Dysphagia screening system and screening method

Abstract

The invention relates to the technical field of computer-aided diagnosis, in particular to a dysphagia screening system and method, and the system comprises a three-dimensional coordinate collection module, a space trajectory analysis module, a frequency domain gravity center extraction module, a bimodal fusion module and an anomaly verification module. According to the method, non-physiological trajectory offset is captured through three-dimensional displacement vector difference and vector included angle calculation, sudden change is detected in combination with myoelectricity frequency domain energy gravity center first-order difference, space fracture and frequency domain drift are synchronously tracked, energy gravity center quantization frequency dominant characteristic changes are extracted through short-time Fourier transform, and the defect that amplitude parameter sensitivity is insufficient is overcome; two-channel event indexes are aligned with a time axis, cross-modal events synchronously verify space-time coupling of swallowing interruption, the single-dimensional misjudgment probability is reduced, a trigger threshold value is set through reverse tracking of multi-muscle-group cooperative offset, local interference and systematic imbalance are distinguished, and the compensatory dysfunction judgment specificity is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer - aided diagnosis, and particularly to a dysphagia screening system and a screening method. Background Art

[0002] The technical field of computer - aided diagnosis includes various methods and systems that use computer technology to improve the accuracy and efficiency of the medical diagnosis process. The core of this technical field is to assist medical staff in making a preliminary judgment and auxiliary evaluation of diseases through technical means such as artificial intelligence image recognition algorithms, clinical big data analysis, and decision - making support systems. In practical applications, computer - aided diagnosis often covers aspects such as medical image processing, physiological signal analysis, electronic medical record information extraction and reasoning analysis, and plays an important role especially in scenarios such as chronic disease screening, early diagnosis, and treatment plan recommendation. With the rapid growth of the types and quantities of medical data, computer - aided diagnosis has gradually formed a systematic technical system of interdisciplinary integration, including the integrated application of multiple branches such as medical engineering, computer science, and bioinformatics.

[0003] Among them, a dysphagia screening system refers to a type of computer - aided screening tool used for non - invasive and rapid assessment of an individual's swallowing function and determination of the potential risk of dysphagia. The technical matters targeted by this system mainly cover the acquisition of swallowing movements, the synchronous acquisition of swallowing sound signals and neck electromyogram signals, the extraction of quantitative indicators based on time - series data, and the determination of swallowing status. Specifically, wearable sensors are used to collect the sound signals and electrophysiological signals during the user's swallowing process, and then feature extraction methods are used to construct a set of swallowing frequency, duration, and interval feature parameters. Combined with a machine - learning model, the samples are classified and identified to complete the dysphagia screening process.

[0004] The existing technology relies on the amplitude feature extraction of swallowing sounds and electromyogram signals, is limited to the single - dimensional analysis of time - series parameters (such as duration and interval), and cannot quantify the continuity deviation of the coordinated movement trajectory of muscle groups or the non - linear drift of the frequency - domain energy center of gravity. For example, a small angular deviation in the displacement trajectory of the laryngeal body or a sudden change in the energy distribution of the electromyogram frequency band may be ignored, resulting in missed detection of the failure of swallowing compensation caused by early neurological control abnormalities. Single - modality signal analysis is easily interfered by sensor noise or individual physiological differences. For example, fluctuations in electromyogram amplitude may be misjudged as swallowing abnormalities, but actually it is a normal muscle fatigue phenomenon. The existing methods lack a cross - modality event alignment mechanism and cannot verify the spatio - temporal correlation of abnormal events. For example, when the abnormal displacement of the laryngeal body and the frequency - domain mutation of the cricopharyngeal muscle occur simultaneously, single - dimensional analysis cannot identify the cascade effect. The existing technology does not construct a reverse tracking process for the coordinated offset of multiple muscle groups, making it difficult to distinguish local abnormalities from systemic coordination imbalances and affecting the clinical applicability of the screening results. Summary of the Invention

[0005] The object of the present invention is to solve the drawbacks existing in the prior art, and a dysphagia screening system and screening method are proposed.

[0006] To achieve the above object, the present invention adopts the following technical solutions: A dysphagia screening system includes: A three-dimensional coordinate acquisition module, which is used to acquire the three-dimensional space coordinate sequences of the hyoid bone, laryngeal body and cricopharyngeal part through body surface marker points, generate a displacement vector set based on fixed interval differences, and transmit the three-dimensional space coordinate sequences and the displacement vector set to the space trajectory analysis module; A space trajectory analysis module, which is used to call the displacement vector set, perform spatial angle calculation on adjacent displacement vectors, judge the direction consistency through cross product, generate spatial continuity break points, and transmit the spatial continuity break points and the three-dimensional space coordinate sequences to the dual-modal fusion module; A frequency domain centroid extraction module, which is used to perform short-time Fourier transform on the original electromyogram signal to generate a frequency domain energy spectrum map, calculate the energy weighted mean value of 0-100Hz to obtain an energy centroid time series, detect the frequency centroid mutation points with the first-order difference exceeding twice the standard deviation, and transmit the energy centroid time series and the frequency centroid mutation points to the dual-modal fusion module; A dual-modal fusion module, which is used to receive the spatial continuity break points and the frequency centroid mutation points, establish a dual-channel event index mapping table to align the time axis, generate activation interruption event markers, and transmit the activation interruption event markers to the anomaly verification module.

[0007] As a further solution of the present invention, the three-dimensional space coordinate sequences include hyoid bone coordinates, laryngeal body coordinates, and cricopharyngeal part coordinates. The displacement vector set is specifically a time interval difference vector, a three-dimensional direction component, and a displacement modulus. The spatial continuity break points include an angle overrun marker, a direction difference marker, and a timestamp index. The energy centroid time series specifically refers to a frequency band weighted mean value, a time window energy distribution, and a centroid offset. The frequency centroid mutation points include a mutation amplitude threshold, a drift direction identifier, and a duration count. The dual-channel event index mapping table includes a time window index, a spatial break event ID, and a frequency domain mutation event ID. The activation interruption event marker is specifically a synchronous trigger flag, a spatial-frequency domain correlation weight, and an anomaly confidence score.

[0008] As a further solution of the present invention, the three-dimensional coordinate acquisition module includes: A coordinate sequence acquisition sub-module acquires the three-dimensional coordinate data of the hyoid bone, laryngeal body and cricopharyngeal part through body surface marker points, records them in chronological order as three-dimensional space coordinate sequences, and stores them in a matrix structure aligned with timestamps; The differential time series processing sub-module calls the three-dimensional space coordinate sequence, extracts the horizontal displacement of the hyoid bone marker point, the vertical displacement of the laryngeal body, and the depth displacement of the cricopharyngeal part, calculates the mixed displacement feature quantity, and generates a differential time series set in chronological order; The displacement vector generation sub-module extracts the three-dimensional direction components of the differential scalar values at multiple time points based on the differential time series set, combines them in a vector form after normalization, and generates a displacement vector set.

