KOA rehabilitation condition monitoring method based on intestinal flora and multi-source data collaboration

Abstract

The invention provides a KOA rehabilitation condition monitoring method based on intestinal flora and multi-source data collaboration, and relates to the technical field of rehabilitation monitoring, the method comprises the following steps: screening KOA patients and healthy subjects, obtaining intestinal flora data, inflammatory factor data and electroencephalogram data, and obtaining electroencephalogram feature information; determining physiological state change information and rehabilitation condition information of the KOA patient in each treatment course, further determining an intervention effectiveness evaluation index, and obtaining a rehabilitation condition monitoring result. According to the method and the system, a healthy subject can be screened as a reference, and whether the physiological state of the KOA patient is changed or not and whether the KOA patient is rehabilitated or not is judged by monitoring intestinal flora, inflammatory factors and electroencephalogram data of the KOA patient, so that the rehabilitation condition of the KOA patient is accurately analyzed, and a plurality of KOA patients can be diagnosed based on the rehabilitation condition of the KOA patient. Whether the intervention treatment on the KOA patient is effective or not is determined, and an accurate data basis is provided for analysis of the intervention effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rehabilitation monitoring, and in particular to a KOA rehabilitation condition monitoring method based on intestinal flora and multi-source data collaboration. BACKGROUND

[0002] Osteoarthritis (OA) is a degenerative chronic joint disease that mainly damages joint cartilage, leading to pain, swelling and stiffness around the joint, and is one of the main causes of disability in the elderly population. The disease mainly involves weight-bearing joints, of which knee osteoarthritis (KOA) is the most common. It is estimated that KOA accounts for about 60% of the total disability burden of osteoarthritis worldwide; it is estimated that the number of KOA cases will increase by 74.9% by 2050 compared to 2020. Among them, the prevalence of symptomatic KOA is 8.1%, and the years lived with disability (YLD) reaches 1.97 million people / year, which has brought a heavy burden to society and individuals. Acupuncture treatment of KOA has become an important part of the key technology research of "basic treatment" in its stepwise treatment. However, the related technology is difficult to quantitatively evaluate the effect of acupuncture treatment and the rehabilitation status of patients. SUMMARY

[0003] The present application provides a KOA rehabilitation condition monitoring method based on intestinal flora and multi-source data collaboration, which can solve the technical problem that the related technology is difficult to quantitatively evaluate the effect of acupuncture treatment and the rehabilitation status of patients.

[0004] According to a first aspect of the present application, a KOA rehabilitation condition monitoring method based on intestinal flora and multi-source data collaboration is provided, comprising: screening KOA patients and healthy subjects according to a preset screening rule; obtaining first intestinal flora data of the healthy subjects and second intestinal flora data of the KOA patients, and first inflammatory factor data of the healthy subjects and second inflammatory factor data of the KOA patients; at the end of each treatment course in the process of intervention on the KOA patients, obtaining third intestinal flora data, third inflammatory factor data and electroencephalogram data of the KOA patients, respectively; processing the electroencephalogram data by an electroencephalogram data processing model to obtain electroencephalogram feature information; determining physiological state change information of the KOA patients according to the electroencephalogram feature information, the second intestinal flora data, the second inflammatory factor data, the third intestinal flora data and the third inflammatory factor data; According to the physiological state change information of the KOA patient and the first intestinal flora data and the first inflammatory factor data, the rehabilitation condition information of the KOA patient is determined; According to the rehabilitation condition information of the KOA patient, the intervention effectiveness evaluation index is determined; According to the intervention effectiveness evaluation index, the rehabilitation condition monitoring result is determined.

[0005] According to the present application, the electroencephalogram data is processed by the electroencephalogram data processing model to obtain the electroencephalogram feature information, including: Obtaining the electroencephalogram data of multiple detection points of multiple brain regions of the KOA patient; The electroencephalogram data of each detection point is analyzed in frequency domain to obtain the frequency domain feature vector of each detection point; The number of detection points in the brain region is determined; If the number of detection points in the brain region is greater than 1, the weighted phase lag index between each detection point is determined according to the electroencephalogram data of each detection point in the brain region; According to the weighted phase lag index, the adjacency matrix corresponding to the brain region is determined; According to the subgraph model, the frequency domain feature vector and the adjacency matrix of each detection point in the brain region are processed to obtain the first output vector of each detection point in the brain region; The first output vector of each detection point in the brain region is averaged to obtain the brain region feature vector; If the number of detection points in the brain region is equal to 1, the frequency domain feature vector of the detection point is determined as the brain region feature vector; The brain region feature vectors of the multiple brain regions are processed by the graph attention network model to obtain the second output vector of each brain region; The second output vector of each brain region is input into the first multi-layer perception network level to obtain the electroencephalogram feature information of the KOA patient.

[0006] According to the present application, according to the electroencephalogram feature information, the second intestinal flora data, the second inflammatory factor data, the third intestinal flora data and the third inflammatory factor data, the physiological state change information of the KOA patient is determined, including: The third intestinal flora data of the current course is processed by the second multi-layer perception network level to obtain the second intestinal flora feature information; The third inflammatory factor data of the current course is processed by the third multi-layer perception network level to obtain the second inflammatory factor feature information; The second intestinal flora feature information, the second inflammatory factor feature information and the electroencephalogram feature information are processed by the first cross attention mechanism to obtain the second flora influence feature information and the second inflammatory factor influence feature information; The second flora influence feature information is spliced with the second inflammation factor influence feature information to obtain physiological state description information of the current course of treatment; The physiological state description information of the current course of treatment and the first hidden state information of the previous course of treatment are input into a physiological state change prediction model to obtain the first hidden state information of the current course of treatment, wherein if the current course of treatment is the first course of treatment, the first hidden state information of the previous course of treatment is the first hidden state information obtained according to the second intestinal flora data and the second inflammation factor data before the start of the course of treatment, and when the first hidden state information before the start of the course of treatment is obtained, a zero vector of the input first hidden state information; The first hidden state information of the current course of treatment is input into a fourth multi-layer perception network level to obtain physiological state change information of the KOA patient.

[0007] According to the present application, the rehabilitation condition information of the KOA patient is determined according to the physiological state change information of the KOA patient and the first intestinal flora data and the first inflammation factor data, comprising: The first intestinal flora data and the second intestinal flora data are clustered to determine one or more cluster clusters containing the least number of second intestinal flora data, and the first intestinal flora data in the one or more cluster clusters is averaged to obtain reference flora data, and the reference flora data is processed through a second multi-layer perception network level to obtain reference flora feature information; The first inflammation factor data and the second inflammation factor data are clustered to determine one or more cluster clusters containing the least number of second inflammation factor data, and the first inflammation factor data in the one or more cluster clusters is averaged to obtain reference inflammation factor data, and the reference inflammation factor data is processed through a third multi-layer perception network level to obtain reference inflammation factor feature information; The first hidden state information of the previous course of treatment and the second hidden state information of the previous course of treatment are spliced to obtain comprehensive hidden state information of the previous course of treatment; The comprehensive hidden state information of the previous course of treatment is processed through a fifth multi-layer perception network level to obtain input hidden state information; The physiological state change information of the current course of treatment, the reference flora feature information and the reference inflammation factor feature information are spliced to obtain treatment effectiveness feature information of the current course of treatment; The input hidden state information and the treatment effectiveness feature information of the current course of treatment are processed through a rehabilitation condition prediction model to obtain the second hidden state information of the current course of treatment, wherein if the current course of treatment is the first course of treatment, the second hidden state information of the previous course of treatment is the second hidden state information obtained according to the second intestinal flora data and the second inflammation factor data before the start of the course of treatment, and when the second hidden state information before the start of the course of treatment is obtained, a zero vector of the input second hidden state information; The second hidden state information of the current course is input into the sixth multi-layer perception network level for processing to obtain the rehabilitation condition information of the KOA patient.

