A system and method for auxiliary detection of adenoid hypertrophy based on dynamic image fusion

Abstract

The present invention proposes an auxiliary detection system and method for adenoids hypertrophy based on dynamic graph fusion. According to the left and right nasal breathing airflow velocity data of the tester, a certain learning method is used to model the unknown nonlinear dynamics contained in the respiratory airflow velocity signal to obtain left and right respiratory dynamic graphs; and a weight matrix is ​​constructed to fuse the left and right respiratory dynamic graphs. Adenoids hypertrophy disease is diagnosed based on the morphological characteristics of the fused respiratory dynamic graphs. The constructed weight matrix can effectively reflect the pathological respiratory characteristics of adenoids hypertrophy, can weaken the influence caused by the above-mentioned data asymmetry, and improve the diagnostic accuracy of adenoids hypertrophy and the adaptability to adenoids hypertrophy disease.

Description

Technical Field

[0001] The present invention belongs to the technical field related to adenoids auxiliary detection, and in particular relates to a system and method for auxiliary detection of adenoids hypertrophy based on kinetic image fusion. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Adenoidal hypertrophy (AH) is the enlargement of the adenoids (the pharyngeal tonsils). It is commonly caused by recurrent episodes of acute and chronic nasopharyngitis, as well as inflammation of adjacent organs such as the nasal cavity, sinuses, and tonsils, which stimulate pathological hyperplasia of the adenoid tissue. Because the adenoids naturally shrink during puberty, children are most commonly affected by adenoid diseases and conditions, although some adults may also develop the condition. Adenoid hypertrophy reduces or obstructs the upper airway and is the most common cause of upper airway obstruction in children and adolescents. Upper airway obstruction can contribute to health problems such as mouth breathing, nasal diseases, asthma, speech problems, and sleep apnea, and may even lead to more serious conditions such as obstructive sleep apnea, craniofacial changes, and cognitive impairment. If left untreated, adenoid hypertrophy often leads to long-term health consequences. The prevalence of adenoid hypertrophy in a randomly representative sample of children is estimated to be approximately 34.46%.

[0004] Currently, the main medical methods for diagnosing adenoids are nasopharyngeal endoscopy and nasopharyngeal radiology, such as lateral skull X-rays. However, these existing diagnostic methods for adenoids have their own clinical pain points: nasopharyngeal endoscopy is the gold standard for diagnosing adenoids, but the endoscopic procedure requires the endoscope to penetrate deep into the child's nasopharynx, which may cause fear and physical discomfort in the child, hindering the examination specialist's operation. In addition, the results of nasopharyngeal endoscopy may be affected by the examiner's subjective bias. Nasopharyngeal radiology is an objective and non-invasive tool for evaluating adenoids in the nasopharynx, but children's easy movement makes imaging difficult, and X-rays can also cause certain radiation damage.

[0005] Patent document CN117690580A proposes a respirodynamicsgram-based adenoids hypertrophy detection system. This system captures nasal airflow changes caused by abnormal nasal structure or related diseases, providing a convenient new approach for evaluating nasal diseases. Based on deterministic learning, a new dynamic environment machine learning technique, the system generates a respirodynamicsgram (RDG) from airflow velocity data collected by a nasal resistance meter. By extracting features from the respirodynamicsgram, the system enables dynamic assessment and diagnosis of adenoids hypertrophy.

[0006] However, there are several shortcomings in the existing adenoids hypertrophy detection system based on pneumodynamics:

[0007] First, the respiratory dynamics map is independently generated using the airflow velocity data of the patient's left and right nasal cavities collected from the nasal resistance meter. This means that the respiratory dynamics map of each nasal cavity is modeled separately. Given that the left and right nasal cavities of a patient may have physiological or pathological differences, such differences may cause asymmetry in the airflow velocity data of the two nasal cavities. If the airflow velocity data of the left and right noses differ significantly, the independently modeled respiratory dynamics maps of the left and right noses may not consistently reflect the child's respiratory characteristics, which may affect the accurate diagnosis of the child's adenoid hypertrophy.

