Intelligent assessment and diagnosis system for kawasaki disease based on multi-modal data

Abstract

The present application relates to the technical field of physiological data processing, in particular to a kawasaki disease intelligent evaluation and diagnosis system based on multi-modal data. The system collects a first pulsation characteristic signal of the left ventricular posterior wall in a cardiac cycle as an endogenous reference, and collects second pulsation characteristic signals of the left and right coronary artery walls respectively; a comparison reference weight is constructed using the statistical fluctuations of the first pulsation characteristic signal, the feature signal differences are weighted based on the weight, the matching cost is calculated combining the time delay between signal points, the matching signal point pairs and the waveform morphology difference values; the feature trajectory is reconstructed in the velocity phase space to obtain the viscoelastic damping degree of the vessel wall; the unilateral vessel wall motion deviation index is generated by fusing the morphology difference and the damping degree, and the difference between the left and right deviation indexes is calculated to output the diagnosis warning signal. The present application realizes the accurate early screening of occult kawasaki disease before the change of coronary morphology.

Description

Technical Field

[0001] This invention relates to the field of physiological data processing technology, specifically to an intelligent assessment and triage system for Kawasaki disease based on multimodal data. Background Technology

[0002] Kawasaki disease is an acute febrile rash illness characterized by systemic vasculitis, with coronary artery injury being its most serious complication. The key to clinical intervention lies in accurate identification and guidance during the early stages of the disease, specifically within the "window period" before morphological dilation of the coronary arteries occurs. In the early stages, vascular lesions primarily manifest as inflammatory edema of the vessel walls and decreased smooth muscle compliance. At this time, the vessel diameter is often within the normal range, making traditional morphological diagnostic methods based on two-dimensional ultrasound measurements of diameter or Z-scores prone to missed diagnoses. Although existing multimodal ultrasound technology can extract tissue velocity or strain parameters, the lack of individualized reference standards means these parameters are easily affected by fluctuations in the child's heart rate, blood pressure changes, and the gain settings of different ultrasound equipment, making it difficult to clearly delineate the boundary between subtle pathological changes and systemic physiological responses. For example, tachycardia caused by common fever can also lead to changes in the amplitude of vessel wall motion; this non-specific hemodynamic change often masks the true inflammatory stiffness. Furthermore, existing assessment methods typically view anatomical structure and functional status in isolation, failing to establish an endogenous correlation between the hemodynamic environment and tissue mechanical response, and thus failing to capture the key pathological feature of phase hysteresis caused by changes in the viscoelasticity of the vascular wall. Summary of the Invention

[0003] To address the technical problems of existing technologies neglecting individualized fluctuations and failing to establish an endogenous correlation between hemodynamic environment and tissue mechanical response, the present invention aims to provide an intelligent assessment and triage system for Kawasaki disease based on multimodal data. The specific technical solution adopted is as follows: This invention proposes an intelligent assessment and triage system for Kawasaki disease based on multimodal data, the system comprising: The multimodal cardiac cycle information acquisition module is used to acquire the first pulsation characteristic signal of the posterior wall of the left ventricle and the second pulsation characteristic signals of the walls of the left and right coronary arteries during the cardiac cycle; the pulsation characteristics include pulsation velocity and pulsation intensity; The unilateral cardiac feature matching module is used to match the signal points of the second pulsation feature signal and the first pulsation feature signal on each side respectively. During the matching process, a comparison benchmark weight is constructed based on the first pulsation feature signal. The signal difference between the first pulsation feature signal and the second pulsation feature signal is weighted based on the comparison benchmark weight. The matching cost is obtained by combining the time delay between the signal points. The matching signal point pair on each side is obtained based on the matching cost. The unilateral cardiac feature analysis module is used to statistically analyze the overall matching cost of the matched signal point pairs and obtain the waveform morphology difference value; analyze the phase lag and nonlinear distortion of the first and second pulsation feature signals under the pulsation velocity dimension to obtain the degree of viscoelastic damping of the blood vessel wall; and fuse the waveform morphology difference value and the degree of viscoelastic damping of the blood vessel wall to obtain the unilateral vessel wall motion deviation index. The lesion assessment module is used to provide a diagnostic warning signal based on the deviation value of the deviation index of a single vessel wall movement between the left and right sides.

[0004] Furthermore, methods for obtaining any pulsation characteristic signal include: The short-axis video stream of the myometrial tissue in the middle layer of the posterior wall of the left ventricle was used as the endogenous reference section signal, and the short-axis video stream of the coronary artery wall of the hyperechoic ring structure on either side was used as the target section signal to be tested. For any frame of the endogenous reference section signal and the target section signal, the radial displacement modulus of all pixels in the section is obtained. The first derivative of the displacement modulus with respect to time is taken as the pulsation velocity, and the average gray value of all pixels in the section is taken as the pulsation intensity. The pulsation velocity signal and the pulsation intensity signal are taken as pulsation feature signals.

