological response; the expected physiological response is calculated by multiplying the frequency domain transformation of the test input signal with the patient-specific peripheral visceral blood flow transfer function.

[0014] Preferably, the specific method for calculating the structural-functional miscoupling index in the intervention response closed-loop analysis module is as follows: calculate the frequency domain phase angle of the actually observed physiological response and the frequency domain phase angle of the expected physiological response respectively; calculate the phase difference spectrum between the frequency domain phase angle of the actually observed physiological response and the frequency domain phase angle of the expected physiological response, and use the spectral energy distribution of the test input signal as a weighting function to perform integral calculation on the phase difference spectrum within the analysis frequency band covering heart rate and its harmonic components to obtain the structural-functional miscoupling index; this index is used to characterize the degree of deviation of the physiological functional response from the anatomical structural basis.

[0015] Preferably, the output control module is preset with a resistance threshold and a decoupling threshold; the generation logic of the graded warning signal includes: when the visceral vascular resistance index exceeds the resistance threshold and the structural-functional decoupling index is lower than the decoupling threshold, a first-level warning signal indicating the risk of tissue ischemia is generated; when the structural-functional decoupling index exceeds the decoupling threshold, a second-level warning signal indicating functional failure or microcirculatory disorder is generated.

[0016] Preferably, the attribution analysis results output by the output control module through the interactive terminal include: for the first-level warning signal, prompting microcirculation resuscitation intervention; for the second-level warning signal, indicating that the physiological mechanism leading to the warning is attributed to increased resistance due to structural rigidity or response mismatch due to functional dysfunction.

[0017] This invention provides an early warning system for acute and critical illnesses of the digestive system based on AI and big data. It has the following beneficial effects: 1. This invention constructs a patient-specific peripheral visceral blood flow transfer function and uses frequency domain inverse operation to map easily obtainable clinical peripheral high-frequency signs data into perfusion waveforms of deep organs; it solves the problem of delayed detection of organ ischemia due to reliance on invasive catheter monitoring or only relying on systemic indicators, and can identify the hidden low perfusion risk of the digestive system in advance through the abnormal increase of visceral vascular resistance index during the systemic compensation period.

[0018] 2. This invention utilizes a lesion topological stiffness extraction module to quantify anatomical features such as the degree of tissue fibrosis and vascular bed density reflected in medical images into physical stiffness parameters, and uses these parameters to dynamically correct the damping ratio and natural frequency of the fluid dynamics model; ensuring that the model calculation basis conforms to the patient's current pathophysiological structure, and avoiding calculation deviations caused by using general model parameters in the monitoring of patients with specific lesions.

[0019] 3. This invention treats routine clinical medical interventions as systemic stimulus signals and calculates the structural-functional decoupling index by comparing the expected response based on physical structure with the actual observed physiological response. The index can quantify the degree of impairment of organ regulatory function and help clinicians distinguish whether the increase in vascular resistance is due to irreversible structural rigidity or reversible functional vasoconstriction, thereby providing a clear pathophysiological attribution basis for the formulation of subsequent treatment plans. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the system functional framework of the present invention; Figure 2 This is a schematic diagram of the system operation process of the present invention.

[0021] The module includes: 100, Multimodal Data Acquisition and Alignment Module; 200, Lesion Topological Stiffness Extraction Module; 300, Blood Flow Transfer Function Reconstruction Module; 400, Visceral Blood Flow Inverse Mapping Module; 500, Intervention Response Closed-Loop Analysis Module; and 600, Output Control Module. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see the appendix Figure 1 This invention provides an AI and big data-based early warning system for acute and critical illnesses of the digestive system. The early warning system is connected to an external data source and includes a data processing server, a storage unit, and an interactive terminal. The data processing server is connected to the hospital's internal medical image storage and transmission system (PACS), hospital information system (HIS), and bedside monitoring network through a communication interface.

[0024] The early warning system is configured to execute computer-readable instructions to monitor critical illnesses in the occult compensatory stage of the digestive system. Functionally, the early warning system includes: a multimodal data acquisition and alignment module 100, a lesion topology stiffness extraction module 200, a blood flow transfer function reconstruction module 300, a visceral blood flow reverse mapping module 400, and an intervention response closed-loop analysis module 500.

[0025] The multimodal data acquisition and alignment module 100 connects to an external medical data source and is configured to acquire three types of heterogeneous data from the target patient. The first type of data is medical imaging data, including abdominal computed tomography (CT) or magnetic resonance imaging (MRI) data; the second type of data is high-frequency time-series vital sign data, including photoplethysmography (PPG), arterial blood pressure waveforms, and electrocardiogram data; and the third type of data is medical intervention event data, including drug infusion records and fluid resuscitation records.

[0026] The multimodal data acquisition and alignment module 100 is also configured to perform timeline synchronization operations. This operation uses the timestamps of medical intervention event data as a reference, extracts high-frequency time-series vital sign data within the corresponding time window, and associates it with the medical image data generated from the most recent scan, integrating the above data into a standardized multimodal input vector.