[0008] Second, to overcome the shortcomings of independently modeling the left and right nasal respiratory dynamics, a possible solution is to extract the pathological respiratory features that reflect adenoid hypertrophy from both respiratory dynamics. However, there is currently a lack of an effective method to extract these consistent features from the respiratory dynamics. Summary of the Invention

[0009] In order to overcome the shortcomings of the above-mentioned existing technologies, the present invention provides a system and method for auxiliary detection of adenoids hypertrophy based on dynamic graph fusion. By obtaining the airflow velocity data of the left and right nasal cavities and modeling the respiratory dynamic graphs respectively, and fusing the independently modeled respiratory dynamic graphs, and then extracting the features of the fused respiratory dynamic graphs, the accuracy of auxiliary detection of adenoids hypertrophy is improved.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions:

[0011] In a first aspect, the present invention provides a system for assisting detection of adenoid hypertrophy based on kinetic image fusion, the system comprising:

[0012] A preprocessing module is used to obtain the respiratory airflow velocity signals in the left and right nasal cavities of the test subject and perform preprocessing;

[0013] A respiratory dynamics graph module is used to construct discrete-time nonlinear systems based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities. Based on deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear systems are locally modeled using an RBF neural network. The modeled nonlinear dynamics data and the corresponding nasal airflow velocity data are combined to form respiratory dynamics trajectories to obtain respiratory dynamics graphs for the left and right noses.

[0014] A respiratory dynamics graph fusion module is used to extract nonlinear dynamics data from the test subject's left and right nasal respiratory dynamics graphs, respectively, construct a weight matrix based on the intensity of RBF neural network neuron activation corresponding to the left and right nasal cavities, and use the weight matrix to fuse the nonlinear dynamics data extracted from the left and right nasal respiratory dynamics graphs to obtain a fused respiratory dynamics graph;

[0015] The output module is used to obtain auxiliary detection results of adenoid hypertrophy according to the morphology of the fused respiratory dynamics graph.

[0016] In a second aspect, the present invention provides a method for auxiliary detection of adenoid hypertrophy based on kinetic image fusion, the method comprising:

[0017] Obtain the respiratory airflow velocity signals in the left and right nasal cavities of the tester and perform preprocessing;

[0018] Based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities, discrete-time nonlinear systems are constructed separately. Based on deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear systems are locally modeled using an RBF neural network. The modeled nonlinear dynamic data and the corresponding nasal airflow velocity data are combined to form respiratory dynamics trajectories to obtain respiratory dynamics maps for the left and right noses.

[0019] Extracting nonlinear dynamic data from the test subject's left and right nasal respiratory dynamics maps respectively, constructing a weight matrix based on the intensity of RBF neural network neuron activation corresponding to the left and right nasal cavities, and fusing the nonlinear dynamic data extracted from the left and right nasal respiratory dynamics maps using the weight matrix to obtain a fused respiratory dynamics map;

[0020] According to the morphology of the fused respiratory dynamics graph, the auxiliary detection results of adenoid hypertrophy are obtained.

[0021] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed by the processor, wherein when the computer instructions are executed by the processor, the following steps are performed:

[0022] Obtain the respiratory airflow velocity signals in the left and right nasal cavities of the tester and perform preprocessing;

[0023] Based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities, discrete-time nonlinear systems are constructed separately. Based on deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear systems are locally modeled using an RBF neural network. The modeled nonlinear dynamic data and the corresponding nasal airflow velocity data are combined to form respiratory dynamics trajectories to obtain respiratory dynamics maps for the left and right noses.

[0024] Extracting nonlinear dynamic data from the test subject's left and right nasal respiratory dynamics maps respectively, constructing a weight matrix based on the intensity of RBF neural network neuron activation corresponding to the left and right nasal cavities, and fusing the nonlinear dynamic data extracted from the left and right nasal respiratory dynamics maps using the weight matrix to obtain a fused respiratory dynamics map;

[0025] According to the morphology of the fused respiratory dynamics graph, the auxiliary detection results of adenoid hypertrophy are obtained.