[0005] Furthermore, the method for obtaining the comparison benchmark weights includes: The comparison benchmark weights include a first benchmark weight and a second benchmark weight; The first variance of the pulsation velocity and the second variance of the pulsation intensity in the first pulsation signal are statistically analyzed. The first variance is mapped by reciprocal to obtain the first benchmark weight, and the second variance is mapped by reciprocal to obtain the second benchmark weight.

[0006] Furthermore, the method for obtaining the matching cost includes: For two signal points, the squared difference between pulsation velocities is taken as the pulsation velocity difference, and the squared difference between pulsation intensities is taken as the pulsation intensity difference; the first benchmark weight is taken as the weight of the pulsation velocity difference, and the second benchmark weight is taken as the weight of the pulsation intensity difference, and the weighted sum is taken as the weighted signal difference; the sum of the weighted signal difference and the normalized time delay is taken as the matching cost.

[0007] Further, obtaining the matching signal point pair based on the matching cost includes: A matching matrix is ​​constructed using the signal point indices of the first pulsation feature signal as rows and the signal point indices of the second pulsation feature signal as columns. The element value of each element in the matching matrix is ​​the matching cost between the signal points at the corresponding index. The matching matrix is ​​then optimized by finding the minimum matching cost to obtain the matching signal point pairs.

[0008] Furthermore, the method for obtaining the waveform morphology difference value includes: The waveform morphology difference value is obtained by summing the matching costs of all matching signal point pairs and normalizing the sum.

[0009] Furthermore, the method for obtaining the degree of viscoelastic damping of the blood vessel wall includes: Using the pulsation velocity of the first pulsation characteristic signal as the abscissa and the pulsation velocity of the second pulsation characteristic signal as the ordinate, two-dimensional discrete points are obtained; the area enclosed by the trajectory formed by the two-dimensional discrete points is calculated using the discrete Green's formula to obtain the degree of viscoelastic damping of the blood vessel wall.

[0010] Furthermore, the method for obtaining the unilateral pipe wall motion deviation index includes: After standardizing the degree of vascular viscoelastic damping using a comparison benchmark weight, it is amplified by a preset amplification weight to obtain the weighted degree of vascular wall viscoelastic damping. The weighted sum of the difference between the weighted degree of vascular wall viscoelastic damping and the waveform morphology value according to the preset weight is used as the unilateral vessel wall motion deviation index.

[0011] Furthermore, the absolute value of the difference between the deviation index of a single pipe wall movement between the left and right sides is taken as the deviation value.

[0012] Furthermore, the method for providing feedback on the triage and early warning signal includes: The preset ratio of the maximum value of the single tube wall movement deviation index between the left and right sides is used as the first judgment threshold; the preset multiple of the preset health benchmark is used as the second judgment threshold. If the deviation value is greater than the preset health benchmark and greater than the first judgment threshold, a high-risk triage warning signal is fed back. If the deviation value is less than or equal to the preset health benchmark, and the maximum value of the single tube wall movement deviation index between the left and right sides is greater than the second judgment threshold, then a medium-risk triage warning signal is fed back. If the deviation value is less than or equal to the preset health benchmark, and the maximum value of the single tube wall movement deviation index between the left and right sides is less than or equal to the second judgment threshold, then a low-risk triage warning signal is fed back.

[0013] The present invention has the following beneficial effects: This invention establishes a data foundation for homologous comparison by acquiring the first pulsational characteristic signal of the left ventricular posterior wall (posterior ventricular wall) and the second pulsational characteristic signals of both sides (coronary artery walls), eliminating interference from heart rate variability introduced by asynchronous acquisition. A comparison benchmark weight is constructed based on the pulsational characteristic signal of either side, and this weight is used to weight the differences between the two sides. The matching cost is calculated in conjunction with time delay, achieving normalization based on individual physiological benchmarks. This technique effectively solves the problems of existing technologies lacking individualized reference standards and being highly susceptible to fluctuations in the child's heart rate and equipment gain, clearly defining the boundary between minor vascular pathological changes and systemic physiological responses. The unilateral cardiac characteristic analysis module not only statistically analyzes the overall matching cost to obtain waveform morphology difference values, but also deeply analyzes the phase lag and nonlinear distortion in the pulsation velocity dimension, quantifies the degree of viscoelastic damping of the vessel wall, and fuses the two to generate a unilateral vessel wall motion deviation index. This process establishes an endogenous correlation between the hemodynamic environment and tissue mechanical response, successfully capturing key pathological features of the vessel wall caused by inflammatory edema and decreased compliance. Finally, based on the deviation values ​​of the unilateral vessel wall motion deviation index on both sides, a diagnostic warning signal is fed back. Utilizing the focal distribution characteristics of early-stage vasculitis in Kawasaki disease, highly specific lesion identification and stratified diagnostic guidance are achieved. This invention can accurately identify occult vessel wall functional damage in the early stages before significant dilation of the coronary arteries, assisting in timely clinical intervention. Attached Figure Description

[0014] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a block diagram of a Kawasaki disease intelligent assessment and triage system based on multimodal data, provided as an embodiment of the present invention. Detailed Implementation

[0016] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a Kawasaki disease intelligent assessment and triage system based on multimodal data proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0018] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent assessment and triage system for Kawasaki disease based on multimodal data provided by this invention.