[0009] As a further solution of the present invention, the space trajectory analysis module includes: The vector angle calculation sub-module calls the displacement vector set, extracts adjacent displacement vectors, calculates the spatial angle between adjacent vectors based on the dot product and modulus of the vectors, and generates a spatial angle sequence; The direction consistency judgment sub-module calls the spatial angle sequence, performs a cross product operation on each pair of adjacent displacement vectors, and uses the formula: ; Performs an operation to obtain a direction consistency index, compares the index with a preset dynamic threshold, and generates a direction consistency index set; Among them, represents adjacent displacement vectors, represents the time interval, represents the spatial angle, is the direction consistency index; The breakpoint screening sub-module calls the direction consistency index set and the spatial angle sequence, selects the positions where the index is lower than the threshold and the angle is greater than the critical angle, and merges adjacent abnormal points to generate spatial continuity breakpoints.

[0010] As a further solution of the present invention, the frequency domain centroid extraction module includes: The frequency domain conversion sub-module performs a short-time Fourier transform on the original EMG signal, divides the signal according to a fixed time window, calculates the complex spectral amplitudes of multiple frequency points in the 0-100 Hz frequency band within each window, and generates a frequency domain energy map; The energy centroid sequence generation sub-module calls the frequency domain energy map, extracts the energy values of the frequency points within multiple time windows, and uses the formula: ; Calculates the weighted mean of the frequency energy within each window, splices the calculation results of adjacent windows in chronological order, and generates an energy centroid sequence; Among them, represents the energy centroid value of the th time window, is the frequency value corresponding to the th frequency point, is the th window and the The complex spectral amplitude of a frequency point is the adjustment factor for the total energy difference between adjacent windows is the absolute value of the change in the total energy of the previous window is the total number of discrete frequency points within the frequency band The mutation detection sub-module calculates the first-order difference sequence based on the energy centroid sequence, statistically calculates the standard deviation of the difference sequence, screens the data points where the difference value exceeds 2 times the standard deviation of the difference sequence, and marks the corresponding time points as frequency centroid mutation points

[0011] As a further solution of the present invention, the dual-modal fusion module includes The event index alignment sub-module obtains the spatial continuity break points and the frequency centroid mutation points, establishes a dual-channel event index mapping table, and simultaneously calls the time axis alignment strategy to perform cross-modal matching on the timing coordinates of the break points and mutation points, screens the synchronous event pairs with the absolute value of the time difference less than the set tolerance threshold, and generates a time synchronization index table The interruption activation determination sub-module extracts the break strength value and mutation amplitude of the synchronous event pair based on the time synchronization index table, and uses the formula ; Calculate the cross-modal event correlation weight value, compare the weight value with the dynamic determination threshold. If three consecutive groups of weight values exceed the threshold and show an increasing trend, it is determined as an effective activation interruption event, and an interruption activation weight column is generated Among them represents the cross-modal event correlation weight value between the th spatial continuity break point and the th frequency centroid mutation point represents the strength value of the th spatial continuity break point represents the amplitude of the th frequency centroid mutation point represents the maximum value of the spatial break point strength value and the frequency mutation point amplitude is the absolute value of the time difference between the two event points represents the modal difference compensation coefficient The event marking generation sub-module calls the time synchronization index table and the interruption activation weight column, performs weighted fusion on the spatio-temporal coordinates of the cross-modal event pairs, eliminates redundant markings, and generates activation interruption event markings covering the time axis

[0012] As a further solution of the present invention, the system further includes Anomaly verification module, configured to perform reverse retrieval of the displacement vector set and the energy center of gravity time series of the associated muscle groups based on the activation interruption event marker, and compare the multi-node synchronous offsets to generate a cascaded interruption determination result; The cascaded interruption determination result includes the muscle group chain node number, the synchronous offset count, and the systemic disorder level.

[0013] As a further solution of the present invention, the anomaly verification module includes: The event marker parsing sub-module obtains the activation interruption event marker, extracts the event occurrence timestamp and the associated node position coordinates, calls the preset node displacement sampling interval parameter, aligns the timestamp with the energy center of gravity acquisition period, and generates an event alignment time series; The muscle group vector retrieval sub-module retrieves the muscle group displacement vector set within the corresponding time interval based on the event alignment time series, performs vector superposition operations on the three-dimensional coordinate data of each node, and generates the energy center of gravity mean value of multiple nodes; The cascaded offset determination sub-module calls the muscle group displacement vector set and the energy center of gravity mean value, calculates the standard deviation of the multi-node displacement vector and the energy center of gravity, compares the preset synchronous offset threshold, and generates a cascaded interruption determination coefficient when the number of nodes exceeding the threshold reaches 70% of the total number of nodes.

[0014] A swallowing disorder screening method, which is performed based on the above-mentioned swallowing disorder screening system, and includes the following steps: S1: Obtain the three-dimensional spatial coordinate sequences of the hyoid bone, laryngeal body, and cricopharyngeal part during swallowing activities through the body surface marker point signal acquisition device, and perform differential operations on the three-dimensional spatial coordinate sequences based on a fixed time interval to generate a displacement vector set; S2: Call the displacement vector set, perform spatial angle calculations on adjacent displacement vectors, and judge the direction consistency through vector cross product operations. When the angle exceeds 15 degrees and the cross product direction difference triggers a set threshold, generate a spatial continuity break point; S3: Perform short-time Fourier transform on the original electromyogram signal of the target muscle group to generate a frequency domain energy distribution map, calculate the weighted average based on the energy values in the 0-100 Hz frequency band to generate an energy center of gravity time series, and detect the frequency center of gravity mutation point where the first-order difference exceeds twice the standard deviation; S4: Establish a two-channel event index mapping table to align the time axes of the spatial continuity break point and the frequency center of gravity mutation point, and generate an activation interruption event marker when the two are synchronously triggered within the same time window; S5: Based on the activation interruption event marker, perform reverse retrieval of the displacement vector set and the energy center of gravity time series of the associated muscle groups, and compare the synchronous offset characteristics of multiple muscle group nodes in adjacent time periods to generate a cascaded interruption determination result.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, through the calculation of the three-dimensional space displacement vector difference and the included angle between vectors at consecutive time points, the non-physiological deviation of the muscle group movement trajectory is captured. Combining the first-order difference mutation detection of the myoelectric frequency-domain energy centroid time series, the spatial trajectory break and the frequency-domain energy drift events are synchronously tracked. The frequency-domain energy centroid parameter extracted by the short-time Fourier transform quantifies the dynamic change of the frequency-dominant characteristics during the muscle group activation process, overcoming the insufficient sensitivity of the amplitude parameter to abnormal nerve control. The two-channel event index mapping table realizes the time-axis alignment of the spatial break point and the frequency-domain mutation point, and verifies the spatio-temporal coupling characteristics of the swallowing action interruption through the cross-modal event synchronization trigger mechanism, reducing the probability of single-dimensional misjudgment. Backtracking the collaborative offset characteristics of multiple muscle groups and setting the synchronization trigger quantity threshold to distinguish local signal interference from systematic coordination imbalance, and improving the discrimination specificity for compensatory dysfunction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is the system flow chart of the present invention; Figure 2 is the flow chart of the three-dimensional coordinate acquisition module of the present invention; Figure 3 is the flow chart of the spatial trajectory analysis module of the present invention; Figure 4 is the flow chart of the frequency-domain centroid extraction module of the present invention; Figure 5 is the flow chart of the dual-modal fusion module of the present invention; Figure 6 is the flow chart of the abnormality verification module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0018] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0019] Embodiment 1 Please refer to Figure 1 , the present invention provides a technical solution: a dysphagia screening system includes: A three-dimensional coordinate acquisition module, which is used to collect the three-dimensional spatial coordinate sequences of the hyoid bone, laryngeal body and cricopharyngeal part through body surface marking points, generate a displacement vector set based on fixed interval differences, and transmit the three-dimensional spatial coordinate sequences and the displacement vector set to the spatial trajectory analysis module; A spatial trajectory analysis module, which is used to call the displacement vector set, perform spatial angle calculations on adjacent displacement vectors, judge the direction consistency through cross products, generate spatial continuity break points, and transmit the spatial continuity break points and the three-dimensional spatial coordinate sequences to the dual-modal fusion module; A frequency domain centroid extraction module, which is used to perform short-time Fourier transform on the original electromyogram signal to generate a frequency domain energy spectrum diagram, calculate the energy weighted mean value of 0-100Hz to obtain an energy centroid time series, detect the frequency centroid mutation points with the first-order difference exceeding twice the standard deviation, and transmit the energy centroid time series and the frequency centroid mutation points to the dual-modal fusion module; A dual-modal fusion module, which is used to receive the spatial continuity break points and the frequency centroid mutation points, establish a dual-channel event index mapping table to align the time axis, generate activation interruption event markers, and transmit the activation interruption event markers to the anomaly verification module; An anomaly verification module, which is used to retrieve the displacement vector set and the energy centroid time series of the associated muscle groups in reverse based on the activation interruption event markers, and compare the multi-node synchronous offsets to generate a cascaded interruption determination result.