[0026] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the following steps are performed:

[0027] Obtain the respiratory airflow velocity signals in the left and right nasal cavities of the tester and perform preprocessing;

[0028] Based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities, discrete-time nonlinear systems are constructed separately. Based on deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear systems are locally modeled using an RBF neural network. The modeled nonlinear dynamic data and the corresponding nasal airflow velocity data are combined to form respiratory dynamics trajectories to obtain respiratory dynamics maps for the left and right noses.

[0029] Extracting nonlinear dynamic data from the test subject's left and right nasal respiratory dynamics maps respectively, constructing a weight matrix based on the intensity of RBF neural network neuron activation corresponding to the left and right nasal cavities, and fusing the nonlinear dynamic data extracted from the left and right nasal respiratory dynamics maps using the weight matrix to obtain a fused respiratory dynamics map;

[0030] According to the morphology of the fused respiratory dynamics graph, the auxiliary detection results of adenoid hypertrophy are obtained.

[0031] One or more of the above technical solutions have the following beneficial effects:

[0032] In the present invention, based on the test subject's left and right nasal respiratory airflow velocity data, a learning method is determined to model the unknown nonlinear dynamics inherent in the respiratory airflow velocity signals, resulting in left and right respiratory dynamics maps. A weight matrix is ​​then constructed to fuse the left and right respiratory dynamics maps, and adenoids hypertrophy is diagnosed based on the morphological characteristics of the fused respiratory dynamics maps. Due to possible physiological or pathological differences between the left and right nasal cavities, asymmetry in the nasal airflow velocity data can occur. This asymmetry is introduced into the respiratory dynamics map as the respiratory dynamics are modeled, ultimately affecting the accuracy of adenoids hypertrophy detection. The constructed weight matrix can effectively reflect the pathological respiratory characteristics of adenoids hypertrophy, mitigate the effects of this data asymmetry, and improve the diagnostic accuracy of adenoids hypertrophy and its adaptability to adenoids hypertrophy.

[0033] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0035] Figure 1 This is an overall block diagram of an auxiliary detection system for adenoids hypertrophy in Example 1 of the present invention;

[0036] Figure 2 is the respiratory airflow velocity signal after preprocessing in the first embodiment of the present invention;

[0037] Figure 3 This is an example diagram of right nasal respiratory dynamics of a patient with adenoid hypertrophy and differences in the left and right nasal cavities in Example 1 of the present invention;

[0038] Figure 4 This is an example diagram of the respiratory dynamics of the left nose of a patient with adenoid hypertrophy and differences in the left and right nasal cavities in Example 1 of the present invention;

[0039] Figure 5 This is an example diagram of the right nasal respiratory dynamics after fusion for a patient with adenoid hypertrophy and differences in the left and right nasal cavities in Example 1 of the present invention;

[0040] Figure 6 This is an example diagram of the fused left nasal respiratory dynamics of a patient with adenoid hypertrophy and differences in the left and right nasal cavities in Example 1 of the present invention. DETAILED DESCRIPTION

[0041] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0042] It should be noted that the terms used herein are for describing particular embodiments only and are not intended to limit the exemplary embodiments according to the present invention.

[0043] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0044] All data in this embodiment is obtained in compliance with laws and regulations and based on the consent of the user, and is used legally.