[0008] According to the present application, the method further comprises: obtaining sample electroencephalogram signals, sample intestinal flora data and sample inflammatory factor data of a sample KOA patient at multiple rehabilitation stages, and labeled rehabilitation conditions at the multiple rehabilitation stages; processing the sample electroencephalogram signals at the i-th rehabilitation stage through the sub-graph model, the graph attention network model and the first multi-layer perception network level to obtain sample electroencephalogram feature information at the i-th rehabilitation stage; obtaining sample physiological state change information at the i-th rehabilitation stage according to the second multi-layer perception network level, the third multi-layer perception network level, the fourth multi-layer perception network level, the first cross-attention mechanism, the physiological state change prediction model, the sample electroencephalogram feature information at the i-th rehabilitation stage, the sample intestinal flora data and the sample inflammatory factor data; determining a state change loss function according to the sample physiological state change information at the multiple rehabilitation stages; obtaining predicted rehabilitation condition information at the i-th rehabilitation stage according to the rehabilitation condition prediction model, the fifth multi-layer perception network level and the sixth multi-layer perception network level, the reference flora data, the reference inflammatory factor data and the sample physiological state change information; determining a rehabilitation condition loss function according to the predicted rehabilitation condition information at each rehabilitation stage and the labeled rehabilitation conditions; obtaining a training loss function according to the rehabilitation condition loss function and the state change loss function; training the sub-graph model, the graph attention network model, the first multi-layer perception network level, the second multi-layer perception network level, the third multi-layer perception network level, the fourth multi-layer perception network level, the first cross-attention mechanism, the physiological state change prediction model, the rehabilitation condition prediction model, the fifth multi-layer perception network level and the sixth multi-layer perception network level according to the training loss function.

[0009] According to the present application, the state change loss function is determined according to the sample physiological state change information at the multiple rehabilitation stages, comprising: inputting the sample physiological state change information into the seventh multi-layer perception network level to obtain sample rehabilitation condition information; according to the formula determining the state change loss function wherein, is the labeled rehabilitation condition at the i-th rehabilitation stage, is the labeled rehabilitation condition at the i-1-th rehabilitation stage, sample rehabilitation status information of an i-th rehabilitation stage, sample rehabilitation status information of an (i-1)-th rehabilitation stage, n is a number of rehabilitation stages, and is a preset weight.

[0010] According to the present application, the rehabilitation status loss function is determined according to the predicted rehabilitation status information of each rehabilitation stage and the labeled rehabilitation status, comprising: According to the formula determining the rehabilitation status loss function wherein, is the predicted rehabilitation status information of the i-th rehabilitation stage, is the predicted rehabilitation status information of the (i-1)-th rehabilitation stage, and is a preset weight, and if is a conditional function.

[0011] According to the present application, the intervention effectiveness evaluation index is determined according to the rehabilitation status information of the KOA patient, comprising: obtaining an average value of rehabilitation status information of a plurality of KOA patients, and a historical average value of rehabilitation status information of the plurality of KOA patients at the end of a previous course; determining a relative gap between the average value of the rehabilitation status information and the historical average value; determining the relative gap as the intervention effectiveness evaluation index.

[0012] According to the second aspect of the present application, a KOA rehabilitation status monitoring system based on intestinal flora and multi-source data collaboration is provided, comprising: a screening module for screening KOA patients and healthy subjects according to a preset screening rule; an acquisition module for acquiring first intestinal flora data of healthy subjects and second intestinal flora data of KOA patients, and first inflammatory factor data of healthy subjects and second inflammatory factor data of KOA patients; a detection module for acquiring third intestinal flora data, third inflammatory factor data and electroencephalogram data of KOA patients at the end of each course during the intervention of KOA patients; an electroencephalogram feature module for processing electroencephalogram data through an electroencephalogram data processing model to obtain electroencephalogram feature information; a physiological state change module for determining physiological state change information of KOA patients according to the electroencephalogram feature information, the second intestinal flora data, the second inflammatory factor data, the third intestinal flora data and the third inflammatory factor data; The rehabilitation condition module is configured to determine rehabilitation condition information of the KOA patient according to the physiological state change information of the KOA patient and the first intestinal flora data and the first inflammatory factor data. The evaluation module is configured to determine an intervention effectiveness evaluation index according to the rehabilitation condition information of the KOA patient. The result module is configured to determine a rehabilitation condition monitoring result according to the intervention effectiveness evaluation index.

[0013] By adopting the technical solutions described above, the present application can achieve the following technical effects: According to the present application, healthy subjects can be screened as a reference, and by monitoring various data such as gut flora, inflammatory factors and EEG data of KOA patients, it can be determined whether the physiological state of KOA patients changes and whether KOA patients are recovering, so as to accurately analyze the recovery status of KOA patients, and based on the recovery status of multiple KOA patients, it can be determined whether the intervention treatment for KOA patients is effective, and accurate data basis is provided for the analysis of intervention effect. And through the sub-graph model, the first output vector of the detection point can be fused with the data features of each detection point in the brain area, and the important data features of the brain area for the reaction of KOA can be highlighted, the description accuracy of the EEG data for the reaction of KOA can be improved, and through the graph attention network model, the second output vector of the brain area can be fused with the data features of the whole brain for the reaction of KOA, and the important data features of the brain for the reaction of KOA can be highlighted, so as to accurately express the whole brain for the reaction of the feeling caused by KOA, and the accuracy of feature expression can be improved. In combination with the physiological characteristics of each course in the past, the physiological state change prediction model can be used to determine whether the recovery status of KOA patients changes and whether it changes towards the direction away from the initial state, and data basis is provided for determining the effectiveness of intervention. And by screening the cluster with the least number of second gut flora data and second inflammatory factor data after clustering, the feature significance of the physiological state of healthy subjects can be improved, which can provide data basis for verifying whether the physiological state of KOA patients is recovering and whether the intervention is effective. And by splicing the first hidden state information and the second hidden state information of the previous course, it can be described whether the physiological state of KOA patients changes after the end of the previous course and whether it changes towards the healthy state, and it can be used as a reference and constraint for the direction of physiological state change in the training process, and the training efficiency and effect can be improved. And the reference flora feature information and the reference inflammatory factor feature information can be used as reference information of the healthy state, which provides a basis for judging whether the physiological state of KOA patients changes towards the healthy state, and provides a constraint for the training process, which helps to further improve the training efficiency and effect. In the training process, the seventh multi-layer perception network level can be used for auxiliary training, and when determining the state change loss function, the error of the labeled recovery status and the sample recovery status information, and the error of the change rule of the labeled recovery status and the sample recovery status information are used to construct the loss function, which can further constrain the prediction rule of the physiological state change prediction model for the sample physiological state change information, and improve the training precision and efficiency. And by setting the condition function, the physiological state change prediction model and the recovery status prediction model can be compared and competed in terms of accuracy, the model training intensity can be improved, the prediction accuracy of the predicted recovery status information and its change trend can be improved, and the prediction accuracy and training efficiency of each model can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 An exemplary flowchart of a KOA rehabilitation status monitoring method based on gut microbiota and multi-source data synergy according to an embodiment of the present invention is shown. Figure 2 An exemplary schematic diagram illustrating the determination of rehabilitation status information according to an embodiment of the present invention is shown; Figure 3 A block diagram of a KOA rehabilitation status monitoring system based on gut microbiota and multi-source data synergy, according to an embodiment of the present invention, is shown as an example. Detailed Implementation

[0015] 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0017] Figure 1 An exemplary flowchart illustrates a method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy according to an embodiment of the present invention, the method comprising: Step S1: Screen KOA patients and healthy subjects according to preset screening rules; Step S2: Obtain the first gut microbiota data of healthy subjects and the second gut microbiota data of KOA patients, as well as the first inflammatory factor data of healthy subjects and the second inflammatory factor data of KOA patients; Step S3: At the end of each treatment course during the intervention of KOA patients, the third gut microbiota data, third inflammatory factor data and EEG data of KOA patients are obtained respectively. Step S4: Process the EEG data using the EEG data processing model to obtain EEG feature information; Step S5: Based on the electroencephalogram (EEG) characteristic information, the second gut microbiota data, and the second inflammatory factor data, determine the physiological state changes of the KOA patient; Step S6: Determine the recovery status information of KOA patients based on the physiological status change information, first gut microbiota data, and first inflammatory factor data. Step S7: Determine the indicators for evaluating the effectiveness of the intervention based on the rehabilitation status information of KOA patients; Step S8: Determine the monitoring results of rehabilitation status based on the intervention effectiveness assessment indicators.

[0018] The KOA rehabilitation monitoring method based on gut microbiota and multi-source data synergy according to embodiments of the present invention can screen healthy subjects as references and determine whether the physiological state of KOA patients has changed and whether KOA patients are recovering by monitoring various data such as gut microbiota, inflammatory factors and electroencephalogram data. This allows for accurate analysis of the rehabilitation status of KOA patients and, based on the rehabilitation status of multiple KOA patients, determines whether the intervention treatment for KOA patients is effective, providing an accurate data basis for the analysis of intervention effects.