[0019] Please see Figure 1 The diagram illustrates a block diagram of a Kawasaki disease intelligent assessment and triage system based on multimodal data provided in an embodiment of the present invention. The system includes a multimodal cardiac cycle information acquisition module 101, a unilateral cardiac feature matching module 102, a unilateral cardiac feature analysis module 103, and a lesion assessment module 104.

[0020] The system proposed in this invention can be integrated into existing echocardiography workstations or PACS imaging centers for targeted intelligent monitoring of Kawasaki disease. Its applications include early outpatient screening for suspected cases in children with fever for 3-5 days but without coronary artery dilation (normal Z-score). By comparing the hysteresis of the vessel wall relative to the myocardium, this system identifies occult vessel wall inflammation before morphological changes, assisting in emergency triage. Another application is inpatient drug efficacy assessment. Before and after intravenous immunoglobulin (IVIG) treatment, by monitoring the dynamic changes in the degree of viscoelastic damping of the vessel wall, it assesses whether vascular compliance has recovered, determines treatment sensitivity, and guides clinical adjustments to hormone regimens. This system relies on existing medical monitoring systems and requires no additional hardware support. High-end color Doppler ultrasound machines in existing medical monitoring systems are equipped with ECG trigger interfaces and tissue Doppler (TDI) modes. Therefore, the multimodal cardiac cycle information acquisition module 101 directly calls the DICOM stream data with ECG R-wave markers exported from the existing ultrasound machine. Using the R-wave as a logic switch, the system automatically extracts the velocity and echo sequences within the cardiac cycle to construct the corresponding pulsational characteristic signals. Existing medical monitoring systems generally possess a DICOM image processing plug-in architecture; therefore, this system operates as a workstation plug-in to implement the matching process in the unilateral cardiac feature matching module 102. Similarly, based on the existing wall motion analysis algorithm framework in the ultrasound system, this system adds the logic process in the unilateral cardiac feature analysis module 103 to obtain the corresponding output results. Finally, utilizing the structured interface of the existing medical monitoring system, the system can directly visualize the warning signals from the lesion assessment module 104 as an assessment report.

[0021] This invention aims to quantify the asymmetry of the pathological states of the left and right vessel walls, thereby determining the vessel wall status. For example, when the pathological states of the left and right vessel walls are significantly inconsistent, it indicates that one vessel wall has exhibited specific dynamic changes not seen on the other side, consistent with the focal characteristics of early vasculitis in Kawasaki disease. The system proposed in this invention requires extracting the states of the left and right vessel walls separately for comparative analysis. Considering that the left ventricular posterior wall is thick and located deep within the heart, its information features are relatively clear; therefore, the left ventricular posterior wall can represent the patient's current systemic myocardial local dynamic state. Therefore, the multimodal cardiac cycle information acquisition module 101 is used to acquire the first pulsational feature signal of the left ventricular posterior wall and the second pulsational feature signals of the left and right coronary artery walls during the cardiac cycle. The pulsational features include two dimensions: pulsation velocity and pulsation intensity. In subsequent modules, the first pulsational feature is used as a physiological benchmark to construct a comparison benchmark weight to ensure comparability in matching or other data analysis processes.

[0022] Preferably, in this embodiment of the invention, the method for obtaining any pulsation feature signal includes: Within one cardiac cycle, the short-axis video stream of the myocardial layer in the middle layer of the posterior wall of any ventricle is used as the endogenous reference section signal, and the short-axis video stream of the coronary artery wall of the hyperechoic ring structure on any side is used as the target section signal.

[0023] As a specific example, in one implementation of this invention, because the coronary arteries are small in diameter and easily affected by lung air, while the posterior wall of the left ventricle is thick and located deep within the heart, it is difficult to simultaneously obtain clear images that meet the requirements for quantitative analysis in a single ultrasound section. Therefore, in the actual acquisition process, time-division independent acquisition is required, locking the optimal acoustic window to obtain the target video stream, specifically including: First, the ultrasound probe is positioned at the 3rd-4th intercostal space along the left parasternal border of the patient to acquire a standard long-axis or short-axis video stream containing the posterior wall of the left ventricle. The myocardial tissue in the middle layer of the posterior wall of the left ventricle is identified and located within the image, defined as the endogenous reference region. The motion state of this region reflects the patient's myocardial baseline strain capacity under the current hemodynamic environment, thus providing an endogenous reference section signal. The ultrasound probe angle is then adjusted to acquire a short-axis video stream containing the wall of the left coronary artery (e.g., the left anterior descending artery) or the right coronary artery. The vessel wall exhibiting a hyperechoic ring structure is identified in the image and defined as the target region. That is, the acquisition sequence of the endogenous reference section signal and the target section signal is not simultaneous, but rather based on a sequential search for suitable ultrasound sections. Considering that the patient's ECG cycle length may differ at different acquisition times (e.g., heart rate 80 when acquiring the endogenous reference section signal and 85 when acquiring the target section signal), the resulting signal lengths may differ, making subsequent matching inconvenient. Therefore, temporal alignment is also required. For any slice signal containing a complete cardiac cycle, an electrocardiogram (ECG) signal acquired synchronously is obtained. The timestamps of two adjacent R-wave peaks in the ECG signal are identified, and the interval between these two timestamps is the cardiac cycle interval. Within the cardiac cycle interval, the signal is discretized into a corresponding number of time points according to a preset resolution. Cubic spline interpolation is used to map the original slice signal onto the discretized time points. Cubic spline interpolation ensures the continuity of the resampled signal on its second derivative, thereby preserving the acceleration characteristics of the blood vessel wall motion and avoiding the introduction of high-frequency computational noise. Thus, two types of slice signals with the same signal length can be obtained for each cardiac cycle interval. In the specific implementation of this invention, the resolution is set to 100, meaning the number of discretized time points is 100. The specific slice extraction method, cubic spline interpolation, cardiac cycle interval acquisition method, and discretization method are all well-known techniques to those skilled in the art and will not be elaborated upon here.