[0020] The three-dimensional spatial coordinate sequences include hyoid bone coordinates, laryngeal body coordinates, and cricopharyngeal part coordinates. The displacement vector set specifically includes time interval difference vectors, three-dimensional direction components, and displacement modulus lengths. The spatial continuity break points include angle overrun markers, direction difference markers, and timestamp indexes. The energy centroid time series specifically refers to frequency band weighted mean values, time window energy distributions, and centroid offset amounts. The frequency centroid mutation points include mutation amplitude thresholds, drift direction identifiers, and duration counts. The dual-channel event index mapping table includes time window indexes, spatial break event IDs, and frequency domain mutation event IDs. The activation interruption event markers specifically include synchronous trigger flags, spatial-frequency domain association weights, and anomaly confidence scores. The cascaded interruption determination result includes muscle group chain node numbers, synchronous offset counts, and systematic disorder levels.

[0021] Please refer to Figure 2 , the three-dimensional coordinate acquisition module includes: The coordinate sequence acquisition sub-module collects the three-dimensional coordinate data of the hyoid bone, laryngeal body and cricopharyngeal part through body surface marking points, records them in chronological order as three-dimensional spatial coordinate sequences, and stores them in a matrix structure aligned with timestamps; This process first precisely pastes optical reflection marker points at specific anatomical positions on the subject's neck. The specific positions include the midpoint of the body of the hyoid bone, the midpoint of the upper edge of the thyroid cartilage (representing the position of the laryngeal body), and the bilateral arytenoid joint processes (taking the average of the coordinates on both sides to represent the center position of the cricopharyngeal region). Subsequently, a motion capture system consisting of at least three high-speed infrared cameras is activated. This system operates at a sampling frequency of 100 Hz, that is, the position of the marker points is precisely recorded every 0.01 seconds. The cameras simultaneously capture the movement trajectories of each marker point in space. The system software uses the triangulation algorithm to reconstruct the three-dimensional spatial coordinates of each marker point in real time based on the two-dimensional image information provided by multiple cameras. , the coordinate system is defined as: The x-axis points in the horizontal direction (left and right), the y-axis points in the vertical direction (up and down), the z-axis points in the depth direction (front and back). The system assigns a high-precision timestamp to the coordinate data captured in each frame. This timestamp is precisely synchronized with the system's internal clock or an external synchronization signal to ensure a time resolution of the millisecond level. During a complete swallowing motion (the duration is generally 1.5 seconds to 2.5 seconds, here 2.0 seconds are recorded, a total of 200 frames of data), the coordinates of each marker point are continuously recorded. For example, the hyoid bone coordinates at time t1 , the laryngeal body coordinates , the cricopharyngeal region coordinates , and the corresponding coordinates at time t2.

[0022] After the acquisition is completed, the three-dimensional coordinate data of all marker points at each time point are strictly organized in chronological order to form a matrix structure aligned with timestamps. Each row of this matrix represents a timestamp , and the columns contain the coordinate values of all marker points at that moment, in millimeters (mm). The data in the first row corresponds to , containing the coordinates of the hyoid bone, laryngeal body, and cricopharyngeal region at this moment. The second row corresponds to the data at time t2, and so on, until . Table 1 shows a partial data segment of this matrix. The complete matrix is finally stored as a digital file in a specific format (such as.csv or.mat), constituting the three-dimensional spatial coordinate sequence for subsequent analysis.

[0023] Table 1 Example data table of three-dimensional coordinate sequence

[0024] As shown in Table 1, the table lists the three-dimensional coordinate data of the hyoid bone, laryngeal body, and cricopharyngeal marker points accurately measured at different time points, in millimeters, forming a three-dimensional spatial coordinate sequence arranged in time order.

[0025] The differential time series processing submodule calls the three-dimensional space coordinate sequence, extracts the horizontal displacement of the hyoid bone marker, the vertical displacement of the laryngeal body, and the depth displacement of the cricopharynx, calculates the mixed displacement feature, and generates a differential time series set in chronological order; The specific operation is: for the hyoid bone landmark point, extract its The horizontal direction ( Axis) coordinate value , calculate adjacent timestamps arrive The coordinate change between the two is used to obtain the horizontal displacement of the hyoid bone. , using the data in Table 1, When , the horizontal displacement of the hyoid bone is ,exist When .

[0026] Similarly, extract the laryngeal landmarks at each time stamp The vertical direction ( Axis) coordinate value , calculate the vertical displacement of the throat ,exist When the vertical displacement of the throat is ,exist When , and then extract the cricopharyngeal landmarks at each timestamp The depth direction ( Axis) coordinate value , calculate the cricopharyngeal depth displacement ,exist When ,exist When , each time point The three calculated displacements , and (units are all mm) are combined into a three-dimensional vector as the mixed displacement feature at that time point ,exist When , the mixed displacement characteristic is ,exist When , all time points arrive Calculated Generate a differential time series set in chronological order , which represents the instantaneous motion rate information of key structures in a specified direction during swallowing.

[0027] Based on the differential time series set, the displacement vector generation sub-module extracts the three-dimensional direction components of the differential scalar values at multiple time points, normalizes them and combines them into a vector form to generate a displacement vector set.

[0028] Based on the differential time series set , where each element represents to the mixed displacement characteristics during the time period. The goal of this sub-module is to convert these displacement characteristics into standardized direction vectors. First, each mixed displacement characteristic quantity is directly regarded as an original three-dimensional displacement vector , whose component unit is mm. At , , at , , at , calculated according to the data in Table 1 , , , so .