[0019] According to an embodiment of the present invention, in step S1, patients with knee osteoarthritis (KOA) and healthy subjects can be screened. When screening KOA patients, it can be first determined whether the patient has been diagnosed with KOA. According to the Western medical diagnostic criteria, the following diagnostic conditions are included: (1) knee pain for most of the past month; (2) X-ray showing osteophyte formation; (3) joint fluid examination consistent with osteoarthritis; (4) age ≥ 40 years; (5) morning stiffness ≤ 30 min; (6) bone crepitus. Among them, if conditions (1) and (2) are met simultaneously, or conditions (1), (3), (5), and (6) are met simultaneously, or conditions (1), (4), (5), and (6) are met simultaneously, then the patient can be diagnosed with KOA. According to the diagnostic criteria of traditional Chinese medicine, if a patient experiences dull pain in the joints of the limbs, soreness and weakness in the lower back and knees, difficulty in bending and straightening the lower back and legs, difficulty in turning over when lying on their back, and the pain worsens with changes in weather, is milder during the day and more severe at night, increases with cold and is slightly relieved by heat, or is accompanied by dizziness, tinnitus, hearing loss, and vertigo; and has a pale red tongue with a thin white coating and a deep, thready, and slow pulse, then they can be diagnosed with KOA.

[0020] According to one embodiment of the present invention, patients diagnosed with KOA can be screened based on preset screening rules. The inclusion criteria in the preset screening conditions include: age 40-70 years and body mass index ≤30 kg / m². 2 Patients meeting all of the following criteria can be selected as KOA patients: K / L classification (Kellgren-Lawrence radiological diagnostic criteria) grade 2 or 3; right-handed (using the Edinburgh Hand Bias Questionnaire); average pain score >2 on the BPI pain scale; Mini-Mental State Examination (MMSE) score ≥24; BDI score <14; informed consent and voluntary participation.

[0021] According to one embodiment of the present invention, when screening healthy subjects, the inclusion criteria in the preset screening conditions include: good health, no history of KOA; age 40-70 years, and body mass index ≤30 kg / m². 2Individuals must possess clear verbal communication skills to understand and cooperate with the research; be willing to participate and sign an informed consent form. Those who meet all of the above criteria can be selected as healthy participants.

[0022] According to one embodiment of the present invention, the preset screening conditions during the screening process also include exclusion conditions. That is, even if the above conditions are met, if the person meets the following exclusion conditions, they should still be excluded from the observed KOA patients and healthy subjects. The exclusion conditions include: bleeding tendency; history of knee joint surgery within 6 months or intra-articular injection of corticosteroids within 3 months; knee pain related to rheumatic diseases or caused by inflammation; primary or secondary muscle diseases (such as idiopathic inflammatory myopathy, progressive muscular dystrophy, glycogen storage disease, etc.); severe internal or external deformities of the knee joint; abnormal mental state, lack of autonomous behavior, and non-cooperation with the examination; joint instability or lower limb alignment caused by severe knee joint trauma or lower limb fracture, or joint surface unevenness caused by near-joint fracture; serious cardiovascular and cerebrovascular diseases, musculoskeletal system diseases, severe organ failure, and other contraindications to exercise. If any of the above conditions are met, the person should be excluded from the observed KOA patients and healthy subjects.

[0023] According to one embodiment of the present invention, during participation, adverse reactions may occur in observed KOA patients or healthy subjects, and the observation of these KOA patients or healthy subjects may be terminated. The conditions for termination or exclusion include the following: Exclusion conditions: Subjects who are mistakenly included but do not meet the inclusion criteria; subjects who meet the inclusion criteria but have not received intervention. Termination conditions: Subjects who experience serious adverse events and are no longer suitable for this study; subjects who voluntarily withdraw from this study; subjects who experience various serious complications or worsening of their condition during the study and require emergency treatment; subjects who do not receive treatment as required or whose observation data is incomplete and affects the assessment.

[0024] According to one embodiment of the present invention, in step S2, KOA affects the patient's electroencephalogram (EEG) signals, inflammatory factors, and gut microbiota due to factors such as pain and inflammation. Therefore, the patient's recovery level can be determined based on EEG signals, inflammatory factors, and gut microbiota data, and it can also be used to evaluate the treatment effect.

[0025] According to one embodiment of the present invention, before the start of the treatment course, first gut microbiota data of healthy subjects and second gut microbiota data of KOA patients, as well as first inflammatory factor data of healthy subjects and second inflammatory factor data of KOA patients, can be obtained. Furthermore, in step S3, at the end of each treatment course, third gut microbiota data and third inflammatory factor data of KOA patients can be obtained.

[0026] According to one embodiment of the present invention, the collection method for gut microbiota data (e.g., first gut microbiota data, second gut microbiota data, and third gut microbiota data) is as follows: First, stool samples are collected. One week prior to sample collection, patients should not consume probiotic products, maintain a bland diet, and avoid greasy and spicy foods. KOA patients and healthy subjects are informed to collect stool samples on an empty stomach in the morning (stool collection should not be performed during menstruation for women). The sample is transferred to the sample collection tube using the sampling spoon, sampling only from the middle of the stool sample and avoiding contact between the sampling spoon and other parts of the sample. Approximately 2g is transferred to each sample collection tube, and the cap is immediately tightened. The samples are preserved by wrapping the collection tubes in a plastic bag and placing them in a refrigerator at -80°C for testing. After collection, the stool morphology is recorded using the Bristol Stool Classification Scale. Following the acquisition of stool samples, gut microbiota analysis is performed based on the stool samples, and total genomic DNA of fecal bacteria is extracted using a modified TIANamp Stool DNA Kit method. The diversity and composition of bacterial community structure in each sample were assessed using the method described by Caporaso et al. Paired-end sequencing of 16S rDNA PCR products was performed using an Illumina MiSeq sequencer. Using UPARSE software, OTU clustering was performed on the sequences based on 97% similarity, removing single sequences and chimeras during the clustering process. Species classification annotation was performed on each sequence using an RDP classifier, and the sequence was compared against the Silva database with a comparison threshold of 70%. Based on these analyses, the quantity of various types of gut microbiota was determined, serving as gut microbiota data.

[0027] According to one embodiment of the present invention, the inflammatory factor data (first inflammatory factor data, second inflammatory factor data, and third inflammatory factor data) are collected as follows: First, blood samples are collected. Patients and healthy subjects are instructed to fast in the morning, and 3 mL of venous blood is collected from the elbow vein in a sitting position in a blood collection tube. After standing for 1 hour, the blood is centrifuged at 3000 r / min for 15 min. The supernatant is collected after centrifugation and placed in a refrigerator at -80°C for testing. After obtaining the blood samples, the following inflammatory factors are detected using ELISA: Typical pro-inflammatory cytokines: interleukin-1beta (IL-1β), interleukin-6 (IL-6), and interleukin-12 (IL-12). Anti-inflammatory cytokines: interleukin-4 (IL-4) and interleukin-10 (IL-10). Tumor necrosis factor-α (TNF-α), β-nerve growth factor (β-NGF), neuron-specific enolase (NSE), intercellular adhesion molecule (ICAM), and interferon-γ (INF-γ) are all inflammatory factors. After obtaining the levels of these multiple inflammatory factors, inflammatory factor data can be obtained.

[0028] According to one embodiment of the present invention, in step S3, at the end of each treatment course, the electroencephalogram (EEG) data of the KOA patient can also be acquired. When acquiring the EEG data, the head can be divided into multiple brain regions, namely, the prefrontal lobe, left frontal lobe, right frontal lobe, central frontal lobe, left parietal lobe, right parietal lobe, central parietal lobe, left temporal lobe, right temporal lobe, and occipital lobe. Each brain region may include one or more detection points. For example, the prefrontal lobe includes two detection points, FP1 and FP2; the left frontal lobe includes four detection points, F3, FC3, F7, and FT7; the right frontal lobe includes four detection points, F4, FC4, F8, and FT8; the central frontal lobe includes two detection points, Fz and FCz; the left parietal lobe includes three detection points, C3, CP3, and P3; the right parietal lobe includes three detection points, C4, CP4, and P4; the central parietal lobe includes three detection points, Cz, CPz, and Pz; the left temporal lobe includes three detection points, T3, TP7, and T5; the right temporal lobe includes three detection points, T4, TP8, and T6; and the occipital lobe includes three detection points, Oz, O1, and O2. When collecting EEG data, EEG electrodes can be attached to the detection points for collection. EEG data can be detected at each detection point. After bandpass filtering, the EEG signal can be separated into 5 frequency bands: Delta (1-4Hz), Theta (4-8Hz), Alpha (8-15Hz), Beta (15-30Hz), and Gamma (30-45Hz). Furthermore, the spectral information of the EEG signal can be obtained through Fast Fourier Transform, and the relative power of each frequency band can be determined according to the following formula (1): (1) in, Represents frequency With frequency The relative power between them Represents frequency With frequency The power between, This represents the total power across all frequency bands.

[0029] According to an embodiment of the present invention, the relative power of each frequency band can be obtained by the above formula (1), and the relative power of each frequency band can be combined into a vector to obtain the EEG data vector.

