Pediatric neural data intelligent management system based on artificial intelligence

Abstract

The invention relates to the technical field of data analysis, in particular to a pediatric neural data intelligent management system based on artificial intelligence. The system comprises a pediatric neural data acquisition module, a pediatric multi-modal fusion module, a personalized management recommendation module and a recommendation scheme abnormity feedback optimization module, and pediatric neural physiological data can be acquired in real time and uploaded to a cloud platform; performing abnormal noisy point correction and cross-data fusion on the pediatric nerve physiological data based on a convolutional neural network to generate pediatric nerve multi-modal fusion data; performing personalized recommendation based on the pediatric nerve multi-modal fusion data to generate a pediatric patient personalized management recommendation scheme; and applying the pediatric patient personalized management recommendation scheme and collecting corresponding pediatric neural data, and performing pediatric health abnormality feedback analysis and abnormality feedback optimization adjustment at the same time to generate a pediatric personalized optimization adjustment management scheme. The pediatric nerve personalized medical management recommendation system can realize pediatric nerve personalized medical management recommendation work.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to an intelligent management system for pediatric neurological data based on artificial intelligence. Background Art

[0002] With the continuous development of medical technology, especially the rapid progress of artificial intelligence (AI) technology, intelligent healthcare has become a key means to solve the management of complex diseases. In the diagnosis and treatment of pediatric neurological diseases, doctors face challenges such as large individual differences among patients, complex conditions, and variable treatment responses. Therefore, how to provide personalized treatment plans through accurate data collection and analysis, combined with artificial intelligence technology, has become an important direction to improve the treatment effect of pediatric neurological diseases.

[0003] Currently, the intelligent management of pediatric neurological diseases usually relies on traditional medical means, such as electroencephalogram (EEG), heart rate monitoring, blood sample analysis, etc. However, the data generated by these analysis tools are often isolated and heterogeneous, making it difficult to comprehensively consider the correlation between various indicators. At the same time, traditional data analysis methods often rely on manual intervention, with the defects of incomplete and untimely data processing, thus reducing the efficiency of data anomaly detection. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide an intelligent management system for pediatric neurological data based on artificial intelligence to solve at least one of the above technical problems.

[0005] To achieve the above object, an intelligent management system for pediatric neurological data based on artificial intelligence includes the following modules:

[0006] A pediatric neurological data acquisition module, configured to collect pediatric neurological physiological data corresponding to pediatric patients in real time through corresponding intelligent hardware devices, and transmit the data to the corresponding intelligent medical devices of pediatric patients in real time through wireless technology; upload the collected pediatric neurological physiological data to the cloud platform through the intelligent medical devices;

[0007] A pediatric multi-modal fusion module, configured to use the cloud platform and based on a convolutional neural network to correct abnormal noise points in the pediatric neurological physiological data to obtain corrected pediatric neurological abnormal data; perform cross-data fusion on each sub-data in the corrected pediatric neurological abnormal data based on a multi-modal neural network to generate pediatric neurological multi-modal fusion data;

[0008] A personalized management recommendation module, configured to perform personalized recommendation based on the pediatric neurological multi-modal fusion data and combined with reinforcement learning to generate a corresponding personalized management recommendation plan for pediatric patients;

[0009] The recommended solution abnormal feedback optimization module is used to apply the personalized management recommended solution for pediatric patients to the corresponding pediatric patients, monitor and collect the corresponding pediatric nerve data after personalized management using various intelligent emergency devices, and use the edge computing node to perform pediatric health abnormal feedback analysis on the corresponding pediatric nerve data to obtain the health abnormal signal feedback status of the corresponding pediatric patients; based on the health abnormal signal feedback status of the corresponding pediatric patients, the cloud platform is used to perform abnormal feedback optimization adjustment on the corresponding personalized management recommended solution for pediatric patients to generate the corresponding pediatric personalized optimization adjustment management plan.

[0010] Furthermore, the pediatric nerve data acquisition module includes the following functions:

[0011] Real-time collect the electroencephalogram signal data corresponding to pediatric patients through an electroencephalogram monitor;

[0012] Real-time collect the heart rate signal data corresponding to pediatric patients through a heart rate monitor;

[0013] Real-time collect the respiratory blood oxygen saturation data corresponding to pediatric patients through a blood oxygen sensor;

[0014] Real-time collect the nerve movement state data corresponding to pediatric patients through a motion sensor;

[0015] Synthesize the electroencephalogram signal data, heart rate signal data, respiratory blood oxygen saturation data, and nerve movement state data corresponding to pediatric patients into a dataset to obtain the corresponding pediatric neurophysiological data, and transmit it to the intelligent medical device worn by the corresponding pediatric patient in real time through wireless technologies such as Bluetooth or Wi-Fi;

[0016] Upload the collected pediatric neurophysiological data to the cloud platform through the intelligent medical device.

