Fd assisted evaluation information processing method based on gastrointestinal electrical signals

Abstract

The application discloses a kind of FD auxiliary evaluation information processing methods based on gastrointestinal electrical signal, it is related to medical auxiliary diagnosis technical field, it is applied to the functional dyspepsia auxiliary diagnosis scene of obese crowd, including obtaining the individualized medical record data of target object, the individualized medical record data includes at least two in body shape parameter, weight loss surgery history and metabolic drug history;Three evaluation paths of parallel operation are constructed.The present application solves the core technical problems of EGG signal systematic distortion and main frequency detection distortion caused by abdominal fat layer of obese crowd, breaks through the long-term technical prejudice of existing EGG diagnostic equipment to exclude obese patients, realizes reliable auxiliary evaluation information under the condition that patients cannot maintain standard signal quality, eliminates the systematic misjudgment of fat layer to EGG diagnosis, so that the signal entering subsequent analysis link can truly reflect the gastric electrical physiological state of target object.

Description

Technical Field

[0001] This invention relates to the field of medical auxiliary diagnostic technology, specifically to a method for processing FD-assisted assessment information based on gastrointestinal electrical signals. Background Technology

[0002] Functional dyspepsia (FD) is a brain-gut interaction disorder characterized by postprandial fullness, early satiety, upper abdominal pain, or burning sensation, with a global prevalence of approximately 10%–20%. Electroastrogram (EGG) records slow-wave electrical activity of gastric smooth muscle using abdominal electrodes, extracting temporal indicators such as dominant frequency, percentage of normal gastric rhythms, and line length characteristics. It is currently one of the most promising non-invasive diagnostic techniques for FD. However, EGG technology faces several fundamental obstacles when applied to obese individuals (BMI ≥ 28 kg / m²).

[0003] First, the abdominal fat layer physically attenuates and distorts the EGG signal. Computer simulations show that increased fat layer thickness can lead to distortion in EGG dominant frequency detection and a systematic underestimation of the percentage of normal slow waves, while the gastric magnetoencephalogram (GEG) signal is minimally affected by the fat layer.

[0004] Second, the history of bariatric surgery has a permanent effect on the electrophysiological structure of the stomach. After sleeve gastrectomy, the dominant frequency and amplitude of gastric electrical activity are significantly reduced. If the surgical history parameters are not validated, normal postoperative changes can easily be misinterpreted as pathological rhythm disorders of functional gastrointestinal dysregulation (FD).

[0005] Third, GLP-1 receptor agonists have a functional additive modulation effect on gastric electrical activity. The slow-wave rhythm modulation effect of these drugs can be superimposed on abnormal FD rhythm symptoms, which current diagnostic systems do not distinguish.

[0006] The shortcomings of existing technologies are as follows: on the one hand, existing EGG diagnostic devices only improve signal quality by increasing electrode spatial resolution, without utilizing MGG for co-verification of EGG or switching of diagnostic paths; on the other hand, existing adaptive gastric electrophysiological monitoring methods use serial conditional branching logic, selecting one strategy to execute based on the level of interference and abandoning other paths. This selective approach leads to the discarding of multi-source information and completely fails to take into account prior compensation for bariatric surgery history and GLP-1 drug history. Summary of the Invention

[0007] The purpose of this invention is to provide a method for FD-assisted assessment information processing based on gastrointestinal electrical signals, in order to overcome the shortcomings of the prior art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for processing functional dyspepsia (FD)-assisted assessment information based on gastrointestinal electrical signals, applied in the auxiliary diagnosis of functional dyspepsia in obese individuals, comprising: Obtain individualized medical record data of the target subject, wherein the individualized medical record data includes at least two of the following: body shape parameters, history of weight loss surgery, and history of metabolic drugs; Three evaluation paths are constructed to run in parallel, including a first path, a second path, and a third path; the first path, the second path, and the third path run independently in parallel and do not exclude each other. The first path includes classifying the multi-channel gastrointestinal electrical signals on the body surface after individualized compensation and correction, and outputting a first probability value and a first confidence score; The second path includes classifying the normalized compensated gastric magnetoencephalogram signal and outputting a second probability value and a second confidence score. The third path includes calculating a multidimensional heterogeneity measure between the multi-channel gastrointestinal electrical signal on the body surface and the gastric magnetocardiogram signal, generating dynamic fusion weights based on the multidimensional heterogeneity measure, weighting and fusing the first probability value and the second probability value, and outputting a third probability value and a third confidence score. The first probability value and the first confidence score, the second probability value and the second confidence score, and the third probability value and the third confidence score are input into the decision fusion module; The decision fusion module determines the output strategy based on the first confidence score, the second confidence score, and the third confidence score, and allocates the auxiliary evaluation information to the corresponding output queue, outputting the functional dyspepsia auxiliary evaluation information with the corresponding output delay time.

[0009] In a preferred embodiment, the body shape parameters include at least one of body mass index, waist circumference, subcutaneous fat thickness, and visceral fat area; The individualized compensation correction includes constructing an individualized volumetric conductor digital twin model based on the body shape parameters, specifically including: The body shape parameters are mapped to a baseline abdominal anatomical model, and the body shape characteristics of the target object are adapted through parameterized deformation. In the deformed model, the electrophysiological simulation parameters of the gastric dipole were set, and the expected electric field intensity distribution and expected magnetic field intensity distribution at each electrode position on the body surface were calculated. The signal attenuation coefficient of each acquisition channel is estimated based on the ratio of the expected electric field intensity distribution to the preset electric field intensity reference value, and the quality confidence estimate of the gastromagnetic signal is estimated based on the ratio of the expected magnetic field intensity distribution to the preset magnetic field intensity reference value. The output parameter set of the individualized volumetric conductor digital twin model is generated based on the signal attenuation coefficient and the quality confidence estimate.

[0010] In a preferred embodiment, a personalized fusion model arbitration step is also included: Based on the history of weight loss surgery and the history of metabolic drugs, a corresponding dedicated fusion weight generator is selected from a pre-built fusion model library; The fusion model library includes: a first fusion weight generator corresponding to no history of weight loss surgery and no history of metabolic drugs; a second fusion weight generator corresponding to a history of laparoscopic sleeve gastrectomy; a third fusion weight generator corresponding to a history of GLP-1 receptor agonist use; and a fourth fusion weight generator corresponding to a history of both weight loss surgery and metabolic drugs. The network parameters of the first to fourth fusion weight generators are different, and they are optimized for the complementary characteristics of the multi-channel gastrointestinal electrical signals and the gastric magnetocardiogram signals under different interference modes.