[0024] For any frame of the endogenous reference section signal and the target section signal, the radial displacement modulus of all pixels in the section is obtained, and the first derivative of the displacement modulus with respect to time is taken as the pulsation velocity, that is, the pulsation velocity characterizes the feature of tissue mechanical motion.

[0025] The grayscale value on an ultrasound image represents the echo intensity; therefore, the average grayscale value of all pixels within the slice is taken as the pulsation intensity. That is, pulsation intensity characterizes the tissue texture. The pulsation velocity signal and the pulsation intensity signal are taken as the pulsation feature signals. Specifically, the pulsation feature signal corresponding to the endogenous reference slice signal is the first pulsation feature signal, and the pulsation feature signal corresponding to the target slice signal is the second pulsation feature signal.

[0026] It should be noted that, in order to prevent respiratory baseline drift, the system in this embodiment of the invention only selects the period of the patient's stable respiratory and cardiac cycle for information collection.

[0027] The essence of early-stage vasculitis in Kawasaki disease is the alteration of viscoelastic mechanical properties of the blood vessel wall due to inflammatory edema and decreased smooth muscle compliance. Macroscopically, this manifests as "waveform distortion" and "phase lag" in the mechanical response of the blood vessel wall to blood pressure fluctuations, and these subtle changes are often masked by systemic heart rate fluctuations. Therefore, focusing solely on the expansion of the blood vessel diameter or changes in absolute blood flow velocity is insufficient for effectively assessing pathological features. This invention considers that the movement of the blood vessel wall in pathological states may exhibit nonlinear time scaling or shifts. Simple point-to-point linear subtraction cannot capture the intricate dynamic differences. Therefore, to solve this technical problem and achieve effective extraction of pathological features, it is necessary to match two pulsation feature signals across a cross-time dimension, thereby finding potential correspondences between different time points.

[0028] Therefore, the unilateral cardiac feature matching module 102 matches the signal points of the second pulsation feature signal and the first pulsation feature signal on each side. During the matching process, the pulsation feature differences and time delay between signal points are typically considered as matching costs. However, since the pulsation features included in this embodiment of the invention include two modalities—pulsation velocity and pulsation intensity—which belong to different physical dimensions, and the ultrasound equipment gain settings differ among different patients, directly sampling the quantitative differences between features for calculation would lead to a lack of comparability in the matching process. Therefore, to address the problems of existing technologies lacking individualized reference standards and being highly susceptible to fluctuations in the child's heart rate and equipment gain, a clear boundary is drawn between minor vascular pathological changes and systemic physiological responses. The unilateral cardiac feature matching module 102 constructs a comparison benchmark weight based on the first pulsation feature signal, and weights the signal differences between the first and second pulsation feature signals based on this benchmark weight. This benchmark weight eliminates the dimensional influence of the signal differences between the two pulsation features, and then combines this with the time delay to obtain the matching cost, thus obtaining the matched signal point pairs on each side. By constructing two sets of baseline weights, the same set of weights is mandatory for standardization when assessing either the left or right coronary artery in subsequent processes. This mechanism ensures that the assessment results of the left and right coronary arteries are measured based on the same standard, thus providing a reasonable data foundation for subsequent modules to analyze the asymmetric pathological features of the left and right sides.

[0029] Preferably, in this embodiment of the invention, the method for obtaining the comparison benchmark weight includes: The comparison benchmark weights include a first benchmark weight and a second benchmark weight; The first variance of the pulsation velocity and the second variance of the pulsation intensity in the first pulsation signal are statistically analyzed. The first variance is mapped inversely to obtain the first benchmark weight, and the second variance is mapped inversely to obtain the second benchmark weight. Because the mapping uses an inverse method, the variance represents the pulsation amplitude. Therefore, in the subsequent weighting process, dividing by the variance (i.e., multiplying by the benchmark weight) forcibly pulls large and small fluctuation signals to the same standard line of relative intensity, eliminating amplitude differences and making the data from the two different modalities comparable.