[0027] The lesion topology stiffness extraction module 200 is connected to the multimodal data acquisition and alignment module 100 and is configured to receive medical image data. The lesion topology stiffness extraction module 200 performs semantic segmentation on the image data and identifies key anatomical regions such as the liver, pancreas, and mesenteric vessels, as well as their lesion areas.

[0028] The lesion topology stiffness extraction module 200 is further configured to calculate the texture density features and geometric morphology features of the region of interest, and based on a preset tissue elastoplastic mapping relationship, generates an equivalent stiffness matrix reflecting the physical properties of the digestive system vascular bed. This matrix quantifies the biomechanical properties of each organ under the current pathological state.

[0029] The blood flow transfer function reconstruction module 300 is connected to the lesion topology stiffness extraction module 200. The blood flow transfer function reconstruction module 300 contains a pre-built basic fluid dynamics model describing the pulse wave transmission process from the peripheral circulation to the visceral microcirculation.

[0030] The blood flow transfer function reconstruction module 300 is configured to receive the equivalent stiffness matrix and use this matrix to correct the structural parameters of the basic hydrodynamic model. The correction process includes adjusting the damping coefficient and natural frequency of the transfer function based on the stiffness data, thereby generating a patient-specific peripheral visceral blood flow transfer function.

[0031] The visceral blood flow reverse mapping module 400 is connected to the multimodal data acquisition and alignment module 100 and the blood flow transfer function reconstruction module 300, respectively. The visceral blood flow reverse mapping module 400 is configured to receive real-time high-frequency time-series vital sign data and the reconstructed peripheral visceral blood flow transfer function.

[0032] The visceral blood flow reverse mapping module 400 uses frequency domain deconvolution or reverse filtering algorithms to deduce the perfusion waveform of visceral microcirculation from the peripheral monitoring waveform, and calculates the visceral vascular resistance index accordingly, which serves as the first quantitative indicator characterizing the visceral ischemia state.

[0033] The intervention response closed-loop analysis module 500 is connected to the above modules and is configured to monitor the input of medical intervention event data. When a specific medical intervention is detected, the intervention response closed-loop analysis module 500 defines it as an excitation signal for the system.

[0034] The intervention response closed-loop analysis module 500 is configured to calculate the expected physiological response based on the peripheral visceral blood flow transfer function and simultaneously acquire the patient's actual physiological response. By calculating the phase difference spectrum between the expected response and the actual response in the frequency domain, the intervention response closed-loop analysis module 500 generates a structure-function decoupling index.

[0035] The early warning system also includes an output control module 600, which is configured to generate a graded early warning signal and display the specific attribution analysis results through an interactive terminal when the visceral vascular resistance index or the structural-functional decoupling index exceeds a preset threshold.

[0036] See attached document Figure 2 In the data acquisition and spatiotemporal alignment phase, the system collects three types of basic data in parallel from the hospital information system and medical equipment through standardized data interfaces. Medical imaging data includes DICOM format files from abdominal computed tomography (CT) or magnetic resonance imaging (MRI) scans, providing static anatomical information of the patient's digestive system organs. High-frequency time-series vital signs data are input as a continuous waveform stream, with the sampling frequency set to a value that meets waveform feature extraction requirements (e.g., 125Hz or higher), primarily including photoplethysmography (PPG) and arterial blood pressure (ABP) data. Medical intervention event data includes medication order execution records and fluid intake / output records with precise timestamps.

[0037] For the spatiotemporal alignment of the aforementioned heterogeneous data, the system performs a correlation operation based on a time window. The system uses the current time... Based on this, a time window of a set length is traced backward to extract high-frequency time-series vital sign data segments, while simultaneously retrieving data from a distance of time. The system retrieves the most recent and valid medical imaging data (e.g., within 24 hours). For medical intervention event data, the system maps it to the corresponding timeline position, forming a comprehensive state vector that includes anatomical structures, physiological waveforms, and external disturbance events. The parsing of medical imaging data and the reading of DICOM standard files are image processing techniques well-known to those skilled in the art and will not be elaborated upon here.

[0038] In the anatomical feature-based physical parameter extraction stage, an equivalent stiffness matrix reflecting the biomechanical characteristics of the digestive tract vascular bed is constructed using medical imaging data. Semantic segmentation is performed on the liver, pancreas, spleen, and mesenteric regions in the imaging data to identify the texture features and geometric morphology of the lesion areas. Based on a pre-defined tissue elastoplastic mapping relationship, the gray-level gradient distribution in the images is transformed into a stiffness distribution in physical space. Pathological changes in major digestive organs (such as vascular tortuosity caused by cirrhosis, and tissue edema or necrosis caused by severe pancreatitis) are quantified into global equivalent topological stiffness coefficients. It characterizes the impedance characteristics of the digestive system vascular bed to blood flow pulsation and is the physical interface connecting anatomical structure and hemodynamic function.