[0030] According to one embodiment of the present invention, the intervention process may include acupuncture treatment using the "short-needle method," which includes: using 1.5-inch filiform needles at the inner and outer knee eyes, inserting them obliquely into the joint cavity for about 1 inch, stopping when the needle reaches the bone; using 1.5-inch filiform needles at Yinlingquan and Zusanli, employing the Dong's Extraordinary Points acupuncture method, inserting the needle close to the edge of the tibia and close to the bone; using 2-inch filiform needles at Liangqiu, inserting them obliquely to the femur, with the needle tip penetrating the periosteum of the femur and piercing Xuehai. After inserting the needles at each point, perform up-and-down lifting, thrusting, and twisting for 1 minute. Retain the needles for 20 minutes each time, once a day, for 6 consecutive days as one course of treatment, with a 1-day rest between courses, for a total of 8 courses of treatment. After each course of treatment, the recovery status of the KOA patient is determined based on the above-mentioned first intestinal flora data, second intestinal flora data, first inflammatory factor data, second inflammatory factor data, third intestinal flora data, third inflammatory factor data, and EEG data, and the effectiveness of the intervention is judged.

[0031] Figure 2 A schematic diagram illustrating the determination of rehabilitation status information according to an embodiment of the present invention is shown as an example.

[0032] According to an embodiment of the present invention, in step S4, the EEG data is processed by an EEG data processing model to obtain EEG feature information, including: acquiring EEG data of multiple detection points in multiple brain regions of a KOA patient; performing frequency domain analysis on the EEG data of each detection point to obtain the frequency domain feature vector of each detection point; determining the number of detection points in the brain region; if the number of detection points in the brain region is greater than 1, determining the weighted phase lag index between each detection point based on the EEG data of each detection point in the brain region; determining the adjacency matrix corresponding to the brain region based on the weighted phase lag index; and processing the EEG data of the brain region according to the subgraph model. The frequency domain feature vectors and adjacency matrices of each detection point are processed to obtain the first output vector of each detection point in the brain region. The first output vectors of each detection point in the brain region are averaged to obtain the brain region feature vector. If the number of detection points in the brain region is equal to 1, the frequency domain feature vector of the detection point is determined as the brain region feature vector. The brain region feature vectors of multiple brain regions are processed through a graph attention network model to obtain the second output vector of each brain region. The second output vector of each brain region is input into the first multilayer perceptron layer (including multiple fully connected layers and softmax activation layers) to obtain the EEG feature information of KOA patients.

[0033] According to one embodiment of the present invention, as described above, after acquiring EEG data, frequency domain analysis can be performed to obtain EEG data vectors, i.e., frequency domain feature vectors. The functions of the same brain region are similar or identical. For example, multiple detection points in a certain brain region will all respond to sensations such as pain caused by KOA, i.e., generate EEG data. However, the EEG data from different detection points may have some differences. When analyzing EEG data, the EEG data vectors collected from multiple detection points within the same brain region can be fused, thereby analyzing the response of multiple detection points within the same brain region to KOA based on their functions, and thus determining the recovery status of KOA.

[0034] According to one embodiment of the present invention, if the number of detection points in a brain region is greater than 1, the relationship between the EEG data vectors of each detection point can be determined. In an example, the weighted phase lag index between each detection point can be determined based on the EEG data of each detection point in the brain region. The weighted phase lag index between two detection points describes the phase difference between the two detection points, and is used to describe the degree of synchronization between the two detection points, that is, the degree of synchronization of the responses of the two detection points to the sensations induced by KOA, thereby reflecting the relationship between the two detection points. Based on the weighted phase lag index between two detection points, the adjacency matrix corresponding to the brain region can be determined, wherein the data in the x-th row and y-th column of the adjacency matrix is ​​the weighted phase lag index between the x-th detection point and the y-th detection point in the brain region. After obtaining the adjacency matrix, the degree matrix can be determined, wherein the degree matrix is ​​a diagonal matrix, and the x-th data on the diagonal of the degree matrix is ​​the sum of all data in the x-th row of the adjacency matrix. Based on the adjacency matrix and degree matrix, the frequency domain feature vectors of each detection point can be aggregated through the subgraph model to obtain the aggregated feature vectors of each detection point. After processing the aggregated feature vectors through the weight matrix of the subgraph model, the first output vector of each detection point can be obtained. The first output vector is the feature vector after fusing the features of EEG data from multiple detection points in the same brain region. That is, the first output vector of each detection point carries the feature information of EEG data from all detection points in the entire brain region. By averaging the first output vectors of each detection point in the brain region, the brain region feature vector can be obtained. During the aggregation process described above, each detection point aggregates more features with high correlation to itself and fewer features with low correlation to itself based on the synchronization degree between detection points described in the adjacency matrix. Through aggregation at each detection point, important features can be selected from the features carried by multiple detection points (i.e., features with high correlation to each detection point are aggregated more extensively at each detection point), making important features more prominent. Therefore, when the first output vectors of each detection point are combined, the features of important detection points are highlighted more, while the features of unimportant detection points are downplayed. This allows the brain region feature vector to more accurately reflect important features and describe the brain region's response to the sensory input induced by KOA. If there is only one detection point in the brain region, the frequency domain feature vector of the detection point is determined as the brain region feature vector.

[0035] According to one embodiment of the present invention, multiple brain regions can also serve as nodes in a graph structure, but the relationship between the various brain regions is not clear. Therefore, a graph attention network model can be used to process the brain region feature vectors of each brain region. During the processing, the concatenation vector of the brain region feature vectors of two brain regions is processed through the network hierarchy of the graph attention network model itself to obtain the correlation coefficient between the two brain regions, that is, the coefficient used to describe the degree of correlation between the two brain regions. Then, based on the correlation coefficient, the brain region feature vectors of each brain region are aggregated to obtain the second output vector of each brain region. That is, the second output vector of the brain region carries the feature information described by the brain region feature vectors of other brain regions, so that the second output vector can more comprehensively describe the entire brain's response to the sensation caused by KOA and can highlight the important features of the response to the sensation caused by KOA in each brain region. Furthermore, the second output vector of each brain region can be processed through the first multilayer perceptual network layer to highlight the important features of each brain region's response to the sensations induced by KOA, and obtain EEG feature information. This information can be used to describe the entire brain's response to the sensations induced by KOA at the end of the treatment course. If the intervention treatment for KOA is effective, the sensations induced by KOA (e.g., pain) can be reduced, and the brain's response to the sensations induced by KOA can be reduced. This response can be expressed through EEG feature information, and the EEG feature information at the end of each treatment course can be compared to determine whether the intervention is effective.

[0036] In this way, the first output vector of the detection point can be fused with the data features of each detection point in the brain region through the subgraph model, and the important data features of the brain region's response to KOA can be highlighted, thereby improving the accuracy of the description of the EEG data's response to KOA. Furthermore, through the graph attention network model, the second output vector of the brain region can be fused with the data features of the entire brain's response to KOA, and the important data features of the brain's response to KOA can be highlighted, thereby accurately expressing the entire brain's response to the sensations caused by KOA and improving the accuracy of feature representation.

[0037] According to an embodiment of the present invention, in step S5, based on the EEG feature information, second gut microbiota data, second inflammatory factor data, third gut microbiota data, and third inflammatory factor data, the physiological state change information of the KOA patient is determined, including: processing the third gut microbiota data of the current treatment course through a second multilayer perceptual network layer to obtain second gut microbiota feature information; processing the third inflammatory factor data of the current treatment course through a third multilayer perceptual network layer to obtain second inflammatory factor feature information; and processing the second gut microbiota feature information, second inflammatory factor feature information, and EEG feature information through a first cross-attention mechanism to obtain second microbiota influence feature information and second inflammatory factor influence feature information. The process involves: concatenating the influence features of the second gut microbiota with those of the second inflammatory factor to obtain the physiological state description information for the current treatment course; inputting the physiological state description information for the current treatment course and the first hidden state information for the previous treatment course into the physiological state change prediction model to obtain the first hidden state information for the current treatment course, wherein, if the current treatment course is the first treatment course, the first hidden state information for the previous treatment course is the first hidden state information obtained based on the second gut microbiota data and the second inflammatory factor data before the start of the treatment course, and the zero vector of the first hidden state information input when obtaining the first hidden state information before the start of the treatment course; inputting the first hidden state information for the current treatment course into the fourth multilayer perceptual network layer to obtain the physiological state change information of the KOA patient.