[0045] Based on research on the continuous excitation properties of radial basis function (RBF) neural networks, Wang Cong proposed deterministic learning theory in his book, Deterministic Learning Theory for Identification, Recognition, and Control, published in 2007. This theory utilizes radial basis function (RBF) neural networks to model the unknown nonlinear dynamics in nonlinear systems, while satisfying continuous excitation conditions. The dynamically time-varying nonlinear dynamics are accurately modeled and stored locally as constant-valued RBF networks in a time-invariant and spatially distributed manner. Furthermore, utilizing these accurately modeled constant-valued RBF networks, several downstream tasks for deterministic learning have been proposed and developed. Systems in practical engineering applications have discrete-time characteristics, necessitating the expansion of deterministic learning theory to meet these requirements. In the Chinese invention patent application: Rapid diagnosis method of axial compressor rotating stall based on deterministic learning theory (application number CN101887479A), a scheme for applying deterministic learning theory in aircraft engine surge detection is proposed; A myocardial ischemia auxiliary detection method based on deterministic learning theory (application number CN103549949A) proposes a scheme for applying deterministic learning theory in early screening of myocardial ischemia; A perspective-independent gait recognition method based on deterministic learning theory (application number CN104134077A) proposes a scheme for applying deterministic learning theory in gait recognition; and the deterministic learning theory is expanded in the field of temporal data modeling.

[0046] Example 1

[0047] When the respiratory airflow passes through the nasal cavity, factors such as the flow rate and pressure of the airflow will be affected by the nasal structure. Therefore, when the nasal structure undergoes pathological changes due to respiratory diseases (such as adenoids hypertrophy in this embodiment), the respiratory airflow signal can well reflect the abnormal changes in the nasal structure, that is, the respiratory airflow signal contains rich dynamic information for characterizing respiratory diseases. However, due to physiological or pathological reasons, the left and right nasal structures of the same subject may be different. This difference may cause the left and right nasal airflow signals of the same subject to show obvious differences, further leading to different diagnostic results for the same respiratory disease. The goal of fusion in this embodiment is to reduce the decrease in diagnostic accuracy caused by the difference in the left and right nasal structures as much as possible, and to improve the sensitivity and adaptability of the respiratory dynamics graph to respiratory diseases - adenoids hypertrophy.

[0048] like Figure 1 As shown, this embodiment discloses an auxiliary detection system for adenoids hypertrophy based on dynamic image fusion, comprising:

[0049] The preprocessing module is used to obtain the respiratory airflow velocity signals in the left and right nasal cavities of the test subject and perform preprocessing.

[0050] In this embodiment, respiratory airflow velocity signals from the left and right nasal cavities of the subject are obtained, filtered using wavelet filtering, and then normalized. A nasal resistance meter can be used to collect respiratory airflow velocity signals from the left and right nasal cavities of the subject.

[0051] The respiratory dynamics graph module is used to construct discrete-time nonlinear systems based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities. Based on the deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear system are locally modeled using an RBF neural network. The modeled nonlinear dynamics data and the corresponding nasal airflow velocity data are combined to form a respiratory dynamics trajectory to obtain the left and right nasal respiratory dynamics graphs.

[0052] In this embodiment, if Figure 2 As shown in Figure 1, a discrete-time nonlinear system is constructed based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities. First, the respiratory system is modeled as the following nonlinear dynamic form:

[0053]

[0054] Where x1 is the respiratory airflow velocity signal, x2 is the rate of change of the respiratory airflow velocity signal, f(x; p s ) is the nonlinear dynamics of the respiratory system, p s is the system parameter vector, x=[x1,x2] T represents the state of the nonlinear system, the superscript T represents the transpose, and y is the system output, which is the respiratory airflow velocity signal of the tester measured by the nasal resistance meter.

[0055] The nonlinear dynamics of formula (1) is subjected to Euler discretization to obtain the discrete-time nonlinear system:

[0056]

[0057] Among them, T s is the sampling period, i.e., the inverse of the sampling frequency, k is the kth sampling moment, x1(x) is the sampling signal of the respiratory airflow velocity signal, x2(k) is the sampling signal of the rate of change of the respiratory airflow velocity signal, and x(k) = [x1(k), x2(k)] T represents the state of the discrete-time nonlinear system, and y(k) is the sampling signal of the tester's respiratory airflow velocity signal measured by the nasal resistance meter.