[0039] During the dynamic reconstruction phase of the hemodynamic model, a pre-set basic fluid dynamics model is invoked, and the globally equivalent topological stiffness coefficients obtained from the aforementioned calculations are used. The structural parameters of the model are corrected in real time. The basic model is a second-order linear time-invariant system describing the transmission of pulse waves from the central circulation to the visceral microcirculation. The damping ratio of the transfer function is dynamically adjusted according to the patient's current anatomical and pathological state. and natural frequency The modified peripheral visceral blood flow transfer function In the Laplace domain, it is defined as: ; In the formula, For the complex frequency domain variables in the Laplace transform, For the reconstructed peripheral visceral blood flow transfer function, The global equivalent topological stiffness coefficient. The natural frequency function is calculated based on the global equivalent topological stiffness coefficient. This is the damping function calculated based on the global equivalent topological stiffness coefficient. Through this step, the system transforms the general physiological model into a specialized physical model adapted to the current pathological structure of a specific patient.

[0040] In the inverse estimation phase of visceral perfusion status, the system utilizes the reconstructed peripheral visceral blood flow transfer function. Peripheral monitoring signals are inversely processed to infer the blood flow state of visceral microcirculation. A Fast Fourier Transform is performed on the real-time acquired peripheral photoplethysmography (FPT) signals to obtain their frequency domain characteristics, which are then combined with the transfer function. The frequency domain response of the visceral perfusion waveform is calculated. Subsequently, the system reconstructs the visceral perfusion waveform in the time domain using inverse Fourier transform, and calculates the visceral vascular resistance index accordingly. This index reflects the actual degree of contraction or spasm of the visceral vascular bed after excluding the effects of systemic blood pressure fluctuations, and is used to identify whether the body is in an early compensatory state of sacrificing visceral blood supply.

[0041] During the closed-loop validation phase based on intervention response, the system continuously monitors the input of medical intervention event data. When an event with a physiological stimulus effect, such as rapid fluid resuscitation or vasoactive drug infusion, is detected, the system initiates closed-loop analysis logic, treating the intervention event as a step or pulse input to the physiological system, and utilizing the currently determined peripheral visceral blood flow transfer function. The theoretically expected physiological response is calculated. Simultaneously, actual physiological response data from the patient is extracted, and the phase difference spectrum between the expected and actual responses in the frequency domain is calculated. The difference spectrum is defined as: ; In the formula, For phase difference spectrum, For angular frequency variables, For operations that take complex phase angles, This represents the frequency domain of the actual physiological response. To represent the expected physiological response in the frequency domain, a weighted integral of the phase difference spectrum is performed to generate a structural-functional decoupling index. When the index exceeds a preset threshold, it indicates that there is a deviation between the patient's actual visceral microcirculation status and the physical model constructed based on imaging features, suggesting the presence of functional failure or microcirculatory disorder that has not yet been visualized on imaging.

[0042] Finally, based on the real-time calculated visceral vascular resistance index and structural-functional decoupling index, the patient's risk level is comprehensively determined. If any indicator triggers a high-risk threshold, an early warning signal is output through the interactive terminal, indicating the physiological mechanism attribution that led to the warning.

[0043] According to one embodiment, the lesion topology stiffness extraction module 200 is mainly responsible for converting unstructured medical image data into structured physical parameters that can be used for fluid dynamics calculations. Specifically, it is divided into intelligent segmentation of the region of interest, voxel-level texture feature extraction, biomechanical parameter mapping, and aggregation calculation of global equivalent stiffness.

[0044] The lesion topology stiffness extraction module 200 first receives preprocessed medical image data, which is a set of voxels containing spatial information. The lesion topology stiffness extraction module 200 automatically identifies the anatomical structure of the abdominal image using a built-in semantic segmentation algorithm. The segmentation algorithm divides the image space into several non-overlapping regions of interest (ROIs), which constitute the ROL set. ,gather Specifically, this includes: the liver region Pancreatic region Spleen area and the mesenteric vascular network region For each region of interest, the lesion topological stiffness extraction module 200 further identifies whether there are specific pathological change sub-regions within it, such as cirrhotic nodule regions, exudate regions around the pancreas, or intestinal dilatation regions. For semantic segmentation processing of medical images, those skilled in the art can use fully convolutional neural networks (FCN) or UNet architectures to implement it, which are well-known technologies in the field and will not be elaborated here.

[0045] After completing the anatomical region segmentation, the lesion topological stiffness extraction module 200 performs voxel-level texture feature extraction to quantify changes in tissue microstructure. This is because in digestive system pathology, tissue fibrosis, edema, or necrosis often precedes macroscopic morphological changes, manifesting as heterogeneous changes in image texture. For ROI sets... Each voxel in The lesion topological stiffness extraction module 200 calculates its local texture feature vector. It includes gray-level co-occurrence matrix (GLCM) features such as energy, contrast, correlation, and entropy, as well as local binary mode (LBP) features. These features can capture the roughness, uniformity, and directionality within the tissue, thereby reflecting the microscopic physical properties of the tissue.