[0011] In a preferred embodiment, the individualized compensation correction further includes: Based on the signal attenuation coefficients of each channel output by the individualized volume conductor digital twin model, the weighting coefficients of each acquisition channel in the multi-channel gastrointestinal electrical signal of the body surface are redistributed so that the channel with a higher weighting coefficient corresponds to the expected body surface projection area of ​​the gastric electrical activity of the target object. The signal compensation parameters were determined based on the weight loss surgery history and the metabolic drug history. If the weight loss surgery history indicates a history of laparoscopic sleeve gastrectomy, then the gastric electrophysiological structure offset model corresponding to the surgery is obtained. The gastric electrophysiological structure offset model includes the expected decrease range of the postoperative gastric electrical dominant frequency and the expected attenuation range of the gastric electrical amplitude. If the history of metabolic drugs indicates a history of GLP-1 receptor agonist use, then the drug impact factor is calculated based on the type, dosage, and duration of use. The threshold of the normal range of the gastric electrical dominant frequency is adjusted according to the gastric electrophysiological structure offset model, and the amplitude of the multi-channel gastrointestinal electrical signal on the body surface is normalized and compensated according to the drug influence factor.

[0012] In a preferred embodiment, the first path performs classification processing on the individualized compensated and corrected multi-channel gastrointestinal electrical signals from the body surface, including: Temporal features are extracted from the compensated and corrected multichannel gastrointestinal electrical signals of the body surface, and the temporal features include at least one of line length features, percentage of normal gastric electrical rhythm, and postprandial power ratio. The timing features are weighted and corrected using the signal attenuation coefficient to eliminate the attenuation effect of the abdominal fat layer on the signal amplitude. The weighted time-series characteristics were normalized twice using the drug effect factor to eliminate the superimposed modulation effect of GLP-1 receptor agonists on the slow wave rhythm of gastric electrical activity. The features after secondary normalization are input into the first classifier, which outputs the first probability value and the first confidence score.

[0013] In a preferred embodiment, the calculation of a multidimensional heterogeneity measure between the multichannel gastrointestinal electrical signals and the gastric magnetocardiogram signals in the third path includes: The two synchronously acquired signals are segmented using a sliding time window; Power spectral density analysis was performed on two signals within the same window to extract their respective dominant frequency values ​​and calculate the dominant frequency difference value, which was used to evaluate whether the multi-channel gastrointestinal electrical signals on the body surface experienced frequency shift due to attenuation by the fat layer. Calculate the waveform similarity coefficient between two signals; Perform Hilbert transform on the two signals respectively to extract the instantaneous phase, and calculate the phase synchronization deviation value; Statistical analysis is performed on the main frequency difference value, waveform similarity coefficient and phase synchronization deviation value for the entire time period to generate the multidimensional heterogeneous measure.

[0014] In a preferred embodiment, an inter-path consistency monitoring step is also included: Calculate the pairwise differences between the first probability value, the second probability value, and the third probability value; When any of the aforementioned differences exceeds a preset consistency threshold, a consistency anomaly flag is generated, and the possible causes of the differences are analyzed. The possible causes include at least the following: the signal-to-noise ratio of the multi-channel gastrointestinal electrical signal on the body surface is lower than a first preset value; the quality confidence estimate of the gastric magnetograph signal is lower than a second preset value; and the main frequency difference value in the multidimensional heterogeneity measurement is in a preset abnormal range. The functional dyspepsia auxiliary assessment information includes the consistency anomaly identifier and the analysis results of the possible causes.

[0015] In a preferred embodiment, the decision fusion module determines an output strategy based on the first confidence score, the second confidence score, and the third confidence score, and allocates auxiliary evaluation information to the corresponding output queues, outputting it with a corresponding output delay time, specifically including: When the first confidence score, the second confidence score, and the third confidence score are all higher than the first preset threshold, a weighted fusion output strategy is adopted to allocate the auxiliary evaluation information to the first-level output queue and output it with a first delay time. When the first confidence score, the second confidence score, and the third confidence score are all lower than the second preset threshold, a comprehensive output strategy is adopted and a review reminder is triggered. The auxiliary evaluation information is allocated to the third-level output queue and output with a third delay time, which is greater than the first delay time. When only one evaluation path has a confidence score higher than the third preset threshold while the confidence scores of other paths are lower than the fourth preset threshold, the output result of the high-confidence evaluation path is used alone to allocate the auxiliary evaluation information to the second-level output queue and output it with a second delay time, which is between the first delay time and the third delay time.

[0016] In a preferred embodiment, a closed-loop weight reallocation step is also included: Using a sliding time window as the unit, after each time window ends, the first confidence score, the second confidence score, and the third confidence score within the current window are recalculated. If the first confidence score of the current window decreases by more than a preset change threshold relative to the previous window, the weight coefficient of the first path in the dynamic fusion weight is reduced, and the weight coefficients of the second and third paths are increased accordingly; the adjustment range of the weight coefficient is positively correlated with the decrease range of the confidence score. The adjusted weighting coefficients are used for the fusion calculation in the next time window to form a closed-loop adaptive loop.

[0017] In a preferred embodiment, the obese population refers to a population with a body mass index greater than or equal to 28 kg / m²; the history of weight loss surgery includes at least one of laparoscopic sleeve gastrectomy, gastric bypass surgery, and gastric banding; the history of metabolic drugs includes at least information on the type, dosage, and duration of use of GLP-1 receptor agonists.

[0018] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention solves the core technical problems of systematic distortion of EGG signals and main frequency detection caused by abdominal fat layer in obese people. It breaks through the long-standing technical bias of existing EGG diagnostic equipment in excluding obese patients, and realizes the acquisition of reliable auxiliary assessment information under the condition that patients cannot maintain standard signal quality. It eliminates the systematic misjudgment of EGG diagnosis by fat layer, so that the signals entering the subsequent analysis stage can truly reflect the gastric electrophysiological state of the target subject.