[0029] Next, for each non-zero original displacement vector , perform normalization processing, calculate the Euclidean norm (magnitude) of the vector , the unit is still mm, and then divide each component of the vector by its magnitude to obtain a direction vector with unit length (unitless) . For , its magnitude is mm; The normalized vector is ; For , the magnitude mm; The normalized vector ; For , the magnitude is mm; The normalized vector is ; If the original displacement vector at a certain time point is a zero vector (i.e., ), then its normalized vector is also defined as a zero vector . Combine the normalized displacement vectors calculated at all time points Combine them to form a set of displacement vectors , each vector in this set represents the direction of the combined displacement at that moment (relative to the previous moment).

[0030] Please refer to Figure 3 , the spatial trajectory analysis module includes: The vector angle calculation sub-module calls the set of displacement vectors, extracts adjacent displacement vectors, calculates the spatial angle between adjacent vectors based on the dot product and the magnitude of the vectors, and generates a sequence of spatial angles; Call the set of displacement vectors , and sequentially extract two non-zero normalized displacement vectors at adjacent time points from it, denoted as and , where ranges from to . For each pair of such adjacent vectors , calculate their vector dot product . If and , the dot product is calculated as . Since the vectors are normalized, their magnitudes and are both theoretically 1 (ignoring floating-point errors). Therefore, apply the vector angle formula to calculate the spatial angle between these two adjacent displacement vectors, and the result is expressed in radians, with a value range of .

[0031] Taking the normalized vectors at and moments as an example, and , calculate their dot product: ; Then the angle radians; indicates that the displacement directions at these two moments are exactly the same. Considering the case of and again, , calculate : ; The magnitude mm; ; Calculate the dot product: ; Then the angle radians (approx. ); For all valid pairs of adjacent vectors Repeat this calculation process to obtain the spatial angle Arrange them in chronological order to finally generate a sequence of spatial angles .

[0032] The direction consistency judgment sub-module calls the sequence of spatial angles and performs a cross product operation on each pair of adjacent displacement vectors using the formula: ; Calculate the direction consistency index through the operation, compare the index with a preset dynamic threshold, and generate a set of direction consistency indices; Among them, represents adjacent displacement vectors, represents the time interval, represents the spatial angle, is the direction consistency index; Call the sequence of spatial angles and the set of displacement vectors , and calculate the direction consistency index and for each pair of adjacent non-zero normalized displacement vectors . This index comprehensively considers the magnitude of the direction change (through the cross product) and the angle (through the angle penalty term), and the calculation formula is: .

[0033] : Adjacent normalized displacement vectors (unitless). : The magnitude of the cross product of two vectors, which reflects the area of the parallelogram formed by them, and is proportional to, and a larger value usually indicates a greater direction difference (unitless). : The time interval between adjacent time points, with the unit of seconds (s). Here it is a constant value s. : The magnitude of the vector, which is 1 for both due to normalization (unitless). : The spatial angle between two vectors, with the unit of radians (rad), . Obtained from the sequence of spatial angles. : Pi. : The angle penalty term, which decreases from 1 to 0 when the angle increases from 0 to , giving low weight to vectors that are close to being in the opposite direction (unitless). The final index has the unit of .

[0034] Use the vectors of and : , , rad.

[0035] Calculate the cross product: ; Calculate the magnitude of the cross product: ; Calculate : ; Use and vectors: , , rad.

[0036] Calculate the cross product:

[0037] Calculate the magnitude of the cross product: ; Calculate : ; This formula aims to identify sharp, non-smooth transitions in the direction of motion by combining cross product to quantify direction differences and an angular penalty term. The magnitude of the value reflects the degree of direction change per unit time, taking into account the effect of the change angle.

[0038] Compare each calculated value with a preset dynamic threshold to distinguish normal direction fluctuations from potential motion discontinuities. The setting of is based on a statistical analysis of the value distribution during normal swallowing. Select a value that can effectively distinguish abnormal low-value points. This is understood as the lower the value, the worse the consistency (direction mutation or near reversal). By analyzing the value sequence calculated from the swallowing data of a large number of healthy subjects, calculate its mean and standard deviation and set the threshold to where is a coefficient, usually taken as 1.5 or 2, to capture values below the normal fluctuation range. Here, the mean of the normal swallowing value is analyzed, the standard deviation . When the calculated value is lower than When it is, the direction consistency at this time point is considered "low", otherwise the consistency is considered "acceptable" or "high". Record the result of each comparison ( whether it holds) in chronological order to form a direction consistency judgment sequence, or directly store the value sequence .

[0039] In the numerical example , because , this indicates that the direction change at moment relative to is judged to have low consistency. And , because , this indicates that the direction change at moment relative to is judged to be acceptable in consistency. These values and their comparison results with the threshold will be used to screen potential motion breakpoints in the next step.

[0040] The breakpoint screening sub-module calls the set of direction consistency indicators and the spatial angle sequence, selects the positions where the indicators are lower than the threshold and the angle is greater than the critical angle, and merges adjacent abnormal points to generate spatial continuity breakpoints.

[0041] Integrate the set of direction consistency indicators and the spatial angle sequence , and screen out the time points indicating significant discontinuity in the motion trajectory. The screening criteria are: at a certain time point , its corresponding direction consistency indicator is lower than the threshold , and its spatial angle is greater than the critical angle .

[0042] The direction consistency threshold has been determined to be . The critical angle threshold is set based on distinguishing smooth turning in physiological motion from drastic direction changes that may be caused by interruptions or disturbances. During normal swallowing, the direction change of displacement at adjacent time points (interval 0.01s) usually does not have a sharp turn of more than 90 degrees. Therefore, the critical angle is set to radians (i.e., ).

[0043] Check the time point : , rad( ). Condition 1: (satisfied). Condition 2: (Not satisfied). Therefore, is not marked as an outlier. Check time point : , rad( ). Condition 1: (Not satisfied). Therefore, is not marked as an outlier. Assume that at time point , it is calculated that and rad( ). Check the conditions: Condition 1: (Satisfied). Condition 2: (Satisfied). Therefore, the time point is marked as an outlier.

[0044] For all (from to ), perform this dual-condition check to identify all outlier time points that meet the conditions. Subsequently, check the distribution of these outlier points on the time axis. If it is found that there are multiple consecutive adjacent timestamps (such as ) that are all marked as outlier points, then merge them and consider them as a single spatial continuity break event. The start and end times of this event are and respectively. Finally, generate a set of spatial continuity break points, where each element represents a detected interruption of the motion trajectory and records the timestamp (or time period) when it occurs, such as .

[0045] Please refer to Figure 4 , the frequency domain centroid extraction module includes: The frequency domain conversion sub-module performs short-time Fourier transform on the original EMG signal, divides the signal according to a fixed time window, calculates the complex spectral amplitudes of multiple frequency points within the 0 - 100 Hz frequency band in each window, and generates a frequency domain energy map; First, divide the continuous sEMG signal into a series of time windows of a fixed length. Select the Hamming window as the window function to reduce spectral leakage. The window length is set to 256 milliseconds (ms), and a 50% overlap rate is set between windows, which means that each new window slides forward relative to the previous window. For each windowed signal segment , perform the fast Fourier transform (FFT) to obtain the complex spectrum of this time window, where represents the central time point of this window, and represents the frequency.