[0046] Subsequently, the lesion topology stiffness extraction module 200 performs a mapping operation from image features to biomechanical parameters. The lesion topology stiffness extraction module 200 is pre-set with a tissue elastoplastic mapping function. This function is trained based on a pre-built pathomechanical database (containing image features corresponding to different diseases (such as acute pancreatitis, cirrhosis, and septic shock) and different disease stages, stored in the system's storage unit as a pathomechanical database). It is used to establish a nonlinear correspondence between image texture features and tissue Young's modulus. For each voxel, the lesion topological stiffness extraction module 200 extracts its texture feature vector. Input the mapping function to calculate the estimated local stiffness value at that point. : ; In the formula, coordinates Local stiffness estimation of the voxel. For the organization of the elastoplastic mapping function, This is the local texture feature vector at these coordinates. Through this step, the medical image, which originally only contained grayscale information, is transformed into a biomechanical field reflecting the distribution of tissue stiffness.

[0047] Finally, the lesion topology stiffness extraction module 200 performs the aggregation calculation of the global equivalent topology stiffness coefficient. Since visceral hemodynamics is affected by the combined state of multiple organs, it is necessary to aggregate the dispersed local stiffness values ​​into a scalar parameter that can characterize the compliance of the entire digestive system vascular bed. The lesion topology stiffness extraction module 200 calculates the global equivalent topology stiffness coefficient by weighted summation of the average stiffness of each ROI region based on the contribution of each organ to the blood flow resistance of the portal vein and visceral arteries. Global equivalent topological stiffness coefficient The calculation formula is as follows: ; In the formula: The global equivalent topological stiffness coefficient. This represents the total number of regions of interest involved in the calculation. The index number of the region of interest. For the first The weighting factor for blood flow resistance contribution in each region of interest For the first The total volume of the region of interest. In the first Three-dimensional spatial domain of a region of interest The volume integral operation performed on the above, In the spatial domain Interior coordinates The estimated local stiffness at that location.

[0048] Through the above calculations, complex and heterogeneous pathological changes in the digestive system (such as liver cirrhosis increasing portal resistance and intestinal wall edema reducing microcirculatory compliance) are uniformly mapped to a single physical parameter. This parameter directly reflects the physical constraints imposed by the current anatomical structure on the transmission of blood flow pulsation waves, providing personalized structured input for accurate prediction of subsequent basic fluid dynamics models.

[0049] The blood flow transfer function reconstruction module 300 has a pre-set basic fluid dynamics model, which is used to describe the physical attenuation and delay characteristics of the pulse wave as it is transmitted from the central aorta to the peripheral and visceral microcirculation in the human vascular network.

[0050] This fundamental fluid dynamics model is built upon the transmission line theory of fluid mechanics. It treats the vascular bed of the digestive system as a physical system that modulates the input pulse wave. To balance computational efficiency and physiological fit, the transmission process is modeled as a second-order linear time-invariant (LTI) system. In this system, peripheral vital signs are considered as the input excitation, while the perfusion state of the visceral microcirculation is considered as the output response. Although the actual vascular system has nonlinear characteristics, within the scope of small-signal analysis, the second-order LTI system can effectively capture the amplitude-frequency and phase-frequency characteristics of the pulse wave during transmission. In particular, it can characterize the influence of vascular compliance on waveform propagation speed and the attenuation effect of vascular resistance on waveform amplitude.

[0051] The transfer function structure of the basic fluid dynamics model in the complex frequency domain is defined as follows: ; In the formula: The transfer function represents the basic fluid dynamics model. Represents the complex frequency domain variables in the Laplace transform. The Laplace transform of the system output signal corresponds to the visceral microcirculation perfusion waveform to be estimated. The Laplace transform of the system input signal corresponds to the acquired high-frequency time-series vital signs signal from the periphery. This represents the fundamental natural frequency. Physically, this parameter corresponds to the resonance characteristics determined by the elastic modulus of the blood vessel wall and the cross-sectional area of ​​the lumen. Its value determines the phase delay and response speed of pulse wave transmission. This represents the basic damping ratio. Physically, this parameter corresponds to the dissipation of pulse wave energy by blood viscosity and blood vessel wall friction. Its value determines the oscillation decay rate of the waveform during transmission.

[0052] In existing technologies or general monitoring models, the baseline damping ratio and baseline natural frequency are typically set as fixed constants based on large-scale population statistics, assuming that the vascular transmission characteristics of all patients are at an average level. However, the above formula defines the mathematical framework of the model, which acts as a variable-parameter container configured to receive patient-specific physical parameters extracted from medical images in subsequent steps. Specifically, the baseline damping ratio and baseline natural frequency will be dynamically replaced and reconstructed using data output from the lesion topological stiffness extraction module 200, thereby transforming the aforementioned general baseline model into a personalized physical model that reflects the pathological changes in the digestive system of a specific patient.

[0053] The blood flow transfer function reconstruction module 300 is configured to perform parameter mapping operations. The blood flow transfer function reconstruction module 300 first receives the globally equivalent topological stiffness coefficients output by the lesion topological stiffness extraction module 200. The system believes that substantial lesions of digestive system organs (such as liver fibrosis, cirrhosis, or pancreatic necrosis) will change the mechanical environment around the vascular bed, thereby altering the energy dissipation characteristics and frequency response characteristics of blood flowing through the area. In order to quantify this effect, the blood flow transfer function reconstruction module 300 uses a preset constitutive equation to map the scalar stiffness coefficients to two core parameters in the dynamic model: damping ratio and natural frequency.