[0019] The three-path parallel independent operation evaluation architecture constructed in this invention can retain all signal source information and automatically select the optimal evaluation strategy under different signal quality conditions, forming a three-level hierarchical response mechanism of high confidence rapid output, medium confidence optimal selection and trade-off, and low confidence review and reminder, thus optimizing the evaluation efficiency in high confidence scenarios; the closed-loop weight redistribution mechanism can dynamically adjust the fusion strategy to adapt to changes in signal quality, forming a closed-loop adaptive system, and constructing a full-chain cyclic evaluation system from signal acquisition to fusion decision and then to feedback.

[0020] The diagnostic method of this invention does not require expensive large-scale equipment and professional operators, which can significantly reduce the access cost for medical institutions to introduce high-end gastric motility assessment technology; it provides objective quantitative evidence for FD function assessment in obese people, helps to reduce unnecessary invasive diagnoses such as gastroscopy, and promotes the transformation of gastric motility assessment from symptom-based experience judgment to a multimodal quantitative intelligent auxiliary assessment paradigm; it can be embedded in portable and wearable devices, and can be promoted and popularized in primary hospitals, community health check-up scenarios, etc., providing physicians with non-invasive auxiliary assessment tools, helping to reduce the misuse of digestive system medical resources, and creating a profound impact on the development of intelligent auxiliary diagnosis and treatment technology for digestive system diseases. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0022] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1, please refer to Figure 1 As shown in this embodiment, a method for processing functional dyspepsia-assisted assessment information based on gastrointestinal electrical signals is applied to the auxiliary diagnosis of functional dyspepsia in obese individuals, including: Obtain individualized medical record data of the target subject, wherein the individualized medical record data includes at least two of the following: body shape parameters, history of weight loss surgery, and history of metabolic drugs; Three evaluation paths are constructed to run in parallel, including a first path, a second path, and a third path; the first path, the second path, and the third path run independently in parallel and do not exclude each other. The first path includes classifying the multi-channel gastrointestinal electrical signals on the body surface after individualized compensation and correction, and outputting a first probability value and a first confidence score; The second path includes classifying the normalized compensated gastric magnetoencephalogram signal and outputting a second probability value and a second confidence score. The third path includes calculating a multidimensional heterogeneity measure between the multi-channel gastrointestinal electrical signal on the body surface and the gastric magnetocardiogram signal, generating dynamic fusion weights based on the multidimensional heterogeneity measure, weighting and fusing the first probability value and the second probability value, and outputting a third probability value and a third confidence score. The first probability value and the first confidence score, the second probability value and the second confidence score, and the third probability value and the third confidence score are input into the decision fusion module; The decision fusion module determines the output strategy based on the first confidence score, the second confidence score, and the third confidence score, and allocates the auxiliary evaluation information to the corresponding output queue, outputting the functional dyspepsia auxiliary evaluation information with the corresponding output delay time.

[0025] In one embodiment, the method first acquires individualized medical record data of the target subject, which may include at least two of the following: body shape parameters, history of weight loss surgery, and history of metabolic drugs. For example, body shape parameters may include body mass index, waist circumference, etc.; history of weight loss surgery may include information such as whether sleeve gastrectomy was performed; and history of metabolic drugs may include whether GLP-1 receptor agonist drugs were taken. Then, three parallel evaluation paths can be constructed: a first path, a second path, and a third path. These three paths run independently in parallel and are not mutually exclusive. Specifically, the first path can classify the individualized compensated and corrected multi-channel gastrointestinal electrical signals from the body surface and output a first probability value and a first confidence score; the second path can classify the normalized compensated gastric magnetoencephalogram (MEG) signal and output a second probability value and a second confidence score; the third path can calculate a multidimensional heterogeneity measure between the multi-channel gastrointestinal electrical signals from the body surface and the MEG signal, generate dynamic fusion weights based on the multidimensional heterogeneity measure, and perform weighted fusion of the first probability value and the second probability value to output a third probability value and a third confidence score. Subsequently, the first, second, and third probability values ​​and confidence scores can be input into the decision fusion module. This module determines the output strategy based on the confidence scores and allocates the auxiliary assessment information to the corresponding output queues, outputting functional dyspepsia auxiliary assessment information with the corresponding output delay time. Through the above parallel and independent operation path design, the advantages of EGG and MGG can be fully utilized without losing any signal source information. At the same time, through confidence score-driven output queue scheduling, rapid output of high-confidence results and in-depth analysis of low-confidence results can be achieved, thereby significantly improving the accuracy and clinical usability of auxiliary assessment of functional dyspepsia in obese individuals.

[0026] In one embodiment, the body shape parameters may include at least one of body mass index, abdominal circumference, subcutaneous fat thickness, and visceral fat area. The individualized compensation correction may include constructing an individualized volumetric conductor digital twin model based on the body shape parameters. Specifically, the body shape parameters can be mapped to a baseline abdominal anatomy model, and parametric deformation can be used to adapt to the body shape characteristics of the target object.

[0027] For example, a standard 3D anatomical model of the abdomen can be pre-built. Then, based on parameters such as the target subject's body mass index and abdominal circumference, the standard model can be non-rigidly deformed to ensure its geometry and fat layer thickness distribution match the target subject. In the deformed model, electrophysiological simulation parameters for gastric dipoles (such as dipole position, orientation, and intensity) can be set, and the expected electric and magnetic field intensity distributions at each electrode location on the body surface can be calculated. This calculation can be achieved using numerical simulation methods such as the finite element method or the boundary element method. Then, the signal attenuation coefficient of each acquisition channel can be estimated based on the ratio of the expected electric field intensity distribution to a preset electric field intensity reference value, and the quality confidence estimate of the gastric magnetoencephalogram signal can be estimated based on the ratio of the expected magnetic field intensity distribution to a preset magnetic field intensity reference value. Finally, an output parameter set for an individualized volumetric conductor digital twin model can be generated based on the signal attenuation coefficient and the quality confidence estimate.

[0028] The channel signal attenuation coefficient is calculated using a normalized formula: ; in, This represents the dedicated signal attenuation coefficient corresponding to the i-th body surface acquisition electrode; This represents a fixed reference value for the baseline electric field strength corresponding to subjects with standard body types, which is a system preset constant. This represents the real-time expected electric field strength of the i-th electrode channel obtained from the simulation solution of the individualized volumetric conductor digital twin model. By constructing the individualized volumetric conductor digital twin model, the attenuation and distortion effects of abdominal adipose tissue on the propagation path of gastric electrical signals can be quantified into specific technical parameters, providing a numerical basis for subsequent electrode weight allocation and signal switching. The overall effect of this step is that it enables the mathematical recovery of gastric electrical signals that were originally obscured by the fat layer, significantly improving the usability of EGG signals in obese individuals.