[0038] According to one embodiment of the present invention, the second gut microbiota data can be composed of a microbiota data vector. For example, by using the quantity of each gut microbiota as a data point in the vector, a microbiota data vector can be obtained. The microbiota data vector can be processed by a second multilayer perceptron layer (including multiple fully connected layers and softmax activation layers) to extract the data features of the microbiota, thereby obtaining second gut microbiota feature information. Similarly, the second inflammatory factor data can be composed of an inflammatory factor data vector. For example, by using the content of each inflammatory factor as a data point in the vector, an inflammatory factor data vector can be obtained. The inflammatory factor data vector can be processed by a third multilayer perceptron layer (including multiple fully connected layers and softmax activation layers) to extract the data features of the inflammatory factors, thereby obtaining second inflammatory factor feature information.

[0039] According to one embodiment of the present invention, second gut microbiota feature information, second inflammatory factor feature information, and EEG feature information can be processed through a first cross-attention mechanism to fuse the second gut microbiota feature information, the second inflammatory factor feature information, and the EEG feature information to obtain second microbiota influence feature information corresponding to the second gut microbiota feature information, and second inflammatory factor influence feature information corresponding to the second inflammatory factor feature information. In the example, the second gut microbiota feature information can be multiplied with the query matrix corresponding to the first cross-attention mechanism to obtain a query vector; each EEG feature information can be multiplied with the key-value matrix corresponding to the first cross-attention mechanism to obtain a key-value vector corresponding to each EEG feature information; the query vector is multiplied with the key-value vector corresponding to each EEG feature information, and then divided by the square root of the dimension of the key-value vector. After processing the result through a softmax activation function, the relationship weight between the second gut microbiota feature information and the EEG feature information of each brain region can be obtained. Furthermore, by multiplying the EEG feature information with the weight matrix corresponding to the first cross-attention mechanism, a weight vector corresponding to the EEG feature information is obtained. Using the aforementioned relational weights, the weight vector is weighted and summed to obtain the second gut microbiota influence feature information. This allows the second gut microbiota influence feature information to be fused with the EEG feature information, so that the fused second gut microbiota influence feature information can not only express the characteristics of the gut microbiota under intervention treatment but also describe the characteristics of the EEG data. Similarly, the second inflammatory factor influence feature information can be obtained.

[0040] According to one embodiment of the present invention, by splicing the second microbial community influence feature information with the second inflammatory factor influence feature information, physiological state description information that can describe multiple features such as EEG features, inflammatory factor features and gut microbiome features of the current treatment course can be obtained. This information can be used to express the physiological state and recovery status of KOA patients after the intervention of the current treatment course.

[0041] According to one embodiment of the present invention, the physiological state change prediction model can be an RNN model, an LSTM model, etc., and the present invention does not limit the type of physiological state change prediction model. When processing the physiological state description information, the physiological state change prediction model can integrate the features of various physiological states from previous treatment courses, thereby outputting first hidden state information to describe whether the recovery status of the KOA patient has changed after the intervention of the current treatment course, and whether it has changed towards recovery. In the example, the input of the physiological state change prediction model is the physiological state description information of the current treatment course and the first hidden state information of the previous treatment course. To obtain the first hidden state information of the previous treatment course, the physiological state description information of the previous treatment course and the first hidden state information of the two previous treatment courses need to be input… Therefore, the first hidden state information of the previous treatment course can be used to represent the changes in the recovery status of each previous treatment course. If the current treatment course is the first course, the first hidden state information of the previous course is obtained based on the second gut microbiota data and the second inflammatory factor data before the start of the course. That is, the second gut microbiota data and the second inflammatory factor data are used to replace the aforementioned third gut microbiota data and the third inflammatory factor data for the same processing to obtain the first hidden state information. In this processing, the first hidden state information of the previous course input into the physiological state change prediction model is a zero vector. Furthermore, the first hidden state information of the current course can be input into the fourth multilayer perceptron layer (including multiple fully connected layers and softmax activation layers) to obtain the physiological state change information of the KOA patient. This information can be compared with the previous course to determine whether the recovery status of the KOA patient has changed and whether it is changing away from the initial state (the state before treatment).

[0042] In this way, various physiological characteristics from previous treatment courses can be combined to determine whether the recovery status of KOA patients has changed and whether it is moving away from the initial state through a physiological state change prediction model, thus providing a data basis for determining the effectiveness of the intervention.

[0043] According to an embodiment of the present invention, in step S6, the recovery status information of the KOA patient is determined based on the physiological state change information of the KOA patient, the first intestinal flora data, and the first inflammatory factor data. This includes: clustering the first intestinal flora data and the second intestinal flora data to determine one or more clusters containing the fewest amounts of second intestinal flora data, averaging the first intestinal flora data in one or more clusters to obtain reference flora data, and processing the reference flora data through a second multilayer perceptron layer to obtain reference flora feature information; clustering the first inflammatory factor data and the second inflammatory factor data to determine one or more clusters containing the fewest amounts of second inflammatory factor data, averaging the first inflammatory factor data in one or more clusters to obtain reference inflammatory factor data, and processing the reference inflammatory factor data through a third multilayer perceptron layer to obtain reference inflammatory factor feature information; and retrieving the first latent state data from the previous treatment course. The hidden state information of the current treatment course is concatenated with the second hidden state information of the previous treatment course to obtain the comprehensive hidden state information of the previous treatment course. The comprehensive hidden state information of the previous treatment course is processed through the fifth layer of the multilayer perceptual network to obtain the input hidden state information. The physiological state change information, reference microbiota feature information, and reference inflammatory factor feature information of the current treatment course are concatenated to obtain the treatment effectiveness feature information of the current treatment course. The input hidden state information and the treatment effectiveness feature information of the current treatment course are processed through the rehabilitation status prediction model to obtain the second hidden state information of the current treatment course. If the current treatment course is the first treatment course, the second hidden state information of the previous treatment course is the second hidden state information obtained based on the second gut microbiota data and the second inflammatory factor data before the start of the treatment course. When obtaining the second hidden state information before the start of the treatment course, the zero vector of the input second hidden state information is used. The second hidden state information of the current treatment course is input into the sixth layer of the multilayer perceptual network for processing to obtain the rehabilitation status information of the KOA patient.

[0044] According to one embodiment of the present invention, to highlight the differences between healthy subjects and KOA patients in terms of gut microbiota and inflammatory factors, the first gut microbiota data and the second gut microbiota data can be clustered to determine one or more clusters containing the fewest amounts of second gut microbiota data. That is, the first gut microbiota data and features in this cluster have significant differences from the features of the second gut microbiota data of KOA patients. In the example, the first gut microbiota data and the second gut microbiota data can be used to construct a data vector, and then the DBSCAN clustering algorithm can be used for clustering to obtain multiple clusters. Then, one or more clusters containing the fewest amounts of second gut microbiota data can be selected (for example, if the number of second gut microbiota data contained in each cluster is different, the cluster containing the fewest amounts of second gut microbiota data can be selected; if there are multiple clusters containing the fewest amounts of second gut microbiota data, for example, there are multiple clusters that do not contain second gut microbiota data, multiple clusters can be selected). Furthermore, the first gut microbiota data in the multiple clusters can be averaged to obtain reference microbiota data, which is used to describe the microbiota data of healthy subjects. Similarly, reference inflammatory factor data can be obtained. Further, to obtain the characteristics of the microbial community data and inflammatory factor data, a second multilayer perceptron layer can be used to process the reference microbial community data to obtain reference microbial community feature information, and a third multilayer perceptron layer can be used to process the reference inflammatory factor data to obtain reference inflammatory factor feature information.

[0045] According to one embodiment of the present invention, the rehabilitation status prediction model is an RNN or LSTM model, which can process the input latent state information of the current treatment course and the treatment effectiveness feature information of the current treatment course to obtain the second latent state information of the current treatment course. The second latent state information can be used to describe the rehabilitation status of the KOA patient after the current treatment course, i.e., whether the treatment is effective and whether the physiological state is changing towards a healthy state (i.e., the state of a healthy subject). The first latent state information can indicate whether the physiological state has changed relative to the previous treatment course and whether it is changing away from the state before treatment. Therefore, by concatenating the first and second latent state information of the previous treatment course, the resulting comprehensive latent state information can describe the direction of physiological state change of the KOA patient after the previous treatment course and whether there is a trend towards rehabilitation. Compared with using only the second latent state information of the previous treatment course as the input of the rehabilitation status prediction model, the above method increases... Adding the first hidden state information as a constraint on the direction of change in physiological state features makes convergence easier during training. Furthermore, it allows the prediction effect of the rehabilitation status prediction model to more closely approximate the labeled true values. For example, based on training data, using a concatenation of the first and second hidden state information results in a nearly 15% faster convergence speed and a nearly 9% improvement in prediction accuracy compared to using only the second hidden state information. Further, a fifth multilayer perceptron layer (including multiple fully connected layers and ReLU activation layers) can be used to process the comprehensive hidden state information from the previous treatment course, reducing its dimensionality so that the dimension of the input hidden state information equals that of the second hidden state information. On the other hand, reference microbial community features and reference inflammatory factor features can be concatenated with the comprehensive hidden state information from the previous treatment course, thereby increasing the reference microbial community features and reference inflammatory factor features. This provides data for determining whether the physiological state features of KOA patients are changing towards a healthy state, and this reference can also serve as a training constraint, improving the convergence speed and training effect of the rehabilitation status prediction model. After processing the input latent state information and the treatment effectiveness features of the current treatment course by the rehabilitation status prediction model, the second latent state information of the current treatment course can be obtained. This information can be used to describe whether the physiological state of the KOA patient is changing towards a healthy state, and also to describe whether the intervention of the current treatment course is effective. Furthermore, the second latent state information of the current treatment course can be input into the sixth multilayer perceptron layer (e.g., including multiple fully connected layers and ReLU activation layers) for processing to obtain the rehabilitation status information of the KOA patient. This information can represent the rehabilitation status of the KOA patient, for example, describing the degree to which the KOA patient approaches a healthy state (i.e., the state of a healthy subject) after the current treatment course as a percentage.