[0054] For the reconstruction of the damping ratio, a correction calculation based on the logarithmic growth law is performed. When the topological stiffness of organ tissues increases (e.g., increased microcirculatory resistance due to cirrhosis), the viscous resistance encountered by the pulse wave entering this region increases, leading to faster waveform energy decay. This manifests as enhanced overdamping characteristics of the system, resulting in a reconstructed damping ratio. The calculation formula is as follows: ; In the formula: This represents the reconfiguration damping ratio calculated based on the global equivalent topological stiffness coefficient. This represents the baseline damping ratio, a preset constant in the basic fluid dynamics model, and indicates the average vascular damping level in healthy individuals. This represents the damping sensitivity coefficient, used to adjust the weight of the effect of stiffness changes on the damping ratio. This coefficient is obtained by fitting historical clinical data. Represents the natural logarithm operation. This represents the global equivalent topological stiffness coefficient.

[0055] For the reconstruction of natural frequencies, the system performs correction calculations based on the exponential decay law. Physically, as organ lesions lead to decreased compliance (increased stiffness), the vascular bed's responsiveness to high-frequency pulsatile components changes, typically manifesting as a shift or narrowing of the system's passband, thus affecting the reconstructed natural frequencies. The calculation formula is as follows: ; In the formula: This represents the reconstructed natural frequency calculated based on the global equivalent topological stiffness coefficient. This represents the fundamental natural frequency, which is a predefined constant in the basic fluid dynamics model. Represented by natural constant Exponential operations with base 0. This represents the frequency attenuation coefficient, which reflects the degree to which pathological stiffness inhibits the vascular resonance characteristics.

[0056] After completing the independent calculations of the two parameters mentioned above, the blood flow transfer function reconstruction module 300 will... and Substituting these parameters into the algebraic structure of the basic fluid dynamics model, replacing the original fixed parameters, generates the final patient-specific peripheral visceral blood flow transfer function. Patient-specific peripheral visceral blood flow transfer function The complete expression is as follows: ; In the formula: This represents the reconstructed patient-specific peripheral visceral blood flow transfer function. Represents the complex frequency domain variables in the Laplace transform. To reconstruct natural frequencies; To reconstruct the damping ratio.

[0057] Through the aforementioned dynamic reconstruction mechanism, a personalized modeling strategy is achieved. For a patient with severe cirrhosis, the higher... This value will result in a large damping ratio in the basic fluid dynamics model. and smaller natural frequencies This enables the basic fluid dynamics model to accurately predict that even if peripheral blood pressure is normal, the visceral perfusion waveform has already shown significant amplitude decay and phase lag, thus providing a precise physical benchmark for subsequent identification of occult shock.

[0058] The visceral blood flow reverse mapping module 400 first preprocesses and performs frequency domain transformation on the acquired peripheral vital signs signals. The visceral blood flow reverse mapping module 400 then extracts a continuous photoplethysmography (PPG) wave or invasive arterial blood pressure waveform in the form of a sliding time window, which is defined as the peripheral input signal. To eliminate spectral leakage and ensure the accuracy of frequency domain analysis, the 400-pair visceral blood flow reverse mapping module... A windowing function (such as a Hanning or Hamming window) is applied, and then the signal is transformed from the time domain to the frequency domain using a Fast Fourier Transform (FFT) algorithm. This gives the frequency domain representation of the peripheral input signal. The calculation is as follows: ; In the formula: The frequency domain spectrum function representing the peripheral input signal. Represents the imaginary unit. Represents the angular frequency variable. This represents the Fourier transform operator. This represents the peripheral input signal in the time domain. Represents a time variable. It represents the base of the natural logarithm.

[0059] For discretized digital signal processing, those skilled in the art will understand that the above integration operation is specifically implemented using a fast algorithm (FFT) of the Discrete Fourier Transform (DFT).

[0060] After acquiring the spectrum of the input signal, the visceral blood flow inverse mapping module 400 performs a frequency domain synthesis operation of the visceral perfusion waveform. The visceral blood flow inverse mapping module 400 calls the patient-specific peripheral visceral blood flow transfer function output by the blood flow transfer function reconstruction module 300. Since the actual signal analysis is performed in the frequency domain, the visceral blood flow reverse mapping module 400 will map the complex frequency domain variables in the transfer function. Replace with The frequency response function of the system is obtained. According to the response theory of linear systems, the spectrum of the output signal is equal to the product of the spectrum of the input signal and the system's frequency response function. The frequency domain estimate of the visceral perfusion waveform... The calculation is as follows: ; In the formula: This represents the frequency domain spectrum function of the estimated visceral microcirculation perfusion waveform. The frequency domain spectrum function representing the peripheral input signal. This represents the patient-specific peripheral visceral blood flow transfer function. Represents the complex frequency domain variables in the Laplace transform. Represents frequency domain variables, used to replace variables in the transfer function. To obtain the frequency response.