[0029] In one embodiment, the method may further include an individualized fusion model arbitration step. Specifically, a corresponding dedicated fusion weight generator can be selected from a pre-built fusion model library based on the history of bariatric surgery and the history of metabolic drug use. The fusion model library may include: a first fusion weight generator corresponding to no history of bariatric surgery and no history of metabolic drug use; a second fusion weight generator corresponding to a history of laparoscopic sleeve gastrectomy; a third fusion weight generator corresponding to a history of GLP-1 receptor agonist use; and a fourth fusion weight generator corresponding to a history of both bariatric surgery and metabolic drug use. The network parameters of the first to fourth fusion weight generators may be different, and they are optimized for the complementary characteristics of multi-channel gastrointestinal electrical signals and gastric magnetoencephalogram signals under different interference modes.

[0030] For example, in patients with a history of sleeve gastrectomy, the normal pacing area on the greater curvature of the stomach is removed, resulting in a significant decrease in the dominant frequency of gastric electrical activity post-surgery. Therefore, the corresponding fusion weight generator can appropriately reduce the weight of the first path (EGG path) and increase the weight of the second path (MGG path). For patients with a history of GLP-1 receptor agonist use, the drug may cause superposition modulation of the slow wave rhythm of gastric electrical activity. Therefore, the corresponding fusion weight generator can introduce a drug influence factor to compensate for the amplitude of the EGG signal. By introducing individualized fusion model arbitration, physiological structural changes and pathological rhythm abnormalities can be distinguished, avoiding misclassification of normal post-operative changes or drug effects as functional decomposition (FD), thereby improving the assessment specificity for obese individuals with a history of surgery or medication use.

[0031] In one embodiment, the individualized compensation correction may further include the following steps. First, the weighting coefficients of each acquisition channel in the multi-channel gastrointestinal electrical signal on the body surface can be redistributed according to the signal attenuation coefficients of each channel output by the individualized volumetric conductor digital twin model, so that the channel with a higher weighting coefficient corresponds to the expected surface projection area of ​​the gastric electrical activity of the target object.

[0032] For example, if the signal attenuation coefficient at a certain electrode location is large, its weighting coefficient should be reduced; conversely, if the attenuation coefficient is small, its weighting coefficient should be increased. Secondly, signal compensation parameters can be determined based on the history of bariatric surgery and metabolic medications. If the history of bariatric surgery indicates a history of laparoscopic sleeve gastrectomy, a gastric electrophysiological structural shift model corresponding to that surgery can be obtained. This model can include the expected decrease range of the postoperative gastric electrical frequency and the expected attenuation range of the gastric electrical amplitude. If the history of metabolic medications indicates a history of GLP-1 receptor agonist use, a drug influence factor can be calculated based on the type, dosage, and duration of medication. Finally, the normal range threshold of the gastric electrical frequency can be adjusted based on the gastric electrophysiological structural shift model, and the amplitude of the multi-channel gastrointestinal electrical signals on the body surface can be normalized and compensated based on the drug influence factor.

[0033] The normalization compensation for the amplitude of gastrointestinal electrical signals is calculated using a standardized formula: ; Detailed explanation of formula letters: This represents the effective surface gastrointestinal electrical timing signal after amplitude compensation correction; This indicates raw gastrointestinal electrical signals directly acquired in situ without any calibration processing. The drug rhythm modulation influence factor corresponding to GLP-1 receptor agonists is defined as having a fixed value range of 0.75–1.25, which can be linearly fine-tuned according to the drug dosage. Through the above adaptive adjustment, various biases caused by obesity can be corrected from the signal acquisition end, making the signals entering the subsequent analysis stage more realistically reflect the gastric electrophysiological state of the target subject. This forms a closed-loop control system from perception to adjustment, significantly improving the adaptive capability and accuracy of signal preprocessing.

[0034] In one embodiment, the specific implementation of classifying and processing the individualized compensated and corrected multi-channel gastrointestinal electrical signals on the body surface in the first path can be as follows: First, temporal features are extracted from the compensated and corrected multi-channel gastrointestinal electrical signals from the body surface. These temporal features may include at least one of line length features, percentage of normal gastric electrical rhythms, and postprandial power ratio. Line length features reflect subtle and complex changes in the signal waveform, and studies have confirmed significant differences between FD patients and healthy controls. The percentage of normal gastric electrical rhythms assesses the regularity of the gastric electrical rhythm. The postprandial power ratio reflects the stomach's physiological responsiveness to food intake. Then, the temporal features can be weighted and corrected using the signal attenuation coefficient to eliminate the attenuation effect of the abdominal fat layer on the signal amplitude.

[0035] For example, if the attenuation coefficient of a certain channel is 0.4, the feature value contributed by that channel can be divided by 0.4 to recover its original amplitude. Next, the weighted corrected temporal features can be normalized twice using the drug influence factor to eliminate the superimposed modulation effect of GLP-1 receptor agonists on the slow wave rhythm of gastric electrical activity. Finally, the twice-normalized features are input into a first classifier (e.g., support vector machine, random forest, or lightweight neural network), which outputs a first probability value and a first confidence score.

[0036] The confidence score for the first path is calculated using a standardized, piecewise formula. ; in: This represents the FD-assisted evaluation quantitative confidence score output by the first path, with a value range constrained to the interval between 0 and 1. This represents the predicted probability of FD (digestive disease) positivity output by the first classifier, also normalized to the range of 0-1. Segmentation logic is used to symmetrically quantify the confidence level of the classification results; the greater the probability deviates from 0.5, the higher the confidence score. Through multiple corrections (digital twin correction + drug correction) and feature fusion, the true gastric electrical signals interfered with by obesity can be restored to the greatest extent possible, and objective classification is performed using a machine learning model. This provides a high-precision auxiliary assessment process when EGG is available, effectively reducing the false positive rate.