[0058] Since x2(k) cannot be collected by nasal resistance meter, a high-gain observer is used to obtain it. The high-gain observer calculates the respiratory system state x(k) = [x1(k), x2(k)] T Make an estimate and output This output will constitute the regression trajectory of the respiratory system state in phase space; is the estimated signal obtained by observing the respiratory airflow velocity signal through a high-gain observer, It is an estimated signal obtained by observing the rate of change of the respiratory airflow velocity signal through a high-gain observer.

[0059] The unknown nonlinear dynamics of discrete-time nonlinear systems f(x(k); p s ), using RBF neural network for local accurate modeling.

[0060] First, the regression trajectory of the respiratory system state Input into the RBF neural network, and then use the identifier and network weight update rate in the learning mechanism to obtain f(x(k); p s ) is accurately modeled by the neural network as follows:

[0061]

[0062] in, is a constant vector representing the network weights, Represents the RBF regression subvector, N represents the number of neurons in the RBF neural network, ∈ * Modeling error for a sufficiently small network.

[0063] In this embodiment, the respiratory dynamics trajectory refers to The three-dimensional space curve composed of the respiratory dynamics trajectory is visualized to obtain the respiratory dynamics diagram.

[0064] The respiratory dynamics graph fusion module is used to extract the nonlinear dynamics data from the left and right nasal respiratory dynamics graphs of the tester, namely, the constant RBF neural network model, and construct a weight matrix according to the intensity of the activation of the RBF neural network neurons corresponding to the left and right nasal cavities. The weight matrix is ​​used to fuse the nonlinear dynamics data extracted from the left and right nasal respiratory dynamics graphs, namely, the constant RBF neural network, to obtain the fused respiratory dynamics graph.

[0065] In this embodiment, the nonlinear dynamic data of the left and right nasal respiratory dynamics diagrams are extracted, that is, the constant RBF neural network model is built, that is, the respiratory dynamics trajectory of the left nose is extracted from the left and right nasal respiratory dynamics diagrams. and respiratory dynamics of the right nose The subscripts L and R represent left and right, respectively. Because a subject actually needs to collect two respiratory airflow signals from the left and right nostrils, and considering that the left and right nasal cavities of the subject may have physiological or pathological differences, the subscripts L and R are introduced to distinguish them.

[0066] The weight matrix is ​​constructed based on the intensity of neuron activation in the RBF neural network, specifically:

[0067] First, the RBF regression matrices of the left and right nasal respiratory dynamics trajectories are concatenated to obtain

[0068]

[0069] 1≤k≤M, k is the kth data point. and Similarly, M, M′ refer to the data length of the collected left and right nasal airflow signals respectively. In order to facilitate the understanding of the data length, it is assumed that the sampling period T s =0.01s to sample the left (or right) nasal respiratory airflow signal of a tester, and the sampling time is 10s. The sampling data length of the respiratory airflow signal obtained is M or M ′ =1000.

[0070] Next, calculate Take its diagonal elements to construct a diagonal matrix Get the weight matrix Its i-th diagonal element is Characterizes the intensity of activation of the i-th neuron in the RBF neural network, and s(x) is a Gaussian function.

[0071] Second, consider the respiratory dynamics after fusion as f F (x(k);p F ), which can be modeled by the RBF neural network as follows:

[0072]

[0073] Here p F is the respiratory dynamics after fusion F (x(k);p F ) parameter vector, x(k) belongs to f F (x(k);p F ) has no specific meaning. The value range of x(k) is very wide. Since only the respiratory airflow signals of the left and right nasal cavities of a test subject are collected, x(k) is replaced by and Indicates that the RBF network is used to approximate fF (x(k);p F ) is the weight of the neuron, and N represents the number of neurons in the RBF network.

[0074] The dynamic distance D between the fused respiratory dynamics and the respiratory dynamics of the left and right noses is as follows:

[0075]

[0076] Among them, ‖·‖ refers to the Euclidean distance norm, SxLk∈RN represents the RBF regression subvector corresponding to the left nasal cavity, represents the RBF regression subvector corresponding to the right nasal cavity, The constant vector corresponding to the left nasal cavity represents the network weight, The constant vector corresponding to the right nasal cavity represents the network weight.