[0046] Next, extract the information in the frequency band from 0 Hz to 100 Hz from the calculated frequency spectrum. In this frequency band, according to the parameters of the FFT (such as the sampling rate of 1000 Hz, and the window length corresponding to the FFT point number N = 256), a series of discrete frequency points will be obtained. . The frequency resolution is . Therefore, in the range of 0 - 100 Hz, the frequency points are approximately . There are a total of frequency points. For each time window , calculate and store the complex frequency spectrum amplitudes corresponding to these discrete frequency points (unit: microvolt , or the unit after amplification). Combine the calculation results of all time windows to form a two-dimensional frequency-domain energy map (time-frequency spectrum), with the horizontal axis being time , the vertical axis being frequency , and the values in the map being amplitudes . Table 2 shows some of the calculation results.

[0047] Table 2 Example Data Table of Frequency-Domain Energy Map (Amplitude)

[0048] As shown in Table 2, this table lists the example of the myoelectric signal frequency spectrum amplitudes (unit ) at some discrete frequency points within the 0 - 100 Hz frequency band for the central points of different time windows.

[0049] The energy centroid sequence generation sub-module calls the frequency-domain energy map, extracts the energy values of the frequency points within multiple time windows, and uses the formula: ; Calculate the weighted mean of the frequency energy within each window, splice the calculation results of adjacent windows in chronological order, and generate an energy centroid sequence; where, represents the energy centroid value of the th time window, is the frequency value corresponding to the th frequency point, is the complex frequency spectrum amplitude of the th window at the th frequency point, is the adjustment factor for the total energy difference between adjacent windows, is the absolute value of the total energy change of the previous window, is the total number of discrete frequency points within the frequency band.

[0050] Call the frequency-domain energy map data , where is the time window index, is the index of discrete frequency points within the 0 - 100 Hz frequency band. For each time window , calculate the weighted average of the frequency energy, that is, the energy center of gravity , in hertz (Hz), and the calculation formula is: . .

[0051] : The energy center of gravity value (Hz) of the th time window. : The frequency value (Hz) of the th discrete frequency point. . : The spectral amplitude of the th window, the th frequency point ( ). : The corresponding energy value ( ). : The total number of frequency points within the frequency band, . : The energy difference adjustment factor, which is a small positive constant used to avoid the denominator being zero when the signal energy is extremely low and to slightly smooth the result. Its setting should be much smaller than the total energy of a typical window. Set to ensure that its impact on the result is extremely small in signals on the microvolt scale. Setting basis: Select a value that is several orders of magnitude smaller than the expected minimum non - zero total energy squared ( ). : The total energy of the previous window ( ) and the total energy of the window before the previous one ( ) The absolute value of the difference between them, that is ( ).

[0052] Calculate the energy center of gravity of the time window . Using the data in Table 2, assume for simplicity that only 3 frequency points are used to represent: . The corresponding amplitude . (Note: Here, for simplicity of calculation, the frequency points and amplitudes are slightly adjusted compared to the actual data in Table 2 and the previous description. The actual calculation should use all 26 frequency points and their corresponding amplitudes).

[0053] Calculate the numerator (sum of frequency energies): ; Calculate the total energy term in the denominator: ; Calculate the adjustment term in the denominator, which requires the total energy of the first two windows. Obtained from Table 2 or calculated: for window the total energy (assuming calculated using simplified frequency points ), and for window the total energy (assuming calculated as ).

[0054] Then ; Calculate the denominator: (the adjustment term has a minimal impact); Calculate the energy centroid : Hz.

[0055] This formula provides the center point of the frequency distribution of the EMG signal within each time window by calculating the energy-weighted average frequency. The change in reflects the spectral characteristic changes related to the muscle activity state (such as contraction intensity, fatigue degree). The adjustment term in the denominator

[0056] For each time window (with the center time starting from and a step size of ), perform this calculation to obtain a series of energy centroid values . Arrange these values in chronological order to generate the energy centroid sequence .

[0057] The calculated Hz indicates that within the time window centered at 1.024 s, the energy of the EMG signal is mainly distributed around 61 Hz. The change trend of the entire sequence is the basis for subsequent mutation detection.

[0058] The mutation detection sub-module calculates the first-order difference sequence based on the energy centroid sequence, statistically calculates the standard deviation of the difference sequence, screens the data points where the difference value exceeds 2 times the standard deviation of the difference sequence, and marks the corresponding time points as frequency centroid mutation points.

[0059] First, calculate the first-order difference sequence of the energy centroid sequence, which represents the change in the energy centroid between adjacent time windows: , where corresponds to the time points of the second window and later.

[0060] Using the calculation from the previous paragraph in Hz, and assuming the center of gravity of the previous window The calculated value is in Hz, then the difference is in Hz. Then assume the next window (calculated corresponding to the data in Table 2) is in Hz, then the difference is in Hz. Calculate the difference values of the entire sequence .

[0061] Then, count the standard deviation of the entire difference sequence . The standard deviation reflects the typical amplitude of the change in the energy center of gravity value between adjacent windows. By performing statistics on the sequence calculated for the entire swallowing process, we obtain in Hz. Set the mutation detection threshold to twice the standard deviation of the difference sequence: in Hz. This means that if the absolute value of the change in the energy center of gravity between two adjacent 128 - ms step - time windows exceeds 9.6 Hz, a "significant" change or "mutation" in the frequency center of gravity is considered to have occurred.

[0062] Next, traverse each value in the difference sequence and check whether its absolute value is greater than the threshold . For in Hz, , it does not exceed the threshold. For in Hz, , it exceeds the threshold. Mark all the time points that satisfy the condition (i.e., the center timestamp of the window where the energy center of gravity mutates). Finally, generate a set of frequency center of gravity mutation points. This set contains all the time points where a rapid change in the energy center of gravity is detected, such as .

[0063] Please refer to Figure 5 . The dual - mode fusion module includes: The event index alignment sub - module obtains the spatial continuity break points and frequency center of gravity mutation points, establishes a dual - channel event index mapping table, and at the same time calls the time - axis alignment strategy to perform cross - modal matching on the timing coordinates of the break points and mutation points, screening out the synchronous event pairs with the absolute value of the time difference less than the set tolerance threshold to generate a time synchronization index table; Establish the timestamp lists of two events: the break point list and the mutation point list . Adopt the time - window matching strategy for alignment: For each spatial break point , search for mutation points in the mutation point list to see if there are any mutation points such that the absolute value of the time difference between the two is less than the preset time tolerance threshold . The setting of this threshold needs to balance the time accuracy of the two-modal detection and the synchronization of physiological events. Considering the time resolution differences between motion capture (10 ms accuracy) and the sEMG energy center of gravity (window step size 128 ms), as well as the fast conduction of neuromuscular events during swallowing, select ms as the tolerance window.

[0064] Take the spatial break point as an example and search in . Calculate ; Calculate ; Calculate ; No matching mutation point was found for this break point. Take the spatial break point as an example. Calculate ; Calculate ; A matching pair is found! Record the index of this pair of synchronous events and the corresponding timestamp .

[0065] Repeat this matching process for all spatial break points to filter out all synchronous event pairs that meet the time tolerance condition . Store the indices of these matching pairs , the corresponding break point timestamps , the mutation point timestamps , and their time differences to generate a time synchronization index table. Table 3 shows a part of the subsequent table containing this information.