[0037] In one embodiment, the specific method for calculating the multidimensional heterogeneity measure between multichannel gastrointestinal electrical signals and gastric magnetoencephalogram signals in the third path can be: First, the two synchronously acquired signals are segmented using a sliding time window (e.g., window length 60 seconds, step size 30 seconds). Then, power spectral density analysis is performed on the two signals within the same window to extract their respective dominant frequencies, and the dominant frequency difference is calculated. This difference can be used to assess whether the frequency shift of multi-channel gastrointestinal electrical signals on the body surface is caused by attenuation due to the fat layer. Simultaneously, the waveform similarity coefficient between the two signals can be calculated, and the phase synchronization deviation can be calculated after extracting the instantaneous phase using Hilbert transform. Finally, statistical analysis is performed on the dominant frequency difference, waveform similarity coefficient, and phase synchronization deviation over the entire time period to generate a multidimensional heterogeneity measure.

[0038] For example, statistics such as the mean and standard deviation of each indicator can be calculated and combined into a multidimensional vector. This multidimensional heterogeneous metric can be used to drive the generation of dynamic fusion weights.

[0039] The dynamic weighted fusion of dual-path probabilities is calculated using a benchmark normalized formula: ; in: This represents the probability value of the dynamic fusion FD comprehensive evaluation of the final output of the third path; This represents the dynamic fusion weights of the first path, which are adaptively generated in real time by multidimensional heterogeneous metrics, and their values ​​are strictly constrained to be 0 ≤ ≤1; This represents the original probability value of the electrical signal classification output for the first path; This represents the raw probability value of the gastric magnetic signal classification output from the second path; To obtain complementary fusion weights for the second path obtained through adaptive matching, the summation of the two path weights is normalized throughout the process. When the difference in the dominant frequency is small and the waveform similarity is high, it indicates that the EGG signal is relatively reliable, and a higher fusion weight can be assigned to the first path; conversely, when the difference in the dominant frequency is large, it indicates that the EGG signal may be severely attenuated, and the weight of the second path (MGG path) should be increased. By introducing a multidimensional heterogeneous metric, the quality of the EGG signal can be objectively evaluated, providing a scientific numerical basis for subsequent dynamic fusion decisions, and achieving a leap from blindly relying on EGG to adaptive signal source selection.

[0040] In one embodiment, the method may further include an inter-path consistency monitoring step. Specifically, pairwise differences between a first probability value, a second probability value, and a third probability value may be calculated. When any difference value exceeds a preset consistency threshold, a consistency anomaly flag may be generated, and the possible causes of the difference may be analyzed.

[0041] The possible causes may include at least: the signal-to-noise ratio of multi-channel gastrointestinal electrical signals from the body surface is lower than a first preset value; the estimated quality confidence value of the gastric magnetogenomic signal is lower than a second preset value; and the dominant frequency difference value in the multidimensional heterogeneity measurement is within a preset abnormal range. The final output of the functional dyspepsia auxiliary assessment information may include this consistency anomaly identifier and the analysis results of the possible causes. By introducing inter-path consistency monitoring, the degree of divergence between the outputs of the three assessment pathways can be monitored in real time, and the interpretability of the diagnostic process can be provided while outputting auxiliary assessment information.

[0042] For example, when the conclusions of the three paths are highly consistent, the reliability of the output results is high; when significant discrepancies occur, the system can automatically analyze possible causes (such as poor EGG signal quality, poor MGG signal quality, or inconsistent main frequencies between the two), and provide this information to clinicians for reference, thereby avoiding giving misleading assessment conclusions when the signals are unreliable, and guiding clinicians to take more in-depth examination measures.

[0043] In one embodiment, the decision fusion module determines the output strategy based on the first confidence score, the second confidence score, and the third confidence score, and allocates the auxiliary evaluation information to the corresponding output queue in the following specific manner: When the first, second, and third confidence scores are all higher than the first preset threshold, a weighted fusion output strategy can be used to allocate the auxiliary assessment information to the first-level output queue and output it with the first delay time. This indicates that all three paths have high confidence, and the weighted fusion result has the highest reliability, allowing for rapid output for clinical reference. When the first, second, and third confidence scores are all lower than the second preset threshold, a comprehensive output strategy can be used and a review reminder can be triggered, allocating the auxiliary assessment information to the third-level output queue and outputting it with the third delay time, which is longer than the first delay time. This indicates that the confidence of all paths is low, the system needs more time for in-depth analysis, and further examinations (such as gastric emptying radionuclide imaging or gastroscopy) are recommended for the physician. When only one assessment path has a confidence score higher than the third preset threshold while the confidence scores of other paths are lower than the fourth preset threshold, the output result of that high-confidence assessment path can be used alone, allocating the auxiliary assessment information to the second-level output queue and outputting it with the second delay time, which is between the first and third delay times.

[0044] For example, when the confidence level of the EGG path (first path) is high while the confidence level of the MGG path is low, the output of the first path can be used alone. This hierarchical output mechanism reflects the reliability of the evaluation results in a more granular way, providing clinicians with richer decision support information. It forms a three-tiered output strategy: high-confidence direct output, low-confidence suggestion review, and medium-confidence optimal selection, ensuring efficiency in high signal-to-noise ratio scenarios while avoiding error risks in low signal-to-noise ratio scenarios.

[0045] In one embodiment, the method may further include a closed-loop weight reallocation step. Specifically, the first confidence score, second confidence score, and third confidence score within the current window can be recalculated after each sliding time window ends. If the first confidence score in the current window decreases by more than a preset change threshold relative to the previous window, the weight coefficient of the first path in the dynamic fusion weight can be reduced, while the weight coefficients of the second and third paths can be increased accordingly. The adjustment range of the weight coefficients can be positively correlated with the decrease in the confidence score. The adjusted weight coefficients can be used for the fusion calculation in the next time window, forming a closed-loop adaptive system. For example, when a patient's position changes during measurement, causing a sudden deterioration in the EGG signal, the first confidence score will decrease significantly. The system can automatically reduce the weight of the EGG path while increasing the weights of the MGG path and the fusion path, thereby ensuring the stability of the overall assessment results.