[0061] Finally, the visceral blood flow inverse mapping module 400 performs a time-domain restoration operation, converting the frequency-domain synthesis result back into an intuitive time-series waveform. The visceral blood flow inverse mapping module 400 then calculates the visceral perfusion waveform spectrum. An inverse fast Fourier transform (IFFT) was performed to reconstruct the visceral perfusion waveform in the time domain. This waveform simulates the pressure or flow rate changes of blood flowing through the visceral microvascular network after considering the patient's specific organ stiffness, vascular damping, and compliance characteristics. (Time-domain visceral perfusion waveform) The calculation formula is as follows: ; In the formula: This represents the restored time-domain visceral perfusion waveform, which serves as the basis for subsequent calculations of visceral resistance and ischemia risk. This represents the inverse Fourier transform operator. This represents the frequency domain spectrum function of the estimated visceral microcirculation perfusion waveform. Through the above steps, the system completes the inverse visualization at the signal level, from peripheral signs to visceral state.

[0062] After restoring the time-domain waveform, the visceral blood flow reverse mapping module 400 further performs quantitative analysis to calculate the visceral vascular resistance index, which characterizes the vasomotor microcirculation contraction state. This index aims to identify, in advance, the body's sacrificial compensatory behavior of constricting visceral blood vessels to maintain perfusion of vital organs during the compensatory period before a significant drop in systemic arterial blood pressure has occurred.

[0063] The visceral blood flow inverse mapping module 400 first calculates the mean arterial pressure (MAP) based on synchronously acquired peripheral high-frequency time-series vital sign data. When the input data is an invasive arterial blood pressure waveform, the visceral blood flow inverse mapping module 400 calculates the average pressure within the cardiac cycle using an integral method. When the input data is a non-invasive photoplethysmography pulse wave, the visceral blood flow inverse mapping module 400 estimates the equivalent average perfusion pressure using a transfer feature extraction algorithm. Simultaneously, the visceral blood flow inverse mapping module 400 reconstructs the aforementioned time-domain visceral perfusion waveform. Integral processing is performed to obtain the effective visceral perfusion intensity per unit time.

[0064] Subsequently, the visceral blood flow reverse mapping module 400 calculates the visceral vascular resistance index based on the equivalent principle of Ohm's law in fluid mechanics. Physiologically, vascular resistance is defined as the ratio of driving pressure difference to blood flow. In this example, the system considers the systemic mean arterial pressure as the inlet pressure potential energy driving visceral blood flow, and the average amplitude of the reconstructed visceral perfusion waveform as the effective flow flux through the visceral vascular bed. The visceral vascular resistance index... The specific calculation formula is defined as follows: ; In the formula: The visceral vascular resistance index is a dimensionless or unitized value. An elevated value indicates that the visceral vascular bed is in a systolic state of high resistance and low perfusion. This represents a unit conversion factor or calibration constant used to eliminate the dimensional differences between pressure units and waveform amplitude units. This factor is set based on the reference parameters used during system calibration. This indicates the duration of the calculation window, typically taking the duration of one or more complete cardiac cycles. Represents the integral variable. Indicates at time The systemic arterial blood pressure value or the corresponding peripheral monitoring signal amplitude, Indicates at time The instantaneous values ​​of the restored time-domain visceral perfusion waveform. Indicates within the time window Definite integral operations within the range.

[0065] Through the above calculations, the system obtained a quantitative indicator that can dynamically reflect the tension of the visceral vascular bed. When the patient is in the early stage of shock (compensatory phase), although... (Systemic blood pressure) remains within the normal range, but due to the release of catecholamines in the body, visceral blood vessels constrict strongly. The amplitude of (visceral perfusion) will decrease. According to the above formula, the denominator decreases while the numerator remains unchanged, which will lead to... The value increases. Therefore, this index can serve as the first independent quantitative basis for diagnosing latent shock, filling the technical gap that traditional vital sign monitoring cannot directly assess visceral microcirculatory resistance.

[0066] The intervention response closed-loop analysis module 500 connects to the medical intervention event data interface to monitor in real time whether there are incentive events that meet preset conditions. The intervention response closed-loop analysis module 500 identifies two types of core incentives: one is fluid resuscitation events, specifically referring to the rapid infusion of more than a preset volume (e.g., 250 ml) of crystalloid or colloidal solution within a short period of time (e.g., within 15 minutes); the other is vasoactive drug infusion events, specifically referring to the dose adjustment operation of vasoconstrictors or vasodilators such as norepinephrine and dopamine. When either of the above events is detected, the intervention response closed-loop analysis module 500 marks it as a test input signal for the system. It also extracts physiological data within a preset time window (e.g., 30 minutes) after the event occurs as the actual response signal.

[0067] The intervention response closed-loop analysis module 500 then performs simulation calculations of the expected response, and the blood flow transfer function reconstruction module 300 generates the patient-specific peripheral visceral blood flow transfer function. Test input signal From time-domain transformation to complex frequency domain, we obtain The model calculates the theoretically expected visceral circulatory response using a transfer function. This calculation is based on the assumption that when a patient's physiological function is consistent with the anatomical structure shown in medical imaging (i.e., no imaging-invisible functional failure), the actual observed physiological response should closely match the model's predicted response.