[0030] In a specific implementation of this invention, considering that the variance may be 0, when performing the reciprocal mapping, the variance is not directly used as the denominator. Instead, the sum of the variance and the preset adjustment parameter is used as the denominator, and the positive integer 1 is used as the numerator to obtain the benchmark weight. The preset adjustment parameter can be set to 0.01.

[0031] Furthermore, in this embodiment of the invention, considering that the benchmark weight is represented by variance, taking the pulsation velocity dimension as an example, the variance unit of this dimension is (cm / s)². Therefore, in order to further eliminate the influence of dimensions, the method for obtaining the matching cost includes: For two signal points, the squared difference between their pulsation velocities is taken as the pulsation velocity difference, and the squared difference between their pulsation intensities is taken as the pulsation intensity difference. A first benchmark weight is used as the weight for the pulsation velocity difference, and a second benchmark weight is used as the weight for the pulsation intensity difference. These weighted sums are then used to obtain the weighted signal difference. The sum of the weighted signal difference and the normalized time delay is taken as the matching cost. In other words, the weighted signal difference characterizes the signal difference features of the two modal dimensions after standardization. The smaller this difference feature, the more matched the two signal points are. The smaller the time delay, the more similar the positions of the two signal points are within a cardiac cycle, and the more likely they belong to the same matching pair. That is, the smaller the matching cost of two signal points, the more matched the two signal points are, and the more consistent the mechanical response of the coronary artery wall to the same dynamic source (cardiac pulsation / pressure wave).

[0032] As a concrete example, the above process of obtaining the matching cost can be expressed by the following formula: in, The matching cost between the i-th signal point in the first pulsation feature signal and the j-th signal point in the second pulsation signal; As the first benchmark weight; The pulsation velocity of the j-th signal point in the second pulsation signal; The pulsation velocity of the i-th signal point in the first pulsation characteristic signal; As the second benchmark weight; The pulsation intensity of the j-th signal point in the second pulsation signal; The pulsation intensity is the i-th signal point in the first pulsation characteristic signal; The signal length is 100 in one specific implementation of this invention.

[0033] In the above formula, the weighted signal difference quantifies the absolute difference of transient physical characteristics. The time delay term is used to limit the timing matching with too large a span and prevent the physiological timing of the cardiac cycle from being violated during the matching process. Therefore, normalization is performed in the form of signal length as denominator, and feature amplification is also performed in the form of square.

[0034] Preferably, in this embodiment of the invention, obtaining the matching signal point pair based on the matching cost includes: A matching matrix is ​​constructed using the signal point indices of the first pulsation feature signal as rows and the signal point indices of the second pulsation feature signal as columns. The element value of each element in the matching matrix is ​​the matching cost between the signal points at the corresponding index. The matching matrix is ​​then optimized by minimizing the matching cost to obtain the matched signal point pairs. In one specific implementation of this invention, the optimization algorithm employs the Hungarian algorithm or the KM algorithm. This global optimization effectively resists occasional speckle noise or artifact interference in ultrasound images, ensuring that the evaluation results reflect the topological consistency of the overall waveform structure.

[0035] The posterior wall of the left ventricle (myocardium) and the coronary artery wall are driven by the same heartbeat. In a healthy state, their motion trajectories should be highly synchronized and their waveforms similar. The lower the matching cost of the matching signal point pair, the higher the degree of matching between the two signal points, indicating that the motion logic of the coronary artery wall completely follows the pulsating pattern of the myocardium. This represents good elasticity of the vessel wall, which can conform to the overall contraction and relaxation of the heart. Conversely, the higher the matching cost, the lower the degree of matching, indicating that the motion of the coronary artery wall deviates from the regularity of the healthy state, and the motion response has changed. Due to typical pathologies of Kawasaki disease, such as inflammatory edema, fibrosis, or stiffness, the vessel wall cannot achieve the smooth motion of the healthy state, resulting in personalized distortion and damping. Therefore, the unilateral cardiac motion feature analysis module 103 first calculates the overall matching cost of the matching signal point pair to obtain the waveform morphology difference value. That is, the larger the waveform morphology difference value, the more likely the motion of the vessel wall has undergone personalized variation.

[0036] Considering that in a healthy state, the blood vessel wall has good compliance, and its expansion rate completely follows blood pressure fluctuations, showing a linear relationship between the two; however, if the blood vessel wall experiences inflammatory edema or stiffening, the response relative to blood pressure will be delayed, leading to a significant nonlinear difference in pulsation velocity between the posterior wall of the ventricle and the coronary arteries. Pathological motion abnormalities such as premature beats and local stenosis caused by inflammation will further exacerbate this nonlinear difference. Therefore, the unilateral cardiac motion feature analysis module 103 further analyzes the phase lag and nonlinear distortion of the first and second pulsation feature signals under the pulsation velocity dimension to obtain the degree of viscoelastic damping of the blood vessel wall. That is, the degree of viscoelastic damping of the blood vessel wall is a quantitative result of the mismatch between phase lag and waveform morphology. The larger this index, the more sluggish the mechanical response of the blood vessel is relative to blood pressure, and the greater the degree of pathological dynamic mismatch.