[0066] Based on the time synchronization index table, the interruption activation determination sub-module extracts the break strength value and mutation amplitude of the synchronous event pair, and uses the formula: ; Calculate the cross-modal event correlation weight value, compare the weight value with the dynamic determination threshold, and if three consecutive weight values exceed the threshold and show an increasing trend, then determine it as a valid activation interruption event and generate an interruption activation weight column; Among them, represents the th spatial continuity break point and the The cross-modal event association weight value between the frequency center mutation points, Representative The strength value of the spatial continuity breaking point, Representative The amplitude of the frequency centroid mutation point, It represents the maximum value of the spatial breakpoint intensity value and the frequency mutation point amplitude. is the absolute value of the time difference between two event points, represents the modal difference compensation coefficient; First, for each synchronization event pair Extract the respective intensities / amplitudes. The strength value of the spatial fracture point , quantified as the spatial angle value (unit: degree) that leads to the detection of the breakpoint. , assuming The fracture at Caused by Extract the The amplitude of the frequency centroid mutation point , quantified as the absolute value of the first-order difference of the energy center of gravity that leads to the detection of the mutation point (unit: Hz). , assuming that it is obtained by differential Hz, then .

[0067] because (angle) and The units and physical meanings of (frequency change rate) are different, and direct comparison or subtraction lacks clear meaning. Therefore, before calculating the weights, they need to be converted to a unified comparable scale. Using the normalization method, and Mapped to the interval [0,1]. Conversion rules and instructions: For the spatial angle , whose theoretical range is The normalization rule is This rule linearly maps the angle to [0,1], and the closer the value is to 1, the greater the change in direction. , its typical range of variation needs to be determined. Based on the standard deviation calculated previously Hz, set an upper limit for the maximum possible change amplitude, such as Hz (or set according to observed data). The normalization rule is This rule linearly maps the mutation amplitude to [0,1]. Values ​​exceeding the set upper limit are truncated to 1. The larger the value, the more drastic the change in the frequency center of gravity.

[0068] Apply the transformation rules:

[0069] Obtain the time difference 。Set the modal difference compensation coefficient 。This coefficient is used to adjust the influence of the time difference in the weight calculation. Considering the time resolution and synchronization requirements of the two modal detections, set (unitless), indicating a moderate requirement for time synchronization and slightly reducing the influence of the time difference. Setting basis: Empirical adjustment to balance the contributions of intensity matching and time matching in the final weight.

[0070] Calculate the cross-modal event correlation weight , using the normalized intensity / magnitude: 。(If , then ).

[0071] Event label calculation : Numerator: ; Denominator: ; Time term: ; Note: The original formula design may result in the final with a square root with time units ( ). This is not intuitive in physical interpretation. To obtain a unitless weight, modify the formula, such as normalizing the time term or adjusting the structure. Here, calculate according to the original formula structure but point out this unit problem. To obtain a unitless weight, a possible modification is to divide by a reference time constant (such as ) and then take the square root, or adjust the unit of to . Assuming the latter idea is adopted and set , then the time term becomes (unitless); Calculate using the modified unitless time term ; This formula combines the relative difference of the normalized event intensity / magnitude (the smaller the difference, the smaller the first term) and the time proximity (the smaller the time difference, the smaller the second term) to generate a weight value quantifying the correlation degree of the two modal events.

[0072] Compare each calculated weight with a dynamic decision threshold . This threshold is used to distinguish low-weight matches of random coincidences from high-weight matches with indicative significance. The setting of is based on the calculation during non-interrupt periods Analysis of the background noise distribution, select values higher than most of the background noise. Through analysis and calculation, the value distribution (for example, calculate its mean and standard deviation ), set . Here, the analysis obtains , then .

[0073] Determine whether there is a valid activation interruption event: Check whether there are three consecutive groups (pairs of synchronous events arranged in chronological order) of weight values (the indices do not have to be consecutive, referring to adjacent matching pairs in time) that all exceed the threshold ( ), and these three weight values show an increasing trend ( ). If both of these conditions are met, it is determined that a valid activation interruption event has occurred near or . Mark the last pair of events that meet the conditions as "valid activation", and store all the calculated values in the interrupted activation weight column.

[0074] In the numerical example , because , this pair of events does not exceed the threshold and does not constitute a part of the activation condition itself. It is necessary to check the values of subsequent synchronous event pairs to see if a pattern of "three consecutive groups exceeding the threshold and increasing" can be formed. Only events that meet this strict pattern are confirmed as valid swallowing interruption activation events, aiming to improve the specificity of the determination.

[0075] Table 3 Example of synchronous event pairs, weights, and activation determination table

[0076] As shown in Table 3, this table shows the time-synchronized event pairs and their related parameters, the calculated dimensionless correlation weights, and the activation status determined according to the rule of "three consecutive groups exceeding the threshold and increasing" (in this example, it is assumed that the event pairs with indices 4, 5, and 6 meet the activation conditions and are marked at the 6th event).

[0077] Call the time synchronization index table and the interrupted activation weight column to perform weighted fusion on the spatio-temporal coordinates of cross-modal event pairs, eliminate redundant markings, and generate activation interruption event markings covering the time axis.

[0078] Using the time synchronization index table and the interrupt activation weight column (as shown in Table 3), filter out the synchronized event pairs determined to be "effectively activated". For each event pair marked as effectively activated (in the example of Table 3, it is the event pair with index ), determine a spatio-temporal coordinate representing the fusion event.

[0079] Time coordinate Adopt the timestamp of the spatial break point in the activation event pair, that is . In the example, . Spatial coordinate By querying the original three-dimensional coordinate sequence (Table 1), extract the coordinates of the structures (hyoid bone, laryngeal body, cricopharyngeal part) related to the spatial break at the timestamp , and calculate their geometric center. Assume that at , the hyoid bone coordinate , the laryngeal body coordinate , the cricopharyngeal part coordinate (unit: mm), then the spatial position coordinate of the event is calculated as: mm; mm; mm; Therefore, the spatial coordinate of this activation interrupt event is mm.

[0080] Perform redundant marker elimination. If there are multiple effective activation event markers with very close timestamps (for example, the time difference is less than ), merge them into one event, select the marker with the largest associated weight as the representative, or calculate their weighted average spatio-temporal coordinates. In this example, only one activation event is recognized and no merging is required.

[0081] Finally, generate a set of activation interrupt event markers covering the entire analysis time axis. Each marker contains the timestamp when the event occurs and the associated spatial position coordinate . The example output is: .

[0082] Please refer to Figure 6 , the anomaly verification module includes: The event marker parsing sub-module obtains the activation interrupt event marker, extracts the event occurrence timestamp and the associated node position coordinate, calls the preset node displacement sampling interval parameter, aligns the timestamp with the energy centroid acquisition period, and generates an event alignment time series; Receive the set of activation interrupt event markers, such as It is responsible for aligning the time information marked by these events with the time reference required for subsequent analysis.

[0083] Extract the time stamps of the occurrence of events from each event marker . For , . Call the preset node displacement sampling interval parameter, which here refers to the sampling time interval of the original coordinate data, that is . At the same time, it is necessary to align the event time stamps with the time axis of the energy center of gravity calculation (the window center time stamp sequence , step size 128 ms). Determine which energy center of gravity window it belongs to. The coverage range of the window with a center time of 1.920 s is approximately ; the coverage range of the window with a center time of 2.048 s is approximately ; Therefore, falls within the energy center of gravity calculation window with a center time of .