[0077] Using the idea of ​​weighted least squares algorithm, the constructed weight matrix P is introduced into D′ to obtain:

[0078]

[0079] in,

[0080] Introducing the regularization measure into D′ yields:

[0081]

[0082] Among them, λ L and λ R Both are regularization coefficients and are greater than 0.

[0083] Use the least squares algorithm to calculate the optimal solution that minimizes D′ get:

[0084]

[0085] Among them, the superscript + represents the pseudo-inverse of the matrix, and the regularization coefficient λ L ,λ R The purpose is to limit the solution The solution range makes the solution process more stable.

[0086] So far, the respiratory dynamics of fusion have been obtained as

[0087] First, the respiratory dynamics after fusion Maintains breathing dynamics with both noses There is a good dynamic similarity between the two, because the fusion process is based on the respiratory dynamics of the left and right noses, and the fusion result cannot have too much dynamic difference with the respiratory dynamics of the left and right noses. In addition, due to the weighting of the weight matrix P, the respiratory dynamics It includes the respiratory dynamics of both the left and right noses that reflect the pathological respiratory characteristics of adenoid hypertrophy.

[0088] The output module is used to obtain auxiliary detection results based on the morphology of the fused respiratory dynamics graph and the basic information of the tester.

[0089] Among them, the basic information of the tester includes age, medical history and symptoms.

[0090] According to the morphology of the fused respiratory dynamics graph, it refers to observing whether the subject's respiratory dynamics graph is symmetrical and regular or asymmetrical and scattered; when the respiratory dynamics graph appears to be a symmetrical and regular butterfly-shaped graph, it means that the current subject is healthy; when the respiratory dynamics graph appears to be asymmetrical and scattered, it means that the current subject has adenoid hypertrophy.

[0091] according to Figure 3-Figure 6 Before fusion, the respiratory dynamics of the left and right noses differed significantly. The right nose's respiratory dynamics exhibited a distinctly asymmetric and scattered pattern, characteristic of adenoid hypertrophy, while the left nose's respiratory dynamics exhibited a nearly regular, symmetrical butterfly shape. After fusion, both noses exhibited distinctly asymmetric and scattered patterns.

[0092] The present invention uses deterministic learning to accurately model the unknown respiratory dynamics contained in the subject's respiratory airflow signal. The respiratory dynamics map obtained by modeling is used to diagnose the respiratory disease - adenoids hypertrophy. In the diagnosis of adenoids hypertrophy, the respiratory dynamics map of healthy subjects is a symmetrical and regular butterfly shape, while the respiratory dynamics map of patients with adenoids hypertrophy is an asymmetrical and scattered shape. However, for subjects with physiological or pathological differences in the left and right noses, there will be obvious differences in the respiratory dynamics maps of the left and right noses. This difference will seriously affect the accuracy of subsequent diagnosis using the respiratory dynamics map, and may cause misdiagnosis or missed diagnosis of the subject. The fusion method proposed in the present invention extracts the pathological respiratory features of adenoids hypertrophy that are jointly reflected in the respiratory dynamics maps of the left and right noses, making the present invention highly adaptable to adenoids hypertrophy and avoiding the impact of the left and right nose differences on the accuracy of disease diagnosis. In addition, the construction of the weight matrix in the fusion method is not unique and fixed, and it can vary according to different respiratory diseases, thus making the proposed fusion method have a wider range of applicability. In summary, the present invention is non-invasive, radiation-free, easy to operate and accurate, and is particularly suitable for dynamic evaluation of adenoid hypertrophy, and can provide medical professionals with accurate and objective diagnostic results and reports.