[0037] The unilateral cardiac motion feature analysis module 103 integrates two features characterizing the degree of pathological vascular motion mismatch: waveform morphology difference and vessel wall viscoelastic damping, to obtain a unilateral vessel wall motion deviation index. That is, a corresponding unilateral vessel wall motion deviation index can be obtained for both the left and right coronary arteries, representing the degree of dynamic mismatch between the corresponding side coronary artery and the left ventricular posterior wall baseline.

[0038] Preferably, in this embodiment of the invention, the method for obtaining waveform morphology difference values ​​includes: The matching costs of all matched signal point pairs are summed and normalized to obtain the waveform morphology difference value. It should be noted that, to ensure the comparability of indicators among different patients, in one specific implementation of this invention, the autocorrelation energy of the first pulsation characteristic signal is used as the normalization denominator. This autocorrelation energy is expressed by the formula: ;in For autocorrelation energy, As the first benchmark weight, The average pulsation velocity in the first pulsation characteristic signal. Let be the pulsation velocity of the k-th signal point in the first pulsation characteristic signal. As the second benchmark weight, The average pulsation intensity in the first pulsation characteristic signal. The pulsation intensity is the k-th signal point in the first pulsation characteristic signal.

[0039] In the above formula for obtaining autocorrelation energy, the benchmark weight is also used for dimensional unification. The corresponding average value is subtracted from the two dimensions respectively, so that the final autocorrelation energy purely represents the pulsating energy of the cardiac tube wall, and is not affected by the overall brightness of the ultrasound image background and the normal overall translation of the heart, thus extracting pure dynamic feature energy.

[0040] Normalization is achieved by using the sum of the autocorrelation energy and the preset adjustment parameter as the denominator, and the sum of the matching costs of all matched signal point pairs as the numerator, thus obtaining the waveform morphology difference value. The setting of the preset adjustment parameter has been explained above and will not be repeated here.

[0041] Preferably, in this embodiment of the invention, the method for obtaining the degree of viscoelastic damping of the blood vessel wall includes: Using the pulsation velocity of the first pulsation characteristic signal as the abscissa and the pulsation velocity of the second pulsation characteristic signal as the ordinate, two-dimensional discrete points are obtained. The area enclosed by the trajectory formed by the two-dimensional discrete points is calculated using the Discrete Green's formula to obtain the degree of viscoelastic damping of the blood vessel wall. It should be noted that, because the embodiments of this invention obtain two-dimensional discrete points, they are sorted according to the time sequence of the two-dimensional discrete points, with the first two-dimensional discrete point also being used as the last two-dimensional discrete point. This ensures that the Discrete Green's formula can loop back to the first point after the last index processing, realizing the area quantification of the closed trajectory. This area quantification is a physically closed energy dissipation loop. That is, the larger the area, the more prominent the nonlinear distortion in the pulsation velocity dimension; the smaller the area, the closer the trajectory formed by the two-dimensional discrete points is to a straight line, indicating stronger blood vessel compliance and a more likely healthy state.

[0042] Preferably, in some implementations of the present invention, considering that dimensionless processing is also required after obtaining the degree of viscoelastic damping of the blood vessel wall, the degree of viscoelastic damping of the blood vessel is standardized using a comparison benchmark weight. In the embodiments of the present invention, the degree of viscoelastic damping of the blood vessel wall is multiplied by the first benchmark weight, and then multiplied by 0.5 to obtain the final result. The reason for multiplying by 0.5 is that the coefficient 0.5 is the standard geometric coefficient for calculating the area of ​​a closed polygon using the Discrete Green's formula, used to convert the accumulated value of the coordinate cross product into the geometric area enclosed by the trajectory. Finally, the degree of viscoelastic damping of the blood vessel wall that can participate in subsequent feature fusion is obtained.

[0043] After standardization, the signal is amplified using preset amplification weights to obtain a weighted vascular wall viscoelastic damping degree. The weighted sum of the vascular wall viscoelastic damping degree and the waveform morphology difference value is then used as the unilateral vessel wall motion deviation index. The amplification weights increase the weight of the vascular viscoelastic damping degree in the final fused features. The weights in the weighted summation process can all be set to 0.5, meaning the average value of the weighted vascular wall viscoelastic damping degree and the waveform morphology difference value is selected as the unilateral vessel wall motion deviation index. This ensures that even without significant waveform distortion, as long as the viscoelastic hysteresis is significant, the index can still effectively characterize abnormal vascular motion. In this embodiment, the total amplification weight is set to 5, meaning the vascular wall viscoelastic damping degree is amplified by a factor of 5.