[0046] This approach enhances the saliency of the physiological status characteristics of healthy subjects by selecting the clusters with the fewest amounts of data on the second gut microbiota and second inflammatory factors after clustering. This provides a data foundation for subsequent verification of whether the physiological status of KOA patients is recovering and whether the intervention is effective. Furthermore, by concatenating the first and second latent state information from the previous treatment course, it can describe whether the physiological status of KOA patients has changed after the previous treatment course and whether it is moving towards a healthy state. This can serve as a reference and constraint for the direction of physiological status changes during training, improving training efficiency and effectiveness. Moreover, reference gut microbiota and reference inflammatory factor characteristics can be used as reference information for the healthy status, providing a basis for judging whether the physiological status of KOA patients is moving towards a healthy state and providing constraints for the training process, further contributing to improved training efficiency and effectiveness.

[0047] According to an embodiment of the present invention, the above model can be trained before use. The method further includes: acquiring sample EEG signals, sample gut microbiota data, and sample inflammatory factor data of sample KOA patients at multiple rehabilitation stages, as well as labeled rehabilitation status at multiple rehabilitation stages; processing the sample EEG signals of the i-th rehabilitation stage through a subgraph model, a graph attention network model, and a first multilayer perceptron layer to obtain sample EEG feature information of the i-th rehabilitation stage; obtaining sample physiological state change information of the i-th rehabilitation stage based on the second, third, and fourth multilayer perceptron layers, the first cross-attention mechanism, a physiological state change prediction model, sample EEG feature information of the i-th rehabilitation stage, sample gut microbiota data, and sample inflammatory factor data; and determining the state change damage based on the sample physiological state change information of multiple rehabilitation stages. Loss function; based on the rehabilitation status prediction model, the fifth and sixth multilayer perceptron layers, reference microbial data, reference inflammatory factor data, and sample physiological state change information, the predicted rehabilitation status information for the i-th rehabilitation stage is obtained; based on the predicted rehabilitation status information and labeled rehabilitation status for each rehabilitation stage, the rehabilitation status loss function is determined; based on the rehabilitation status loss function and the state change loss function, the training loss function is obtained; based on the training loss function, the subgraph model, graph attention network model, the first, second, third, and fourth multilayer perceptron layers, the first cross-attention mechanism, the physiological state change prediction model, the rehabilitation status prediction model, the fifth and sixth multilayer perceptron layers are trained.

[0048] According to one embodiment of the present invention, the sample KOA patients are KOA patients historically selected through the above screening methods, and the data of these KOA patients are used to train the above multiple models. The sample KOA patients can use various other methods for rehabilitation treatment, and various data are obtained at each rehabilitation stage, and the rehabilitation status is labeled by experts. That is, the labeled rehabilitation status at each rehabilitation stage can be used to represent the true rehabilitation status of the KOA patients.

[0049] According to one embodiment of the present invention, the method of obtaining the sample EEG feature information of the i-th rehabilitation stage is similar to the method of obtaining EEG feature information described above, the method of obtaining the sample physiological state change information of the i-th rehabilitation stage is similar to the method of obtaining physiological state change information described above, and the method of obtaining the predicted rehabilitation status information of the i-th rehabilitation stage is similar to the method of obtaining rehabilitation status information described above, and will not be repeated here.

[0050] According to an embodiment of the present invention, during the training process, not only can the predicted rehabilitation status information be made closer to the labeled information, but the physiological state change prediction model can also be further trained to make the prediction of physiological state changes more accurate. Based on the physiological state change information of samples from multiple rehabilitation stages, the state change loss function is determined, including: inputting the sample physiological state change information into the seventh multilayer perceptron layer to obtain the sample rehabilitation status information; and determining the state change loss function according to formula (2). , (2) in, This represents the labeled rehabilitation status for the i-th rehabilitation stage. This represents the labeled rehabilitation status for the (i-1)th rehabilitation stage. This provides information on the recovery status of a sample in the i-th recovery stage. This represents the rehabilitation status information for the sample in the (i-1)th rehabilitation stage, where n is the number of rehabilitation stages. and Preset weights.

[0051] According to one embodiment of the present invention, the seventh multilayer perceptual network layer may include multiple fully connected layers and ReLU activation layers, which can process the physiological state change information of the sample, thereby judging the rehabilitation status of the sample KOA patient based solely on the physiological state change information, and reducing the gap between the sample rehabilitation status information determined based on this and the labeled rehabilitation status during the training process, thereby improving the accuracy of the physiological state change prediction model and helping the physiological state change prediction model to obtain more accurate physiological state change information. In formula (2), This means narrowing the gap between the sample recovery status information and the labeled recovery status during training, so that the physiological state change prediction model can obtain more accurate physiological state change information.

[0052] According to one embodiment of the present invention, Let be the change in the recovery status information of the sample in the i-th recovery stage relative to the recovery status information of the sample in the (i-1)-th recovery stage. This represents the change in the labeled rehabilitation status at the i-th rehabilitation stage relative to the labeled rehabilitation status at the (i-1)-th rehabilitation stage. This represents the gap between the two changes mentioned above. During training, this gap can be narrowed so that the change pattern of the sample's rehabilitation status information is the same as the change pattern of the labeled rehabilitation status. This can further constrain the prediction pattern of the physiological state change prediction model on the sample's physiological state change information, thereby improving training accuracy and efficiency.

[0053] According to one embodiment of the present invention, the state change loss function can be obtained by weighted summing of the above two items corresponding to multiple rehabilitation stages. During the training process, the state change loss function can be reduced, thereby improving the prediction accuracy and training efficiency of the physiological state change prediction model.

[0054] In this way, the seventh layer of the perceptual network can be used for assisted training. When determining the loss function for state changes, the loss function is constructed by using two aspects: the error between the labeled rehabilitation status and the sample rehabilitation status information, and the error in the change pattern of the labeled rehabilitation status and the sample rehabilitation status information. This can further constrain the prediction pattern of the physiological state change prediction model for the sample physiological state change information, thereby improving training accuracy and training efficiency.

[0055] According to one embodiment of the present invention, a rehabilitation status loss function is determined based on the predicted rehabilitation status information and labeled rehabilitation status at each rehabilitation stage, including: determining the rehabilitation status loss function according to formula (3). , (3) in, This provides information on the predicted rehabilitation status for the i-th rehabilitation stage. This provides information on the predicted rehabilitation status for the (i-1)th rehabilitation stage. and The preset weights are defined by 'if', which is a conditional function.

[0056] According to one embodiment of the present invention, in formula (3), To reduce the gap between the predicted rehabilitation status information and the labeled rehabilitation status, the training process can be used to improve the accuracy of the various models mentioned above.

[0057] According to one embodiment of the present invention, in formula (3), For a conditional function, in In this case, the value of the conditional function is Otherwise, the condition function value is Furthermore, in or When, directly set the condition function value to That is, discarding meaningless things. .