[0017] Furthermore, the pediatric multimodal fusion module includes the following functions:

[0018] Use the short-time Fourier transform on the cloud platform to perform data feature input conversion on the electroencephalogram, heart rate, blood oxygen saturation, and movement state data in the pediatric neurophysiological data, so as to convert the corresponding data into a feature format suitable for deep learning input to obtain the pediatric nerve time series data features;

[0019] Obtain the labeled pediatric neurooriginal data, including the pediatric nerve data labeled with normal and abnormal noise points;

[0020] Train the corresponding convolutional neural network model using the labeled pediatric neuron data so that it can identify the corresponding normal and abnormal noise points in the pediatric nerve data, and input the pediatric nerve time series data features into the trained convolutional neural network model for abnormal noise point detection and division to obtain pediatric nerve normal data and pediatric nerve abnormal noise data;

[0021] Based on the pediatric nerve normal data and the pediatric nerve abnormal noise data, correct the abnormal noise points at the corresponding positions in the pediatric nerve physiological data to obtain pediatric nerve abnormal correction data;

[0022] Based on the multi-modal neural network, perform cross-data fusion on each sub-data in the pediatric nerve abnormal correction data to generate pediatric nerve multi-modal fusion data.

[0023] Further, the correcting the abnormal noise points at the corresponding positions in the pediatric nerve physiological data based on the pediatric nerve normal data and the pediatric nerve abnormal noise data includes:

[0024] Perform statistical analysis on the data distribution of the pediatric nerve abnormal noise data to obtain the mean value of the pediatric nerve noise data distribution and the standard deviation of the pediatric nerve noise data distribution;

[0025] Based on the mean value of the pediatric nerve noise data distribution and the standard deviation of the pediatric nerve noise data distribution, and combined with linear fitting interpolation, perform noise fitting correction on the corresponding normal data distribution in the pediatric nerve normal data, and interpolate and fit the corresponding normal data distribution according to the mean value of the pediatric nerve noise data distribution and the standard deviation of the pediatric nerve noise data distribution to generate the abnormal correction value corresponding to the abnormal noise point, and obtain the pediatric nerve noise abnormal correction value;

[0026] Based on the pediatric nerve noise abnormal correction value, correct the abnormal noise points at the corresponding positions in the pediatric nerve physiological data to obtain pediatric nerve abnormal correction data.

[0027] Further, the performing cross-data fusion on each sub-data in the pediatric nerve abnormal correction data based on the multi-modal neural network includes:

[0028] Perform data feature analysis on each sub-data in the pediatric data abnormal correction data to realize time-frequency analysis of the electroencephalogram signal to obtain the corresponding time-frequency characteristics of the electroencephalogram signal, perform periodic analysis of the heart rate signal to obtain the corresponding periodic characteristics of the heart rate signal, and perform statistical analysis on the data distribution of blood oxygen saturation and nerve movement state to obtain the corresponding data distribution characteristics of blood oxygen and movement state, and obtain the sub-modal characteristics corresponding to each pediatric sub-data type;

[0029] Each network branch of the multimodal neural network is used to process the sub-modal features corresponding to the pediatric sub-data types, and the attention mechanism is used to weight the importance weights of the sub-modal features. At the same time, based on the importance weights, the sub-modal features corresponding to each pediatric sub-data type are cross-data fused to generate pediatric neural multimodal fusion data.

[0030] Further, the personalized management recommendation module includes the following functions:

[0031] Obtain the feature anomaly correlation between each pediatric data sub-modal of the pediatric patient through the pediatric neural multimodal fusion data;

[0032] Based on the feature anomaly correlation between each pediatric data sub-modal of the pediatric patient and combined with reinforcement learning and the corresponding pediatric data sub-modal modeling, generate the corresponding treatment management recommendation reinforcement learning model, take the corresponding pediatric data sub-modal as the state vector, and recommend the corresponding pediatric treatment management plan according to the original treatment management database. At the same time, design a reward function according to the corresponding effect action to continuously reinforce the learning of the corresponding treatment management plan to generate the corresponding personalized management recommendation plan for the pediatric patient.