[0093] Example 2

[0094] The purpose of this embodiment is to provide a method for assisting the detection of adenoid hypertrophy based on kinetic image fusion, including:

[0095] Obtain the respiratory airflow velocity signals in the left and right nasal cavities of the tester and perform preprocessing;

[0096] Based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities, discrete-time nonlinear systems are constructed separately. Based on deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear systems are locally modeled using an RBF neural network. The modeled nonlinear dynamic data and the corresponding nasal airflow velocity data are combined to form respiratory dynamics trajectories to obtain respiratory dynamics maps for the left and right noses.

[0097] Extracting nonlinear dynamic data from the test subject's left and right nasal respiratory dynamics maps respectively, constructing a weight matrix based on the intensity of RBF neural network neuron activation corresponding to the left and right nasal cavities, and fusing the nonlinear dynamic data extracted from the left and right nasal respiratory dynamics maps using the weight matrix to obtain a fused respiratory dynamics map;

[0098] According to the morphology of the fused respiratory dynamics graph, the auxiliary detection results of adenoid hypertrophy are obtained.

[0099] In further embodiments, there is also provided:

[0100] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed by the processor. When the computer instructions are executed by the processor, the method described in Example 1 is performed. For the sake of brevity, no further details are given here.

[0101] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), off-the-shelf field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0102] The memory may include a read-only memory and a random access memory, and provides instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0103] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in embodiment 1 is performed.

[0104] The method in Example 1 can be directly implemented as being executed by a hardware processor, or by a combination of hardware and software modules within the processor. The software module can be located in a storage medium well-established in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not given here.

[0105] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0106] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions contained in program modules, which are executed in a device on a real or virtual processor of a target to perform the process / method described above. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided between program modules as needed. The machine-executable instructions for the program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0107] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the computer or other programmable data processing device, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0108] In the context of the present invention, computer program code or related data can be carried by any appropriate carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals include electrical, optical, radio, acoustic, or other forms of propagated signals, such as carrier waves, infrared signals, and the like.

[0109] Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0110] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A system for assisting the detection of adenoid hypertrophy based on dynamic image fusion, characterized in that: The system comprises: A preprocessing module is used to obtain the respiratory airflow velocity signals in the left and right nasal cavities of the test subject and perform preprocessing; A respiratory dynamics graph module is used to construct discrete-time nonlinear systems based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities. Based on deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear systems are locally modeled using an RBF neural network. The modeled nonlinear dynamics data and the corresponding nasal airflow velocity data are combined to form respiratory dynamics trajectories to obtain respiratory dynamics graphs for the left and right noses. A respiratory dynamics graph fusion module is used to extract nonlinear dynamics data from the test subject's left and right nasal respiratory dynamics graphs, respectively, construct a weight matrix based on the intensity of RBF neural network neuron activation corresponding to the left and right nasal cavities, and use the weight matrix to fuse the nonlinear dynamics data extracted from the left and right nasal respiratory dynamics graphs to obtain a fused respiratory dynamics graph; The output module is used to obtain auxiliary detection results of adenoid hypertrophy according to the morphology of the fused respiratory dynamics graph.

2. The adenoids hypertrophy auxiliary detection system based on dynamic image fusion according to claim 1, characterized in that: The respiratory dynamics diagram module includes: A modeling unit is used to input the regression trajectory of the respiratory system state into a constant-valued RBF neural network, and locally model the unknown nonlinear dynamics of the discrete-time nonlinear system by using the identifier and the network weight update rate in the determined learning mechanism; The visualization unit is used to visualize the respiratory dynamics trajectory composed of the modeled nonlinear dynamics data and the corresponding nasal airflow velocity data to obtain a respiratory dynamics graph.

3. The adenoids hypertrophy auxiliary detection system based on dynamic image fusion according to claim 1, characterized in that: The respiratory dynamics graph fusion module includes: A construction unit is used to splice the RBF regression matrices in the left and right nasal respiratory dynamics trajectories and construct a weight matrix based on the spliced ​​matrices; A calculation unit for constructing a dynamic distance between the fused respiratory dynamics and the left and right nasal respiratory dynamics; The solving unit is used to introduce the weight matrix by using a weighted least squares algorithm, solve the problem with the goal of minimizing the dynamic distance, and obtain a fused respiratory dynamics graph.