[0058] According to one embodiment of the present invention, as described above, This value indicates the consistency between the changing patterns of the sample's recovery status information and the changing patterns of the labeled recovery status. The smaller the value, the higher the consistency between the changing patterns of the sample's recovery status information and the labeled recovery status. Similarly, This indicates the consistency between the predicted pattern of changes in rehabilitation status information and the labeled pattern of changes in rehabilitation status. The smaller the value, the higher the consistency between the predicted pattern of changes in rehabilitation status information and the labeled pattern of changes in rehabilitation status. This indicates a higher consistency between the predicted and labeled patterns of change in rehabilitation status. In this case, it means that after using the rehabilitation status prediction model and more constraints, the model's output for predicting rehabilitation status information is more accurate, and the predicted trend of change in rehabilitation status information is also more accurate. Therefore, it can be considered... As a conditional function value, this term is further reduced during training to improve the consistency between the predicted and labeled patterns of change in rehabilitation status information. Conversely, if the consistency between the sample and labeled patterns of change in rehabilitation status information is higher, it indicates that after using the rehabilitation status prediction model and more constraints, the model's accuracy in predicting rehabilitation status information is worse, and its judgment of the predicted patterns of change in rehabilitation status information is also worse. In this case, using... As the value of the loss function, where the denominator for and The relative error, This can accurately represent the changing patterns of the sample's recovery status information. In addition to this, during the training process... Besides shrinking, it can also make As the fraction shrinks, the denominator increases continuously, thus... It keeps shrinking, and in During the shrinking process, the accuracy of the physiological state change prediction model can be continuously improved, making it more likely that the change patterns of the sample's recovery status information are more consistent with the change patterns of the labeled recovery status, thus allowing the condition function to take on a value that is more favorable to the target population. The likelihood is greater, thus increasing the training power of the rehabilitation status prediction model by having a denominator less than 1. This enhances the training intensity of the rehabilitation status prediction model, thereby improving its accuracy. This leads to a comparison and competition in accuracy between the physiological state change prediction model and the rehabilitation status prediction model, which is beneficial for further improving the accuracy and training efficiency of both.

[0059] According to one embodiment of the present invention, the weighted sum of the above two items corresponding to multiple rehabilitation stages can be used to obtain the rehabilitation status loss function. During the training process, the rehabilitation status loss function can be reduced, thereby improving the prediction accuracy and training efficiency of each model.

[0060] In this way, by setting conditional functions, the physiological state change prediction model and the rehabilitation status prediction model can be compared and competed in terms of accuracy, thereby improving the training intensity of the models, enhancing the accuracy of predicting rehabilitation status information and its changing trends, and improving the prediction accuracy and training efficiency of each model.

[0061] According to one embodiment of the present invention, after obtaining the above-mentioned rehabilitation status loss function and state change loss function, the two can be weighted and summed to obtain the training loss function. Then, the parameters of the subgraph model, graph attention network model, first multilayer perceptron layer, second multilayer perceptron layer, third multilayer perceptron layer, fourth multilayer perceptron layer, first cross-attention mechanism, physiological state change prediction model, rehabilitation status prediction model, fifth multilayer perceptron layer, and sixth multilayer perceptron layer are adjusted through backpropagation to train these models. After multiple training sessions, the training can be completed, and the trained model can be used in the process of obtaining rehabilitation status information.

[0062] According to one embodiment of the present invention, in step S7, an intervention effectiveness evaluation index is determined based on the rehabilitation status information of KOA patients, including: obtaining the average value of rehabilitation status information of multiple KOA patients, and the historical average value of rehabilitation status information of multiple KOA patients at the end of the previous treatment course; determining the relative difference between the average value of the rehabilitation status information and the historical average value; and determining the relative difference as the intervention effectiveness evaluation index. This relative difference can be used to describe the improvement in the average value of the rehabilitation status information after the intervention of the current treatment course, and can describe the intervention effectiveness of the current treatment course, thus serving as an intervention effectiveness evaluation index.

[0063] According to an embodiment of the present invention, in step S8, the rehabilitation status monitoring results can be determined based on the intervention effectiveness evaluation indicators. The rehabilitation status monitoring results may include the intervention effectiveness evaluation indicators, which may represent the overall intervention effect of multiple KOA patients, and may also include the rehabilitation status information of each KOA patient, as well as the rehabilitation status information of each previous treatment course, for the purpose of analyzing the rehabilitation status of each KOA patient.

[0064] The KOA rehabilitation monitoring method based on gut microbiota and multi-source data synergy, according to embodiments of the present invention, can screen healthy subjects as references and, by monitoring various data such as gut microbiota, inflammatory factors, and EEG data of KOA patients, determine whether the physiological state of KOA patients has changed and whether they are recovering, thereby accurately analyzing the rehabilitation status of KOA patients. Furthermore, based on the rehabilitation status of multiple KOA patients, it can determine whether intervention treatment for KOA patients is effective, providing an accurate data foundation for the analysis of intervention effects. A subgraph model can be used to fuse the data features of each detection point within the brain region into the first output vector of the detection point, highlighting important data features of the brain region's response to KOA, improving the accuracy of the description of the EEG data's response to KOA. A graph attention network model can be used to fuse the data features of the entire brain's response to KOA into the second output vector of the brain region, highlighting important data features of the brain's response to KOA, accurately expressing the entire brain's sensory response to KOA, and improving the accuracy of feature expression. Furthermore, by combining various physiological characteristics from previous treatment courses, a physiological state change prediction model can be used to determine whether the recovery status of KOA patients has changed, and whether it is changing away from the initial state, providing a data foundation for determining the effectiveness of the intervention. Moreover, by clustering and selecting the clusters with the fewest data points containing second gut microbiota and second inflammatory factors, the characteristic significance of the physiological state of healthy subjects can be improved, providing a data foundation for subsequent verification of whether the physiological state of KOA patients is recovering and whether the intervention is effective. Furthermore, by concatenating the first and second latent state information from the previous treatment course, it is possible to describe whether the physiological state of KOA patients has changed after the previous treatment course and whether it is changing towards a healthy state. This can also serve as a reference and constraint for the direction of physiological state changes during training, improving training efficiency and effectiveness. Additionally, reference microbiota characteristic information and reference inflammatory factor characteristic information can be used as reference information for the healthy state, providing a basis for judging whether the physiological state of KOA patients is changing towards a healthy state, and providing constraints for the training process, which helps to further improve training efficiency and effectiveness. During training, a seventh-level multilayer perceptron can be used for assisted training. When determining the loss function for state changes, two aspects are considered: the error between the labeled rehabilitation status and the sample rehabilitation status information, and the error in the variation pattern of the labeled rehabilitation status and the sample rehabilitation status information. This constructs the loss function by further constraining the prediction pattern of the physiological state change prediction model for sample physiological state changes, improving training accuracy and efficiency. Furthermore, by setting conditional functions, the physiological state change prediction model and the rehabilitation status prediction model can be compared and competed in terms of accuracy, enhancing model training intensity and improving the accuracy of predicting rehabilitation status information and its changing trends, thereby improving the prediction accuracy and training efficiency of each model.

[0065] Figure 3 An exemplary block diagram of a KOA rehabilitation status monitoring system based on gut microbiota and multi-source data synergy according to an embodiment of the present invention is shown, the system comprising: The screening module is used to screen KOA patients and healthy subjects according to preset screening rules; The acquisition module is used to acquire the first gut microbiota data of healthy subjects and the second gut microbiota data of KOA patients, as well as the first inflammatory factor data of healthy subjects and the second inflammatory factor data of KOA patients; The detection module is used to acquire third gut microbiota data, third inflammatory factor data and EEG data of KOA patients at the end of each treatment course during the intervention process; The EEG feature module is used to process EEG data through an EEG data processing model to obtain EEG feature information. The physiological state change module is used to determine the physiological state change information of KOA patients based on the electroencephalogram (EEG) feature information, second gut microbiota data, second inflammatory factor data, third gut microbiota data, and third inflammatory factor data. The recovery status module is used to determine the recovery status information of KOA patients based on the physiological status changes, first gut microbiota data, and first inflammatory factor data. The assessment module is used to determine the indicators for evaluating the effectiveness of the intervention based on the rehabilitation status information of KOA patients; The results module is used to determine the monitoring results of rehabilitation status based on the intervention effectiveness assessment indicators.

[0066] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0067] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy, characterized in that, include: KOA patients and healthy subjects were screened according to preset screening rules; We acquired primary gut microbiota data from healthy subjects and secondary gut microbiota data from KOA patients, as well as primary inflammatory factor data from healthy subjects and secondary inflammatory factor data from KOA patients. At the end of each treatment course during the intervention of KOA patients, the third gut microbiota data, third inflammatory factor data and EEG data of KOA patients were obtained respectively. By using an EEG data processing model, EEG data is processed to obtain EEG feature information; Based on the electroencephalogram (EEG) characteristics, second gut microbiota data, second inflammatory factor data, third gut microbiota data, and third inflammatory factor data, the physiological state changes of KOA patients are determined. Based on the physiological status changes of KOA patients, along with the first gut microbiota data and the first inflammatory factor data, the recovery status information of KOA patients was determined. Based on the rehabilitation status information of KOA patients, indicators for evaluating the effectiveness of interventions were determined; The results of rehabilitation status monitoring were determined based on the intervention effectiveness assessment indicators.