[0044] Early lesions of Kawasaki disease vasculitis are characterized by random, patchy, focal distribution, while systemic infection or excessive cardiovascular load is characterized by diffuse, symmetrical distribution. The symmetry of the single-shot vessel wall motion deviation index between the left and right sides can distinguish between these two different pathological manifestations. Therefore, the lesion assessment module 104 provides a warning signal based on the deviation value of the single-shot vessel wall motion deviation index between the left and right sides. That is, the smaller the deviation value, the more consistent the deviation of the bilateral vessel walls relative to the myocardial baseline, which may occur in a normal state where both sides are healthy, or in a systemic reaction state where both sides are affected, but it is not the pathology characterized by Kawasaki disease. Conversely, a larger deviation value indicates that one side of the vessel wall has specific dynamic changes not seen on the other side (such as significant unilateral phase lag or waveform distortion), consistent with the focal characteristics of early Kawasaki disease vasculitis.

[0045] Preferably, in this embodiment of the invention, the absolute value of the difference between the deviation index of a single pipe wall movement between the left and right sides is taken as the deviation value.

[0046] Preferably, the method for providing feedback on the triage and early warning signal includes: A preset ratio of the maximum value of the single-time tube wall movement deviation index between the left and right sides is used as the first judgment threshold; a preset multiple of the preset health benchmark is used as the second judgment threshold. In a specific implementation of this invention, the preset health benchmark can be regarded as the noise base of a healthy population statistical study, representing the allowable range of measurement error, and can be set to 0.1. The preset ratio can be set to 0.5. The first judgment threshold obtained by setting the preset ratio based on the maximum value can serve as a threshold value representing "high risk". The second judgment threshold can be set to three times the health benchmark, that is, the second judgment threshold serves as a threshold value representing significant lesions.

[0047] If the deviation value is greater than the preset health baseline and also greater than the first judgment threshold, it indicates that the patient has a significant symmetry abnormality, with one side of the blood vessel wall showing obvious inflammatory stiffness or viscoelastic hysteresis, while the other side is relatively normal. This is a typical indication of early occult coronary artery damage in Kawasaki disease, which is distinctly different from the systemic symmetrical response caused by ordinary viral infections. Providing a high-risk triage warning signal can suggest that clinicians immediately perform an immunoglobulin (IVIG) sensitivity assessment and increase the frequency of ultrasound follow-up, focusing on the blood vessel with the higher value. If the deviation value is less than or equal to the preset health baseline, and the maximum value of the single-shot vessel wall motion deviation index between the left and right sides is greater than the second judgment threshold, it indicates that although the deviation value is small at this time, the single-sided vessel wall motion deviation index on one side is significantly higher than the normal standard. This state indicates that diffuse dynamic abnormalities have occurred in both bilateral vessel walls and myocardial systems, and to a similar degree. This usually corresponds to systemic hemodynamic disturbances caused by multisystem inflammatory syndrome (MIS-C), severe myocarditis, or septic shock, rather than typical focal coronary vasculitis. Therefore, a mid-risk warning signal is issued, suggesting that clinicians pay attention to cardiac function indicators, rule out other systemic infections or immune system diseases, and not prioritize treatment options for simple coronary vasculitis.

[0048] If the deviation value is less than or equal to the preset health baseline, and the maximum value of the single-time vessel wall movement deviation index between the left and right sides is less than or equal to the second judgment threshold, it indicates that the synchronicity between the left and right coronary arteries and the myocardial baseline is good in this state, and there are no obvious changes in the mechanical properties of the vessel walls. The current fever symptoms have not caused specific damage to the cardiovascular system. A low-risk triage warning signal is issued, suggesting that the fever be treated according to the routine fever procedure, without the need for special cardiovascular drug intervention, to avoid over-treatment.

[0049] It should be noted that the suggested content corresponding to the above warning signals can be generated together with the warning signals through the existing medical system's report generation system to remind medical staff.

[0050] In summary, this invention acquires the first pulsational characteristic signal of the left ventricular posterior wall as an endogenous reference during the cardiac cycle, and acquires the second pulsational characteristic signals of the left and right coronary artery walls respectively. A comparison benchmark weight is constructed using the statistical fluctuation of the first pulsational characteristic signal. Based on this weight, the differences in characteristic signals are weighted, and the matching cost is calculated by combining the time delay between signal points. Global optimization is used to obtain the matching signal point pairs and waveform morphology differences. Simultaneously, the characteristic trajectory is reconstructed in the velocity phase space, and the degree of phase lag and nonlinear distortion is quantified to obtain the degree of viscoelastic damping of the vessel wall. The morphological differences and damping degree are fused to generate a unilateral vessel wall motion deviation index, and the difference between the deviation indices of the left and right sides is calculated. Based on the principle of symmetry breaking, focal vasculitis is identified, and a diagnostic warning signal is output. This invention achieves accurate early screening for occult Kawasaki disease before changes in coronary artery morphology occur.