4. The adenoids hypertrophy auxiliary detection system based on dynamic image fusion according to claim 3, characterized in that: The construction unit comprises: A splicing subunit is used to splice the RBF regression matrices in the left and right nasal respiratory dynamics trajectories to obtain a splicing matrix; a calculation subunit, configured to calculate the product of the splicing matrix and the transposed matrix of the splicing matrix to obtain a product matrix; The construction subunit is used to extract the diagonal elements of the product matrix to construct a diagonal matrix to obtain a weight matrix.

5. The adenoids hypertrophy auxiliary detection system based on dynamic image fusion according to claim 3, characterized in that: The solving unit includes: An introducing subunit is used to introduce the weight matrix and regularization into the dynamic distance using a weighted least squares algorithm; The solving subunit is used to solve the problem by using the least squares algorithm with the goal of minimizing the dynamic distance to obtain a fused respiratory dynamics graph.

6. The adenoids hypertrophy auxiliary detection system based on dynamic image fusion according to claim 1, characterized in that: The preprocessing module includes: The filtering subunit is used to process the acquired respiratory airflow velocity signals in the left and right nasal cavities of the test subject using wavelet filtering; The normalization subunit is used to perform normalization processing on the filtered respiratory airflow velocity signal.

7. The adenoids hypertrophy auxiliary detection system based on dynamic image fusion according to claim 1, characterized in that: The output module includes: The first output subunit is configured to present a symmetrical and regular butterfly-shaped graph based on the fused respiratory dynamics graph, indicating that the current test subject is healthy; The second output subunit is used to indicate that the current test subject is a patient with adenoid hypertrophy when asymmetric dispersion is presented according to the fused respiratory dynamics graph.

8. A computer-readable storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, complete the following steps: Obtain the respiratory airflow velocity signals in the left and right nasal cavities of the tester and perform preprocessing; Based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities, discrete-time nonlinear systems are constructed separately. Based on deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear systems are locally modeled using an RBF neural network. The modeled nonlinear dynamic data and the corresponding nasal airflow velocity data are combined to form respiratory dynamics trajectories to obtain respiratory dynamics maps for the left and right noses. Extracting nonlinear dynamic data from the test subject's left and right nasal respiratory dynamics maps respectively, constructing a weight matrix based on the intensity of RBF neural network neuron activation corresponding to the left and right nasal cavities, and fusing the nonlinear dynamic data extracted from the left and right nasal respiratory dynamics maps using the weight matrix to obtain a fused respiratory dynamics map; According to the morphology of the fused respiratory dynamics graph, the auxiliary detection results of adenoid hypertrophy are obtained.

9. An electronic device, characterized in that: The system includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the following steps are accomplished: Obtain the respiratory airflow velocity signals in the left and right nasal cavities of the tester and perform preprocessing; Based on the preprocessed respiratory airflow velocity signals in the left and right nasal cavities, discrete-time nonlinear systems are constructed separately. Based on deterministic learning theory, the unknown nonlinear dynamics of the discrete-time nonlinear systems are locally modeled using an RBF neural network. The modeled nonlinear dynamic data and the corresponding nasal airflow velocity data are combined to form respiratory dynamics trajectories to obtain respiratory dynamics maps for the left and right noses. Extracting nonlinear dynamic data from the test subject's left and right nasal respiratory dynamics maps respectively, constructing a weight matrix based on the intensity of RBF neural network neuron activation corresponding to the left and right nasal cavities, and fusing the nonlinear dynamic data extracted from the left and right nasal respiratory dynamics maps using the weight matrix to obtain a fused respiratory dynamics map; According to the morphology of the fused respiratory dynamics graph, the auxiliary detection results of adenoid hypertrophy are obtained.

Images