[0084] At the same time, align with the time axis of the original coordinate data to determine the corresponding frame index. Since the sampling interval is 0.01 s and the first frame is 0.01 s, then corresponds to the th frame data.

[0085] Generate an event alignment time series, which records each activation interrupt event and its position in different time coordinate systems (energy center of gravity window index , original data frame index ). For , the alignment information is: .

[0086] The muscle group vector retrieval sub-module retrieves the set of muscle group displacement vectors within the corresponding time interval based on the event alignment time series, performs vector superposition operations on the three-dimensional coordinate data of each node, and generates the average energy center of gravity of multiple nodes; Based on the information provided by the event alignment time series (such as occurring in the 199th frame), retrieve the muscle group movement data related to this event. The "muscle group" defined here is represented by the marker points (hyoid bone, laryngeal body, cricopharynx) participating in the calculation of the spatial break point and the event position.

[0087] According to the original data frame index corresponding to the event time ​, retrieve the three-dimensional coordinate data of these three marker points at the 199th frame from the complete three-dimensional space coordinate sequence (the complete version of Table 1): mm; mm; mm; Execute the "energy centroid mean of multiple nodes" calculation, that is, calculate the geometric center (centroid) coordinates of these multiple marker points representing muscle groups at the event time . This has been calculated in the "event marker generation sub-module", and the result is: mm. mm.

[0088] This step confirms the central position representing the overall relevant muscle group at the occurrence time of the detected interruption event. Repeat this process for all activated interruption events.

[0089] The cascaded offset determination sub-module calls the muscle group displacement vector set and the energy centroid mean, calculates the standard deviation of the multi-node displacement vector and the energy centroid, compares it with the preset synchronous offset threshold, and generates a cascaded interruption determination coefficient when the number of nodes exceeding the threshold reaches 70% of the total number of nodes.

[0090] Call the position coordinate set of each marker point within the muscle group corresponding to the activation interruption event time (such as ) and the geometric center of the muscle group at this moment and , aiming to evaluate whether there are significant and asynchronous offsets in the movement of each component within the muscle group at this interruption moment.

[0091] Calculate at the event time , the displacement vector of each marker point relative to the geometric center of the muscle group : mm; mm; mm; Compare these distances with the preset synchronous offset threshold . This threshold represents the typical maximum offset of points within the muscle group relative to the center during normal coordinated movement. By analyzing the value distribution calculated from the healthy control group during normal swallowing movements, set the threshold to the 95th percentile of this distribution, and obtain mm. This means that if the distance of a marked point from the center of the muscle group exceeds 5.0 mm, it is considered to have a "significant" deviation.

[0092] Statistically count at the event moment , the number of marked points with deviation distance exceeding the threshold. Hyoid bone: (exceeding) Laryngeal body: (not exceeding) Cricopharyngeal part: (exceeding) There are a total of 2 marked points (hyoid bone, cricopharyngeal part) whose deviation distances exceed the threshold.

[0093] Calculate the percentage of the number of nodes exceeding the threshold in the total number of nodes :

[0094] Compare this percentage with the preset cascade determination ratio (set to 70%). This ratio defines what proportion of the muscle group components need to have significant deviations to consider that the overall movement pattern is dysregulated. Comparison result: .

[0095] Since the proportion of nodes exceeding the deviation threshold (66.7%) does not reach 70%, it is determined that at the moment of this interruption event , there is no cascading internal deviation of the muscle group. Generate a cascade interruption determination coefficient, which can be set to 0 (indicating not occurring) or directly use the percentage 66.7%. If the percentage reaches or exceeds 70%, the determination coefficient is 1 (indicating occurring) or the corresponding percentage. This coefficient is ultimately used to confirm whether the detected interruption event is accompanied by a disruption of the internal movement coordination of the muscle group.

[0096] A screening method for dysphagia, the screening method for dysphagia is performed based on the above-mentioned dysphagia screening system, and includes the following steps: S1: Obtain the three-dimensional spatial coordinate sequences of the hyoid bone, laryngeal body and cricopharyngeal part during swallowing activities through the body surface marked point signal acquisition device, and perform differential operations on the three-dimensional spatial coordinate sequences based on a fixed time interval to generate a set of displacement vectors; S2: Call the set of displacement vectors, perform spatial angle calculations on adjacent displacement vectors, judge the direction consistency through vector cross product operations, and generate spatial continuity break points when the angle exceeds 15 degrees and the cross product direction difference triggers a set threshold; S3: Perform short-time Fourier transform on the original electromyogram signal of the target muscle group to generate a frequency domain energy distribution map, calculate the weighted average based on the energy values in the 0 - 100 Hz frequency band to generate an energy center of gravity time series, and detect the frequency center of gravity mutation points where the first-order difference exceeds twice the standard deviation; S4: Establish a dual-channel event index mapping table to align the time axis of the spatial continuity breakpoint and the frequency centroid mutation point. Generate an activation interrupt event flag when the two are synchronously triggered within the same time window. S5: Based on the activation interrupt event flag, inversely retrieve the displacement vector set and the energy centroid time series of the associated muscle groups, compare the synchronous offset characteristics of multiple muscle group nodes in adjacent time periods, and generate a cascaded interrupt determination result.

[0097] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A swallowing disorder screening system, characterized in that: The system comprises: A three-dimensional coordinate acquisition module, used to acquire a three-dimensional spatial coordinate sequence of the hyoid bone, laryngeal body and cricopharynx through body surface markers, generate a displacement vector set based on fixed interval differences, and transmit the three-dimensional spatial coordinate sequence and the displacement vector set to a spatial trajectory analysis module; A spatial trajectory analysis module, used to call the displacement vector set, perform spatial angle calculation on adjacent displacement vectors, determine direction consistency through cross product, generate spatial continuity breakpoints, and transmit the spatial continuity breakpoints and the three-dimensional spatial coordinate sequence to a dual-modal fusion module; The frequency domain centroid extraction module is used to perform short-time Fourier transform on the original electromyographic signal to generate a frequency domain energy spectrum, calculate the 0-100Hz energy weighted average to obtain the energy centroid time series, detect the frequency centroid mutation point where the first-order difference exceeds two times the standard deviation, and transmit the energy centroid time series and the frequency centroid mutation point to the dual-modal fusion module; A dual-modal fusion module, used for receiving the spatial continuity breakpoint and the frequency center of gravity mutation point, establishing a dual-channel event index mapping table to align the time axis, generating an activation interruption event mark, and transmitting the activation interruption event mark to an abnormality verification module; An abnormality verification module, used for reversely retrieving the displacement vector set and the energy center of gravity time series of the associated muscle group based on the activation interruption event mark, and comparing the multi-node synchronization offset to generate a cascade interruption determination result; The cascade interruption determination result includes the muscle group chain node number, synchronization offset count, and systemic disorder level.