2. The method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy according to claim 1, characterized in that, EEG data is processed using an EEG data processing model to obtain EEG feature information, including: Acquire EEG data from multiple brain regions and multiple detection points in KOA patients; Frequency domain analysis was performed on the EEG data at each detection point to obtain the frequency domain feature vector of each detection point; Determine the number of detection points within the brain region; If the number of detection points in the brain region is greater than 1, the weighted phase lag index between each detection point is determined based on the EEG data of each detection point in the brain region. The adjacency matrix corresponding to the brain region is determined based on the weighted phase lag index. Based on the subgraph model, the frequency domain feature vectors and adjacency matrices of each detection point in the brain region are processed to obtain the first output vector of each detection point in the brain region. The brain region feature vector is obtained by averaging the first output vectors of each detection point in the brain region. If the number of detection points in a brain region is equal to 1, then the frequency domain feature vector of the detection point is determined as the brain region feature vector; By using a graph attention network model, the feature vectors of multiple brain regions are processed to obtain the second output vector of each brain region. The second output vector of each brain region is input into the first multilayer perceptual network layer to obtain the EEG characteristic information of KOA patients.

3. The method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy according to claim 1, characterized in that, Based on the aforementioned EEG characteristics, second gut microbiota data, second inflammatory factor data, third gut microbiota data, and third inflammatory factor data, the physiological state changes of KOA patients are determined, including: The third gut microbiota data of the current treatment course is processed through the second multilayer perceptual network layer to obtain the second gut microbiota feature information. The third inflammatory factor data of the current treatment course is processed through the third multi-layer perceptual network layer to obtain the characteristic information of the second inflammatory factor. By using the first cross-attention mechanism, the second gut microbiota feature information, the second inflammatory factor feature information, and the EEG feature information are processed to obtain the second microbiota influence feature information and the second inflammatory factor influence feature information. By splicing the influence characteristics of the second microbial community with the influence characteristics of the second inflammatory factor, the physiological state description information of the current treatment course can be obtained. The physiological state description information of the current treatment course and the first hidden state information of the previous treatment course are input into the physiological state change prediction model to obtain the first hidden state information of the current treatment course. If the current treatment course is the first treatment course, the first hidden state information of the previous treatment course is the first hidden state information obtained based on the second gut microbiota data and the second inflammatory factor data before the start of the treatment course. When obtaining the first hidden state information before the start of the treatment course, the zero vector of the first hidden state information is input. The first hidden state information of the current treatment course is input into the fourth multilayer perceptual network layer to obtain the physiological state change information of the KOA patient.

4. The method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy according to claim 3, characterized in that, Based on the physiological status changes, primary gut microbiota data, and primary inflammatory factor data of KOA patients, the recovery status information of KOA patients was determined, including: The first and second gut microbiota data are clustered to determine one or more clusters containing the fewest second gut microbiota data. The first gut microbiota data in one or more clusters are averaged to obtain reference microbiota data. The reference microbiota data is then processed through a second multilayer perceptron layer to obtain reference microbiota feature information. The first inflammatory factor data and the second inflammatory factor data are clustered to determine one or more clusters containing the fewest second inflammatory factor data. The first inflammatory factor data in one or more clusters are averaged to obtain reference inflammatory factor data. The reference inflammatory factor data is then processed through a third multilayer perceptron layer to obtain reference inflammatory factor feature information. By concatenating the first hidden state information and the second hidden state information of the previous treatment course, the comprehensive hidden state information of the previous treatment course is obtained. The comprehensive hidden state information from the previous treatment course is processed through the fifth multilayer perceptual network layer to obtain the input hidden state information; By splicing together the physiological state change information, reference microbial community characteristic information, and reference inflammatory factor characteristic information of the current treatment course, the treatment effectiveness characteristic information of the current treatment course can be obtained. The rehabilitation status prediction model processes the input latent state information and the treatment effectiveness feature information of the current course to obtain the second latent state information of the current course. If the current course is the first course, the second latent state information of the previous course is the second latent state information obtained based on the second gut microbiota data and the second inflammatory factor data before the start of the course. When obtaining the second latent state information before the start of the course, the zero vector of the input second latent state information is used. The second hidden state information of the current treatment course is input into the sixth multilayer perceptual network layer for processing to obtain the rehabilitation status information of the KOA patient.

5. The method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy according to claim 4, characterized in that, The method further includes: We acquired sample EEG signals, sample gut microbiota data, and sample inflammatory factor data from KOA patients at multiple rehabilitation stages, as well as labeled rehabilitation status at multiple rehabilitation stages. The EEG signals of the samples in the i-th rehabilitation stage are processed by the subgraph model, graph attention network model and the first multilayer perceptual network layer to obtain the EEG feature information of the samples in the i-th rehabilitation stage. Based on the second, third, and fourth multilayer perceptual network layers, the first cross-attention mechanism, the physiological state change prediction model, the EEG characteristics of the sample in the i-th rehabilitation stage, the sample gut microbiota data, and the sample inflammatory factor data, the physiological state change information of the sample in the i-th rehabilitation stage is obtained. Based on the physiological state change information of samples from multiple rehabilitation stages, determine the state change loss function; Based on the rehabilitation status prediction model, the fifth and sixth multilayer sensory network levels, reference microbial data, reference inflammatory factor data, and sample physiological state change information, the predicted rehabilitation status information for the i-th rehabilitation stage is obtained. Based on the predicted rehabilitation status information and labeled rehabilitation status at each rehabilitation stage, the rehabilitation status loss function is determined; The training loss function is obtained based on the rehabilitation status loss function and the state change loss function; The training loss function is used to train the subgraph model, graph attention network model, first multilayer perceptron layer, second multilayer perceptron layer, third multilayer perceptron layer, fourth multilayer perceptron layer, first cross-attention mechanism, physiological state change prediction model, rehabilitation status prediction model, fifth multilayer perceptron layer, and sixth multilayer perceptron layer.

6. The method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy according to claim 5, characterized in that, Based on physiological state change information from multiple rehabilitation stages, a state change loss function is determined, including: Input the physiological state change information of the sample into the seventh multilayer sensory network layer to obtain the sample's recovery status information; According to the formula Determine the state change loss function ,in, This represents the labeled rehabilitation status for the i-th rehabilitation stage. This represents the labeled rehabilitation status for the (i-1)th rehabilitation stage. This provides information on the recovery status of the sample in the i-th recovery stage. This represents the rehabilitation status information for the sample in the (i-1)th rehabilitation stage, where n is the number of rehabilitation stages. and Preset weights.

7. The method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy according to claim 6, characterized in that, Based on the predicted rehabilitation status information and labeled rehabilitation status at each rehabilitation stage, a rehabilitation status loss function is determined, including: According to the formula Determine the loss function for rehabilitation status ,in, This provides information on the predicted rehabilitation status for the i-th rehabilitation stage. This provides information on the predicted rehabilitation status for the (i-1)th rehabilitation stage. and The preset weights are defined by `if`, which is a conditional function.

8. The method for monitoring KOA recovery status based on gut microbiota and multi-source data synergy according to claim 1, characterized in that, Based on the rehabilitation status information of KOA patients, indicators for evaluating the effectiveness of the intervention were determined, including: The average value of the rehabilitation status information of multiple KOA patients was obtained, as well as the historical average value of the rehabilitation status information of multiple KOA patients at the end of the previous treatment course. Determine the relative difference between the average value of the rehabilitation status information and the historical average value; The relative gap was determined as an indicator for evaluating the effectiveness of the intervention.

9. A KOA recovery status monitoring system based on gut microbiota and multi-source data synergy, the system being used to perform the method as described in any one of claims 1-8, characterized in that, include: The screening module is used to screen KOA patients and healthy subjects according to preset screening rules; The acquisition module is used to acquire the first gut microbiota data of healthy subjects and the second gut microbiota data of KOA patients, as well as the first inflammatory factor data of healthy subjects and the second inflammatory factor data of KOA patients; The detection module is used to acquire third gut microbiota data, third inflammatory factor data and EEG data of KOA patients at the end of each treatment course during the intervention process; The EEG feature module is used to process EEG data through an EEG data processing model to obtain EEG feature information. The physiological state change module is used to determine the physiological state change information of KOA patients based on the electroencephalogram (EEG) feature information, second gut microbiota data, second inflammatory factor data, third gut microbiota data, and third inflammatory factor data. The recovery status module is used to determine the recovery status information of KOA patients based on the physiological status changes, first gut microbiota data, and first inflammatory factor data. The assessment module is used to determine the indicators for evaluating the effectiveness of the intervention based on the rehabilitation status information of KOA patients; The results module is used to determine the monitoring results of rehabilitation status based on the intervention effectiveness assessment indicators.

Images