[0046] Among them, the closed-loop adaptive weight iterative update adopts two sets of linkage normalization formulas: ; ; in: This indicates the new first path fusion weight used in the next window after the iterative update; This indicates the original fusion weights currently in use during the sliding time window; This indicates that the step size for real-time adaptive weight adjustment dynamically follows changes in confidence level. This indicates the real-time decrease in the first confidence score of the current window compared to the previous window; This indicates that the system has a preset confidence warning threshold, which is fixed at 0.1. The maximum controllable adjustment range of the weight is fixed at 0.3 to prevent sudden weight changes from causing evaluation instability. This closed-loop weight redistribution mechanism allows for real-time and continuous tracking of signal quality changes caused by movement or breathing in obese individuals, and dynamic adjustment of the fusion weights based on the degree of change, achieving true "perception → fusion → decision → feedback" closed-loop control. Furthermore, the obese population can refer to individuals with a body mass index greater than or equal to 28 kg / m²; the weight-loss surgical history can include at least one of laparoscopic sleeve gastrectomy, gastric bypass surgery, and gastric banding; and the metabolic drug history can include at least the type, dosage, and duration of GLP-1 receptor agonist use. The strict definition of the obese population and the specific limitations on surgical and drug histories make the application targets and scope of protection of this invention clearer, facilitating accurate public understanding of the application boundaries of the solution.

[0047] In one embodiment, the first classifier can be constructed using any of several machine learning models, such as binary support vector machines, random forests, lightweight gradient boosting machines, or lightweight neural networks. Its specific structure can be appropriately selected based on the accuracy and efficiency requirements of clinical deployment. The following example illustrates a specific implementation using random forest as the first classifier.

[0048] The first classifier can be trained on a sample dataset of the target objects. The sample dataset can be constructed as follows: First, recruit at least 100 patients with functional dysplasia (FD) meeting the Rome IV diagnostic criteria and 100 healthy controls, with obese individuals (BMI ≥ 28 kg / m²) comprising at least 50% of the total sample. For each subject, their individualized medical record data and gastrointestinal electrical / magnetic signals were simultaneously collected according to the method described above. Then, the collected EGG signals were individually compensated and corrected according to the method described above, and the temporal features defined by the method were extracted (including line length features, percentage of normal gastric electrical rhythms, and postprandial power ratio, etc.). Each sample corresponds to a combination of "feature vector + label," where the feature vector contains multiple temporal feature values ​​after weighted correction and secondary normalization with drug influence factors, and the label is the FD diagnostic result (positive or negative) of the sample. This diagnostic result can be finally confirmed by a gastroenterologist based on the Rome IV criteria and endoscopic examination.

[0049] During model training, a five-fold cross-validation method can be used to evaluate the model's classification performance. The sample dataset is randomly divided into five folds, with four folds used as the training set and the remaining fold as the validation set in each round of training. The training process aims to minimize the Gini coefficient. The optimal splitting node is selected by traversing features and using a splitting threshold. When splitting a node, log2(M) features (M being the total number of features) are randomly selected as candidates to reduce the risk of overfitting. The number of decision trees can be set to an integer between 100 and 500, with the maximum depth of each tree set to an integer between 10 and 20 layers, and the minimum number of samples per leaf node set to an integer between 2 and 5. After the above training, the first classifier can output a probability value between 0 and 1, representing the probability that the target object belongs to FD positive, and also output a classification confidence score. The confidence score is calculated as follows: when the output probability is greater than 0.5, the confidence score equals (output probability - 0.5) × 2; when the output probability is less than or equal to 0.5, the confidence score equals (0.5 - output probability) × 2. In this way, the first classifier can output a first probability value and a first confidence score.

[0050] In one embodiment, the second classifier can be constructed using a machine learning model structure similar to that of the first classifier. The sample dataset for the second classifier is constructed as follows: For each subject, their gastric magnetoencephalography (MEG) signal is simultaneously acquired according to the method described above. Then, the quality confidence estimate of the MEG signal is estimated according to the method described above. The dominant frequency of slow waves, the percentage of normal slow wave rhythms, and the gastric electrical propagation velocity features in the MEG signal are extracted as feature vectors, labeled with the FD diagnosis result (positive or negative) of the sample. This diagnosis result is also confirmed by a gastroenterologist based on the Rome IV criteria combined with endoscopic examination. Unlike the first classifier, the training data for the second classifier can be stratified according to BMI, meaning that obese patients in different BMI strata can correspond to different classification thresholds.

[0051] Specifically, when the target individual's BMI is between 28 and 32, one set of classification thresholds can be used; when the BMI is between 32 and 40, another set can be used; and when the BMI is greater than or equal to 40, a third set can be used. The rationale for this stratified labeling is that existing studies on gastric electrocoagulation in obese individuals have shown that the percentage of normal slow-wave rhythms systematically decreases with increasing BMI. Therefore, the classification thresholds need to be adjusted according to BMI to establish a specific scoring model for obese individuals. After the above training, the second classifier can output a second probability value and a second confidence score.

[0052] In one embodiment, the fusion weight generator can be constructed using an attention mechanism network or a lightweight neural network. Taking an attention mechanism network as an example: The network's input is the multidimensional heterogeneous metric described in the method, including the main frequency difference value, waveform similarity coefficient, and phase synchronization deviation value, with at least three dimensions. The network's output is a dynamic fusion weight, a scalar value between 0 and 1, used to control the proportion of the first probability value and the second probability value in the weighted fusion process. Specifically, the fusion weight can be denoted as α, with the weight corresponding to the first probability value being α and the weight corresponding to the second probability value being 1-α.

[0053] The training process of the fusion weight generator is as follows: First, a training set containing a large number of data samples is constructed. Each sample contains a set of input-output pairs, where the input is a multidimensional heterogeneous metric and the output is the fusion weight label. The fusion weight label can be determined by: after the first classifier and the second classifier independently output the first probability value and the second probability value, respectively, determining the optimal fusion weight corresponding to the sample through expert judgment or cross-validation (so that the classification accuracy after weighted fusion is the highest). Then, the model is trained using the following objective function: Fusion probability value = α × +(1-α)× The training objective is to minimize the cross-entropy loss between the classification result corresponding to the fusion probability value and the true FD diagnostic label of the sample.