2. The dysphagia screening system according to claim 1, characterized in that: The three-dimensional spatial coordinate sequence includes hyoid coordinates, laryngeal coordinates, and cricopharyngeal coordinates. The displacement vector set is specifically a time interval differential vector, a three-dimensional directional component, and a displacement modulus. The spatial continuity breakpoint includes an angle excess mark, a directional difference mark, and a timestamp index. The energy center of gravity time series specifically refers to the weighted mean of the frequency band, the time window energy distribution, and the center of gravity offset. The frequency center of gravity mutation point includes a mutation amplitude threshold, a drift direction identifier, and a duration count. The dual-channel event index mapping table includes a time window index, a space fracture event ID, and a frequency domain mutation event ID. The activation interrupt event mark is specifically a synchronization trigger flag, a space-frequency domain association weight, and an abnormality confidence score.

3. The swallowing disorder screening system according to claim 2, characterized in that: The three-dimensional coordinate acquisition module includes: The coordinate sequence acquisition submodule collects the three-dimensional coordinate data of the hyoid bone, larynx and cricopharynx through body surface markers, records them as a three-dimensional space coordinate sequence in time sequence, and stores them as a matrix structure aligned with time stamps; The differential time series processing submodule calls the three-dimensional space coordinate sequence, extracts the horizontal displacement of the hyoid bone marker, the vertical displacement of the laryngeal body and the depth displacement of the cricopharynx, calculates the mixed displacement feature, and generates a differential time series set in chronological order; The displacement vector generation submodule extracts the three-dimensional directional components of the differential scalar values ​​at multiple time points based on the differential time series set, normalizes them and combines them into a vector form to generate a displacement vector set.

4. The swallowing disorder screening system according to claim 3, characterized in that: The spatial trajectory analysis module includes: The vector angle calculation submodule calls the displacement vector set, extracts adjacent displacement vectors, calculates the spatial angles of adjacent vectors based on the vector dot product and the modulus, and generates a spatial angle sequence; The direction consistency judgment submodule calls the spatial angle sequence and performs a cross product operation on each pair of adjacent displacement vectors using the formula: ； Obtaining a directional consistency index through operation, comparing the index with a preset dynamic threshold, and generating a directional consistency index set; in, represents the adjacent displacement vector, represents the time interval, represents the spatial angle, is the directional consistency indicator; The breakpoint screening submodule calls the direction consistency index set and the spatial angle sequence, selects positions where the index is lower than a threshold and the angle is greater than a critical angle, and merges adjacent abnormal points to generate spatial continuity breakpoints.

5. The swallowing disorder screening system according to claim 4, characterized in that: The frequency domain centroid extraction module comprises: The frequency domain conversion submodule performs short-time Fourier transform on the original electromyographic signal, divides the signal into fixed time windows, calculates the complex spectrum amplitude of multiple frequency points in the 0-100 Hz frequency band in each window, and generates a frequency domain energy spectrum; The energy center of gravity sequence generation submodule calls the frequency domain energy spectrum to extract the energy values ​​of frequency points in multiple time windows using the formula: ； Calculate the weighted mean of frequency energy in each window, and concatenate the calculation results of adjacent windows in chronological order to generate an energy center of gravity sequence; in, Representative The energy center of gravity value of the time window, For the The frequency value corresponding to the frequency point is For the Window The complex spectrum amplitude at the frequency point, is the adjustment factor for the total energy difference of adjacent windows, is the absolute value of the total energy change in the previous window, is the total number of discrete frequency points in the frequency band; The mutation detection submodule calculates the first-order difference sequence based on the energy center of gravity sequence, calculates the standard deviation of the difference sequence, selects data points whose difference values ​​exceed 2 times the standard deviation of the difference sequence, and marks the corresponding time point as the frequency center of gravity mutation point.

6. The dysphagia screening system according to claim 5, characterized in that: The dual-modal fusion module includes: The event index alignment submodule obtains the spatial continuity breakpoint and the frequency center of gravity mutation point, establishes a dual-channel event index mapping table, and calls the time axis alignment strategy to perform cross-modal matching on the time series coordinates of the breakpoint and the mutation point, screens the synchronization event pairs whose absolute time difference is less than the set tolerance threshold, and generates a time synchronization index table; The interrupt activation determination submodule extracts the fracture intensity value and mutation amplitude of the synchronization event pair based on the time synchronization index table, using the formula: ； Calculate the cross-modal event association weight value, compare the weight value with the dynamic judgment threshold, if three consecutive groups of weight values ​​exceed the threshold and show an increasing trend, it is determined to be a valid activation interrupt event, and generate an interrupt activation weight column; in, Representative The spatial continuity breakpoint and the The cross-modal event association weight value between the frequency center mutation points, Representative The strength value of the spatial continuity breaking point, Representative The amplitude of the frequency centroid mutation point, It represents the maximum value of the spatial breakpoint intensity value and the frequency mutation point amplitude. is the absolute value of the time difference between two event points, represents the modal difference compensation coefficient; The event marker generation submodule calls the time synchronization index table and the interrupt activation weight column, performs weighted fusion on the spatiotemporal coordinates of the cross-modal event pairs, eliminates redundant markers, and generates an activation interrupt event marker covering the time axis.

7. The dysphagia screening system according to claim 1, characterized in that: The abnormal verification module includes: The event marker parsing submodule obtains the activation interrupt event marker, extracts the event occurrence timestamp and the associated node position coordinates, calls the preset node displacement sampling interval parameters, aligns the timestamp with the energy center of gravity collection period, and generates an event alignment time series; The muscle group vector retrieval submodule retrieves the muscle group displacement vector set in the corresponding time interval based on the event alignment time series, performs vector superposition operation on the three-dimensional coordinate data of each node, and generates the energy center of gravity mean of multiple nodes; The cascade offset determination submodule calls the muscle group displacement vector set and the energy center of gravity mean, calculates the standard deviation of the multi-node displacement vector and the energy center of gravity, and compares it with the preset synchronization offset threshold. When the number of nodes exceeding the threshold reaches 70% of the total number of nodes, a cascade interruption determination coefficient is generated.

8. A method for screening dysphagia, characterized in that: The method is used to implement the dysphagia screening system according to any one of claims 1 to 7, comprising the following steps: S1: acquiring a three-dimensional spatial coordinate sequence of the hyoid bone, laryngeal body and cricopharynx during swallowing activity through a body surface marker signal acquisition device, and performing a differential operation on the three-dimensional spatial coordinate sequence based on a fixed time interval to generate a displacement vector set; S2: calling the displacement vector set, performing spatial angle calculation on adjacent displacement vectors, judging direction consistency through vector cross product operation, and generating a spatial continuity breakpoint when the angle exceeds 15 degrees and the cross product direction difference triggers a set threshold; S3: Perform short-time Fourier transform on the original electromyographic signal of the target muscle group to generate a frequency domain energy distribution spectrum, calculate the weighted average based on the energy value of the 0-100 Hz frequency band to generate an energy center of gravity time series, and detect the frequency center of gravity mutation point where the first-order difference exceeds two times the standard deviation; S4: establishing a dual-channel event index mapping table to align the time axis of the spatial continuity breakpoint and the frequency center of gravity mutation point, and generating an activation interrupt event mark when the two are synchronously triggered within the same time window; S5: Based on the activation interrupt event mark, the displacement vector set and the energy center of gravity time series of the associated muscle group are reversely retrieved, the synchronous offset characteristics of multiple muscle group nodes in adjacent time periods are compared, and a cascade interrupt determination result is generated.

Images