[0068] To quantify the deviation between the actual state and the model prediction, the intervention response closed-loop analysis module 500 calculates the structural-functional uncoupling index (SVM). The structural-functional decoupling index, obtained by integrating the phase difference spectrum in the frequency domain and applying energy weighting, aims to capture the sluggish or abnormal overshooting of the visceral microcirculation in response to interventions. The calculation formula is defined as follows: ; In the formula: The structural-functional uncoupling index is a non-negative real number. The larger the value, the greater the deviation between the patient's actual physiological function and the anatomical structure model. and These represent the lower and upper frequency limits of the analysis band, respectively, and are typically set to cover the range of heart rate and its harmonic components. Indicates the frequency of the peripheral input signal. The spectral amplitude at that point is used here as a weighting function to ensure that the phase difference calculation is mainly focused on the frequency band where the signal energy is concentrated. This represents the total energy of the input signal within the analysis frequency band, used for normalizing the weights. This represents the phase angle operator. This represents the frequency domain spectrum of the physiological response signal actually observed. This represents the frequency domain spectrum of the expected physiological response signal calculated based on the transfer function. The calculation method is as follows: ,in To test the spectrum of the input signal, This represents the frequency integral variable.

[0069] In calculating the visceral vascular resistance index ( ) and structural-functional miscoupling index ( After that, the output control module 600 executes the hierarchical early warning logic, and the output control module 600 has a preset first threshold. (Resistance threshold) and second threshold (Discoupling thresholds) These two thresholds are not fixed constants, but are dynamically set based on the patient's basic physiological parameters (such as age and BMI) and historical baseline data.

[0070] The output control module 600 executes the following decision logic: if the real-time calculation... - Less than the first threshold and Less than the second threshold The system determines that it is currently in a safety compensation period, and only records data trends in the background without triggering an alarm; if - Exceeding the first threshold but Still below the second threshold The system generates a yellow warning signal, indicating that the patient's visceral blood vessels have significantly constricted to maintain blood pressure, posing a risk of tissue ischemia. However, vascular reactivity is still present, prompting clinicians to be alert to occult shock and displaying a suggestion for microcirculation resuscitation intervention on the interactive terminal.

[0071] like Exceeding the second threshold ,regardless - Regardless of the numerical value, the output control module 600 generates a red critical warning signal. This state indicates that the patient not only has perfusion abnormalities, but also that the response characteristics of the visceral vascular bed to the regulatory signal have undergone fundamental changes, resulting in functional paralysis or severe microcirculatory short circuits / blockages that have not yet been visualized on imaging, suggesting that the patient is on the verge of decompensation.

[0072] At this point, the system highlights the specific attribution analysis results through the interactive terminal, indicating whether the increased resistance is due to structural rigidity (such as portal hypertension caused by severe cirrhosis) or response mismatch due to functional dysfunction (such as vasoparalysis caused by sepsis), thereby assisting doctors in formulating precise treatment plans and avoiding blind fluid resuscitation that could worsen organ edema.