[0054] The model training cross-entropy loss function uses the normalized standard formula: ; in: This represents the real-time cross-entropy loss function value corresponding to a single training sample, used for backpropagation gradient optimization. This represents the gold standard label for the true clinical diagnosis of the sample, with discrete values ​​of only 0 or 1. =1 corresponds to a positive FD confirmed sample. =0 corresponds to a healthy negative control sample; This represents the fused prediction probability value output by the model in the current iteration round, with global normalization in the range of 0 to 1; This represents the natural logarithm operator, used to quantify the deviation between predicted values ​​and true labels. The training of the fusion weight generator can be performed iteratively using backpropagation and the Adam optimizer.

[0055] In one embodiment, the construction of the gastric electrophysiological structural shift model can be based on the statistical regularities of clinical data from patients after sleeve gastrectomy.

[0056] Specifically, clinical research data on gastric electrophysiology (GE) after at least 30 cases following laparoscopic sleeve gastrectomy were collected to extract changes in GE frequency and amplitude before and after surgery. The data showed that the GE frequency after sleeve gastrectomy significantly decreased from the normal value (approximately 3 Hz) to approximately 2.3 Hz (P < 0.001); the GE amplitude decreased from the normal value (approximately 31.5 μV) to approximately 14.8 μV (P < 0.001), and all enrolled patients exhibited abnormal electrical propagation patterns. Based on these clinical statistical patterns, an empirical model including mean and standard deviation can be constructed: the expected decrease in GE frequency can be set to 20% to 30% of the normal frequency, and the expected decrease in GE amplitude can be set to 40% to 60% of the normal amplitude. In practical applications, these ranges can be fine-tuned based on the target subjects' preoperative baseline data and individual circumstances.

[0057] In one embodiment, the first to fourth fusion weight generators in the fusion model library can be trained independently. The first fusion weight generator can be trained using sample data from individuals with no history of weight loss surgery or metabolic drugs; the second fusion weight generator can be trained using sample data from individuals with a history of laparoscopic sleeve gastrectomy but no history of GLP-1 drugs; the third fusion weight generator can be trained using sample data from individuals with a history of GLP-1 drugs but no history of weight loss surgery; and the fourth fusion weight generator can be trained using sample data from individuals with a history of both weight loss surgery and GLP-1 drugs. The network structure of each fusion weight generator can be the same, but the network parameters can be optimized independently to adapt to the complementary characteristics of EGG and MGG signals under different interference modes.

[0058] For example, for samples with a history of weight loss surgery, the reliability of EGG signals is often systematically low. Therefore, the output weight α obtained by training the second fusion weight generator tends to be smaller, while the weight 1-α of the second path (MGG path) is correspondingly larger. For samples with a history of GLP-1 medication, the amplitude of the EGG signal may be suppressed by the drug. The drug influence factor has already been pre-corrected for the amplitude, so the output weight α of the third fusion weight generator can be close to that of the first fusion weight generator. Through this individualized fusion model arbitration, the optimal fusion strategy can be automatically selected based on the target subject's history of weight loss surgery and metabolic drug use.

[0059] In one embodiment, the initial value of the dynamic fusion weight in the closed-loop weight redistribution can be set to α=0.5. This weight value can be dynamically adjusted based on real-time feedback of the confidence score. Specifically, within the current sliding time window, the first confidence score is calculated. Second confidence score ,like If the decrease exceeds a preset change threshold (e.g., δ=0.1) compared to the previous time window, then the weight α is reduced, and the new weight... = ×(1-η), where η is the adjustment step size, which can be set as ( (Decrease value / preset threshold) × , The maximum adjustment range can be set, for example, to 0.3. Simultaneously, the weight coefficients of the second and third paths are increased accordingly to ensure the sum of the fusion weights is normalized. This closed-loop automated adjustment process can track changes in signal quality in real time, improving the robustness of the fusion system.

[0060] In one embodiment, the parameters and data dimensions (such as the number of samples, the number of decision trees, the number of cross-validation folds, etc.) in the above-described classifier and training process are merely illustrative examples. Those skilled in the art can make appropriate adjustments based on actual clinical needs and computing resources, without departing from the scope of protection of this invention.

[0061] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for processing FD-assisted assessment information based on gastrointestinal electrical signals, characterized in that, Applications in the auxiliary diagnosis of functional dyspepsia in obese individuals include: Obtain individualized medical record data of the target subject, wherein the individualized medical record data includes at least two of the following: body shape parameters, history of weight loss surgery, and history of metabolic drugs; Three evaluation paths are constructed to run in parallel, including a first path, a second path, and a third path; the first path, the second path, and the third path run independently in parallel and do not exclude each other. The first path includes classifying the multi-channel gastrointestinal electrical signals on the body surface after individualized compensation and correction, and outputting a first probability value and a first confidence score; The second path includes classifying the normalized compensated gastric magnetoencephalogram signal and outputting a second probability value and a second confidence score. The third path includes calculating a multidimensional heterogeneity measure between the multi-channel gastrointestinal electrical signal on the body surface and the gastric magnetocardiogram signal, generating dynamic fusion weights based on the multidimensional heterogeneity measure, weighting and fusing the first probability value and the second probability value, and outputting a third probability value and a third confidence score. The first probability value and the first confidence score, the second probability value and the second confidence score, and the third probability value and the third confidence score are input into the decision fusion module; The decision fusion module determines the output strategy based on the first confidence score, the second confidence score, and the third confidence score, and allocates the auxiliary evaluation information to the corresponding output queue, outputting the functional dyspepsia auxiliary evaluation information with the corresponding output delay time.

2. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 1, characterized in that: The body shape parameters include at least one of body mass index, waist circumference, subcutaneous fat thickness, and visceral fat area; The individualized compensation correction includes constructing an individualized volumetric conductor digital twin model based on the body shape parameters, specifically including: The body shape parameters are mapped to a baseline abdominal anatomical model, and the body shape characteristics of the target object are adapted through parameterized deformation. In the deformed model, the electrophysiological simulation parameters of the gastric dipole were set, and the expected electric field intensity distribution and expected magnetic field intensity distribution at each electrode position on the body surface were calculated. The signal attenuation coefficient of each acquisition channel is estimated based on the ratio of the expected electric field intensity distribution to the preset electric field intensity reference value, and the quality confidence estimate of the gastromagnetic signal is estimated based on the ratio of the expected magnetic field intensity distribution to the preset magnetic field intensity reference value. The output parameter set of the individualized volumetric conductor digital twin model is generated based on the signal attenuation coefficient and the quality confidence estimate.

3. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 2, characterized in that: It also includes the arbitration step of the individualized fusion model: Based on the history of weight loss surgery and the history of metabolic drugs, a corresponding dedicated fusion weight generator is selected from a pre-built fusion model library; The fusion model library includes: a first fusion weight generator corresponding to no history of weight loss surgery and no history of metabolic drugs; a second fusion weight generator corresponding to a history of laparoscopic sleeve gastrectomy; a third fusion weight generator corresponding to a history of GLP-1 receptor agonist use; and a fourth fusion weight generator corresponding to a history of both weight loss surgery and metabolic drugs. The network parameters of the first to fourth fusion weight generators are different, and they are optimized for the complementary characteristics of the multi-channel gastrointestinal electrical signals and the gastric magnetocardiogram signals under different interference modes.

4. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 3, characterized in that: The individualized compensation correction also includes: Based on the signal attenuation coefficients of each channel output by the individualized volume conductor digital twin model, the weighting coefficients of each acquisition channel in the multi-channel gastrointestinal electrical signal of the body surface are redistributed so that the channel with a higher weighting coefficient corresponds to the expected body surface projection area of ​​the gastric electrical activity of the target object. The signal compensation parameters were determined based on the weight loss surgery history and the metabolic drug history. If the weight loss surgery history indicates a history of laparoscopic sleeve gastrectomy, then the gastric electrophysiological structure offset model corresponding to the surgery is obtained. The gastric electrophysiological structure offset model includes the expected decrease range of the postoperative gastric electrical dominant frequency and the expected attenuation range of the gastric electrical amplitude. If the history of metabolic drugs indicates a history of GLP-1 receptor agonist use, then the drug impact factor is calculated based on the type, dosage, and duration of use. The threshold of the normal range of the gastric electrical dominant frequency is adjusted according to the gastric electrophysiological structure offset model, and the amplitude of the multi-channel gastrointestinal electrical signal on the body surface is normalized and compensated according to the drug influence factor.

5. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 1, characterized in that: The first path involves classifying and processing the individualized compensated and corrected multi-channel gastrointestinal electrical signals from the body surface, including: Temporal features are extracted from the compensated and corrected multichannel gastrointestinal electrical signals of the body surface, and the temporal features include at least one of line length features, percentage of normal gastric electrical rhythm, and postprandial power ratio. The timing features are weighted and corrected using the signal attenuation coefficient to eliminate the attenuation effect of the abdominal fat layer on the signal amplitude. The weighted time-series characteristics were normalized twice using the drug effect factor to eliminate the superimposed modulation effect of GLP-1 receptor agonists on the slow wave rhythm of gastric electrical activity. The features after secondary normalization are input into the first classifier, which outputs the first probability value and the first confidence score.

6. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 1, characterized in that: The third path calculates the multidimensional heterogeneity measure between the multi-channel gastrointestinal electrical signals and the gastric magnetocardiogram signals, including: The two synchronously acquired signals are segmented using a sliding time window; Power spectral density analysis was performed on two signals within the same window to extract their respective dominant frequency values ​​and calculate the dominant frequency difference value, which was used to evaluate whether the multi-channel gastrointestinal electrical signals on the body surface experienced frequency shift due to attenuation by the fat layer. Calculate the waveform similarity coefficient between two signals; Perform Hilbert transform on the two signals respectively to extract the instantaneous phase, and calculate the phase synchronization deviation value; Statistical analysis is performed on the main frequency difference value, waveform similarity coefficient and phase synchronization deviation value for the entire time period to generate the multidimensional heterogeneous measure.

7. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 1, characterized in that: It also includes inter-path consistency monitoring steps: Calculate the pairwise differences between the first probability value, the second probability value, and the third probability value; When any of the aforementioned differences exceeds a preset consistency threshold, a consistency anomaly flag is generated, and the possible causes of the differences are analyzed. The possible causes include at least the following: the signal-to-noise ratio of the multi-channel gastrointestinal electrical signal on the body surface is lower than a first preset value; the quality confidence estimate of the gastric magnetograph signal is lower than a second preset value; and the main frequency difference value in the multidimensional heterogeneity measurement is in a preset abnormal range. The functional dyspepsia auxiliary assessment information includes the consistency anomaly identifier and the analysis results of the possible causes.

8. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 1, characterized in that: The decision fusion module determines the output strategy based on the first confidence score, the second confidence score, and the third confidence score, and allocates the auxiliary evaluation information to the corresponding output queue, outputting it with the corresponding output delay time. Specifically, this includes: When the first confidence score, the second confidence score, and the third confidence score are all higher than the first preset threshold, a weighted fusion output strategy is adopted to allocate the auxiliary evaluation information to the first-level output queue and output it with a first delay time. When the first confidence score, the second confidence score, and the third confidence score are all lower than the second preset threshold, a comprehensive output strategy is adopted and a review reminder is triggered. The auxiliary evaluation information is allocated to the third-level output queue and output with a third delay time, which is greater than the first delay time. When only one evaluation path has a confidence score higher than the third preset threshold while the confidence scores of other paths are lower than the fourth preset threshold, the output result of the high-confidence evaluation path is used alone to allocate the auxiliary evaluation information to the second-level output queue and output it with a second delay time, which is between the first delay time and the third delay time.

9. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 1, characterized in that: It also includes the closed-loop weight redistribution step: Using a sliding time window as the unit, after each time window ends, the first confidence score, the second confidence score, and the third confidence score within the current window are recalculated. If the first confidence score of the current window decreases by more than a preset change threshold relative to the previous window, the weight coefficient of the first path in the dynamic fusion weight is reduced, and the weight coefficients of the second and third paths are increased accordingly; the adjustment range of the weight coefficient is positively correlated with the decrease range of the confidence score. The adjusted weighting coefficients are used for the fusion calculation in the next time window to form a closed-loop adaptive loop.

10. The method for processing FD-assisted assessment information based on gastrointestinal electrical signals according to claim 1, characterized in that: The obese population refers to individuals with a body mass index greater than or equal to 28 kg / m²; the history of weight loss surgery includes at least one of laparoscopic sleeve gastrectomy, gastric bypass surgery, and gastric banding; the history of metabolic drugs includes at least the type, dosage, and duration of use of GLP-1 receptor agonists.

Images