[0051] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0052] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A Kawasaki disease intelligent assessment and triage system based on multimodal data, characterized in that, The system includes: The multimodal cardiac cycle information acquisition module is used to acquire the first pulsation characteristic signal of the posterior wall of the left ventricle and the second pulsation characteristic signals of the walls of the left and right coronary arteries during the cardiac cycle; the pulsation characteristics include pulsation velocity and pulsation intensity; The unilateral cardiac feature matching module is used to match the signal points of the second pulsation feature signal and the first pulsation feature signal on each side respectively. During the matching process, a comparison benchmark weight is constructed based on the first pulsation feature signal. The signal difference between the first pulsation feature signal and the second pulsation feature signal is weighted based on the comparison benchmark weight. The matching cost is obtained by combining the time delay between the signal points. The matching signal point pair on each side is obtained based on the matching cost. The unilateral cardiac feature analysis module is used to statistically analyze the overall matching cost of the matched signal point pairs and obtain the waveform morphology difference value; analyze the phase lag and nonlinear distortion of the first and second pulsation feature signals under the pulsation velocity dimension to obtain the degree of viscoelastic damping of the blood vessel wall; and fuse the waveform morphology difference value and the degree of viscoelastic damping of the blood vessel wall to obtain the unilateral vessel wall motion deviation index. The lesion assessment module is used to provide a diagnostic warning signal based on the deviation value of the deviation index of a single vessel wall movement between the left and right sides.

2. The Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 1, characterized in that, Methods for obtaining any pulsation characteristic signal include: The short-axis video stream of the myometrial tissue in the middle layer of the posterior wall of the left ventricle was used as the endogenous reference section signal, and the short-axis video stream of the coronary artery wall of the hyperechoic ring structure on either side was used as the target section signal to be tested. For any frame of the endogenous reference section signal and the target section signal, the radial displacement modulus of all pixels in the section is obtained. The first derivative of the displacement modulus with respect to time is taken as the pulsation velocity, and the average gray value of all pixels in the section is taken as the pulsation intensity. The pulsation velocity signal and the pulsation intensity signal are taken as pulsation feature signals.

3. The Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 1, characterized in that, The method for obtaining the benchmark weights includes: The comparison benchmark weights include a first benchmark weight and a second benchmark weight; The first variance of the pulsation velocity and the second variance of the pulsation intensity in the first pulsation signal are statistically analyzed. The first variance is mapped by reciprocal to obtain the first benchmark weight, and the second variance is mapped by reciprocal to obtain the second benchmark weight.

4. The Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 3, characterized in that, The method for obtaining the matching cost includes: For two signal points, the squared difference between pulsation velocities is taken as the pulsation velocity difference, and the squared difference between pulsation intensities is taken as the pulsation intensity difference; the first benchmark weight is taken as the weight of the pulsation velocity difference, and the second benchmark weight is taken as the weight of the pulsation intensity difference, and the weighted sum is taken as the weighted signal difference; the sum of the weighted signal difference and the normalized time delay is taken as the matching cost.

5. The Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 1, characterized in that, The step of obtaining the matching signal point pair based on the matching cost includes: A matching matrix is ​​constructed using the signal point indices of the first pulsation feature signal as rows and the signal point indices of the second pulsation feature signal as columns. The element value of each element in the matching matrix is ​​the matching cost between the signal points at the corresponding index. The matching matrix is ​​then optimized by finding the minimum matching cost to obtain the matching signal point pairs.

6. The Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 1, characterized in that, The method for obtaining the waveform morphology difference value includes: The waveform morphology difference value is obtained by summing the matching costs of all matching signal point pairs and normalizing the sum.

7. The Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 1, characterized in that, The method for obtaining the degree of viscoelastic damping of the blood vessel wall includes: Using the pulsation velocity of the first pulsation characteristic signal as the abscissa and the pulsation velocity of the second pulsation characteristic signal as the ordinate, two-dimensional discrete points are obtained; the area enclosed by the trajectory formed by the two-dimensional discrete points is calculated using the discrete Green's formula to obtain the degree of viscoelastic damping of the blood vessel wall.

8. The Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 7, characterized in that, The method for obtaining the single-sided pipe wall motion deviation index includes: After standardizing the degree of vascular viscoelastic damping using a comparison benchmark weight, it is amplified by a preset amplification weight to obtain the weighted degree of vascular wall viscoelastic damping. The weighted sum of the difference between the weighted degree of vascular wall viscoelastic damping and the waveform morphology value according to the preset weight is used as the unilateral vessel wall motion deviation index.

9. The Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 1, characterized in that, The absolute value of the difference between the deviation index of a single pipe wall movement between the left and right sides is taken as the deviation value.

10. A Kawasaki disease intelligent assessment and triage system based on multimodal data according to claim 1, characterized in that, The method for providing feedback on the triage and early warning signal includes: The preset ratio of the maximum value of the single tube wall movement deviation index between the left and right sides is used as the first judgment threshold; the preset multiple of the preset health benchmark is used as the second judgment threshold. If the deviation value is greater than the preset health benchmark and greater than the first judgment threshold, a high-risk triage warning signal is fed back. If the deviation value is less than or equal to the preset health benchmark, and the maximum value of the single tube wall movement deviation index between the left and right sides is greater than the second judgment threshold, then a medium-risk triage warning signal is fed back. If the deviation value is less than or equal to the preset health benchmark, and the maximum value of the single tube wall movement deviation index between the left and right sides is less than or equal to the second judgment threshold, then a low-risk triage warning signal is fed back.

Images