Claims

1. An AI and big data-based digestive system critical illness early warning system, characterized in that, The method comprises the following steps: a multi-modal data acquisition and alignment module configured to acquire medical image data, high-frequency time-series vital sign data, and medical intervention event data, and perform time axis synchronization and correlation operations on the data based on timestamps of the medical intervention event data as a reference to generate a multi-modal input vector; a lesion topology stiffness extraction module configured to receive the medical image data, perform anatomical region segmentation and lesion region identification on digestive system organs, and calculate an equivalent stiffness matrix reflecting the biomechanical properties of the digestive system vascular bed based on texture density features and geometric morphological features of the region of interest, and aggregate the equivalent stiffness matrix to generate a global equivalent topology stiffness coefficient; a blood flow transfer function reconstruction module configured to receive the global equivalent topology stiffness coefficient, and use the global equivalent topology stiffness coefficient to dynamically correct the structural parameters of a preset basic fluid dynamics model to generate a patient-specific peripheral-visceral blood flow transfer function; a visceral blood flow reverse mapping module configured to receive the high-frequency time-series vital sign data and the patient-specific peripheral-visceral blood flow transfer function, deduce the perfusion waveform of the visceral microcirculation through frequency domain reverse operation, and calculate the visceral vascular resistance index; an intervention response closed-loop analysis module configured to monitor the medical intervention event data, calculate the expected physiological response using the patient-specific peripheral-visceral blood flow transfer function when an incentive event is detected, and calculate the structure-function uncoupling index in combination with the actually observed physiological response; an output control module configured to generate a graded warning signal and output the attribution analysis result through an interactive terminal according to the visceral vascular resistance index and the structure-function uncoupling index. 2.The AI and big data-based digestive system critical condition early warning system according to claim 1, characterized in that, The time axis synchronization and correlation operations performed by the multi-modal data acquisition and alignment module specifically include: based on the current time as a reference, the high-frequency time-series vital sign data is intercepted within a time window of a set length, the medical image data closest to the current time and within the valid period is retrieved, and the medical intervention event data is mapped to the corresponding time axis position. 3.The AI and big data-based digestive system critical condition early warning system according to claim 1, characterized in that, The specific way in which the lesion topology stiffness extraction module calculates the global equivalent topology stiffness coefficient is: using a semantic segmentation algorithm to divide the medical image data into a set of regions of interest including the liver, pancreas, spleen, and mesenteric vascular network; extracting the local texture feature vector of each voxel in the region of interest, and using a preset tissue elastoplastic mapping function to convert the local texture feature vector into a local stiffness estimate; integrating the volume and weighted sum of the local stiffness estimates in each region of interest according to the contribution weight of each region of interest to blood flow resistance to obtain the global equivalent topology stiffness coefficient. 4.The AI and big data-based digestive system critical condition early warning system according to claim 1, wherein, The basic fluid dynamics model preset in the blood flow transfer function reconstruction module is a second-order linear time-invariant system, and the structural parameters of the basic fluid dynamics model include a basic natural frequency and a basic damping ratio; the specific process of dynamic correction includes: using the global equivalent topology stiffness coefficient to calculate the reconstruction damping ratio based on the logarithmic growth law, and calculating the reconstruction natural frequency based on the exponential decay law; The blood flow transfer function reconstruction module substitutes the reconstructed damping ratio and the reconstructed natural frequency into the base fluid dynamics model to replace the base damping ratio and the base natural frequency, thereby constructing the patient-specific peripheral-visceral blood flow transfer function. 5.The AI and big data-based digestive system critical condition early warning system according to claim 1, wherein, The process of deriving the perfusion waveform of the visceral microcirculation by the visceral blood flow reverse mapping module includes: a window function is applied to the high-frequency time-series vital sign data to eliminate spectral leakage, and then a fast Fourier transform is performed to obtain an input spectrum; a complex frequency domain variable in the patient-specific peripheral-visceral blood flow transfer function is replaced by a frequency domain variable to obtain a system frequency response function; a product of the input spectrum and the system frequency response function is calculated to obtain a frequency domain estimated value of the visceral perfusion waveform; and an inverse fast Fourier transform is performed on the frequency domain estimated value of the visceral perfusion waveform to reconstruct the perfusion waveform of the visceral microcirculation in the time domain. 6.The AI and big data-based digestive system critical condition early warning system according to claim 5, wherein, The visceral vascular resistance index is calculated by the visceral blood flow reverse mapping module in the following manner: According to the Ohm's law equivalent principle of fluid mechanics, the average systemic arterial pressure or equivalent perfusion pressure in the calculation window is calculated based on the high-frequency time-series vital sign data as a driving pressure potential energy; The reconstructed perfusion waveform of the visceral microcirculation in the time domain is integrated to obtain the effective visceral perfusion intensity in the calculation window as an effective flow flux; The ratio of the driving pressure potential energy to the effective flow flux is calculated, and the ratio is multiplied by a preset unit conversion coefficient to obtain the visceral vascular resistance index. 7.The AI and big data-based digestive system critical condition early warning system according to claim 1, wherein, The excitation events identified by the intervention response closed-loop analysis module include liquid resuscitation events and vasoactive drug infusion events; The intervention response closed-loop analysis module marks the excitation events as test input signals, and intercepts the physiological data in a preset time window after the event occurs as the actually observed physiological response; The expected physiological response is calculated by multiplying the frequency domain transform of the test input signal with the patient-specific peripheral-visceral blood flow transfer function. 8.The AI and big data-based digestive system critical condition early warning system according to claim 7, wherein, The specific way in which the structure-function decoupling index is calculated by the intervention response closed-loop analysis module is as follows: The frequency domain phase angles of the actually observed physiological response and the expected physiological response are calculated respectively; a phase difference spectrum between the frequency domain phase angles of the actually observed physiological response and the expected physiological response is calculated, and the spectrum energy distribution of the test input signal is used as a weighting function to perform integral operation on the phase difference spectrum in an analysis frequency band covering the heart rate and its harmonic components, thereby obtaining the structure-function decoupling index. 9.The AI and big data-based digestive system critical condition early warning system according to claim 1, wherein, The output control module is preset with a resistance threshold and a decoupling threshold; and the generation logic of the hierarchical early warning signal includes: When the visceral vascular resistance index exceeds the resistance threshold and the structure-function decoupling index is lower than the decoupling threshold, a first-level early warning signal is generated to indicate that there is a risk of tissue ischemia; When the structure-function decoupling index exceeds the decoupling threshold, a second-level early warning signal is generated to indicate that there is a functional failure or microcirculatory disorder, regardless of the value of the visceral vascular resistance index. 10.The AI and big data-based digestive system critical condition early warning system according to claim 9, wherein, The output control module outputs the attribution analysis result through the interactive terminal, and the attribution analysis result includes: For the first level early warning signal, prompting microcirculation resuscitation intervention; For the second level early warning signal, indicating that the physiological mechanism causing the early warning is attributed to resistance increase caused by structural stiffness or response mismatch caused by functional disorder.

Images