[0033] Further, the personalized management recommendation plan for the pediatric patient includes the types and doses of drugs corresponding to the pediatric neural data of the pediatric patient in the clinic.

[0034] Further, the recommendation plan anomaly feedback optimization module includes the following functions:

[0035] Support and apply the personalized management recommendation plan for the pediatric patient to the corresponding pediatric patient, and at the same time use various intelligent emergency devices to monitor and collect the corresponding pediatric neural data after personalized management, including the electroencephalogram signal, heart rate signal, blood oxygen signal and motion state signal corresponding to the pediatric patient;

[0036] Upload the corresponding pediatric neural data after personalized management to the edge computing node for pediatric health anomaly feedback analysis to obtain the health anomaly signal feedback status corresponding to the pediatric patient;

[0037] Use the edge computing node to transmit the health anomaly signal feedback status corresponding to the pediatric patient to the cloud platform in real time with reduced transmission delay;

[0038] Through the health anomaly signal feedback status corresponding to the pediatric patient on the cloud platform, perform anomaly feedback optimization adjustment on the corresponding personalized management recommendation plan for the pediatric patient, and adjust the corresponding drug types and doses according to the anomaly fluctuation values of each electroencephalogram, heart rate, blood oxygen and motion state sub-signals in the health anomaly signal feedback status to generate the corresponding personalized optimization adjustment management plan for the pediatric patient.

[0039] Further, the uploading of the corresponding pediatric nerve data after personalized management to the edge computing node for pediatric health anomaly feedback analysis includes:

[0040] Upload the corresponding pediatric nerve data after personalized management to the edge computing node;

[0041] Use the edge computing node to perform statistical analysis on the fluctuation amplitude and frequency of each sub-signal in the pediatric nerve data to obtain the fluctuation amplitude and fluctuation frequency corresponding to the pediatric nerve sub-signal;

[0042] Obtain the signal distribution mean and signal distribution standard deviation corresponding to a period of time through each sub-signal in the pediatric nerve data, and based on the signal distribution mean and signal distribution standard deviation corresponding to a period of time, use the pediatric signal anomaly calculation formula to calculate the abnormal fluctuation of the fluctuation amplitude and fluctuation frequency corresponding to the pediatric nerve sub-signal, so as to obtain the abnormal fluctuation value corresponding to the pediatric nerve sub-signal;

[0043] Based on the abnormal fluctuation value corresponding to the pediatric nerve sub-signal, perform pediatric health anomaly feedback analysis on each sub-signal in the pediatric nerve data during the same period to obtain the health anomaly signal feedback status corresponding to the pediatric patient.

[0044] Further, the specific formula for calculating pediatric signal anomalies is:

[0045]

[0046] In the formula, Δε is the abnormal fluctuation value, T is the time period interval, t is the time variable parameter, X(t) is the specific value corresponding to the pediatric nerve sub-signal at time t, μ is the signal distribution mean corresponding to the time period T, σ is the signal distribution standard deviation corresponding to the time period T, ω is the fluctuation amplitude, and f is the fluctuation frequency.

[0047] Advantages of the present invention:

[0048] The intelligent management system for pediatric neurological data based on artificial intelligence proposed by the present invention is generally composed of a pediatric neurological data acquisition module, a pediatric multi-modal fusion module, a personalized management recommendation module, and a recommendation plan abnormal feedback optimization module. Compared with the prior art, the beneficial effect of this application is that it can collect the neurophysiological data of pediatric patients in real time through intelligent hardware devices, and can obtain various key neurophysiological indicators including electroencephalogram (EEG), electrocardiogram (ECG), blood oxygen saturation, and motion state. Through wireless technology, these data can be quickly transmitted to intelligent medical devices, and these devices can ensure the timeliness and security of the data by accessing the cloud platform. As the data distribution center, the cloud platform can not only store and process large-scale data, but also provide powerful computing resources for subsequent data analysis. The key to this process is that the real-time and efficient upload of data avoids errors or omissions caused by manual recording or too long intervals, ensuring the accuracy of information. In addition, the cloud platform can realize the centralized management of data, providing a solid foundation for subsequent analysis and personalized management. Secondly, after obtaining the initial neurophysiological data, the convolutional neural network (CNN) is used to correct abnormal noise in the data, which can effectively remove the noise interference in the sensor acquisition process and improve the accuracy and reliability of the data. The convolutional neural network is good at image processing and can effectively extract features from waveform data, removing redundant and noise parts, thereby improving the efficiency of anomaly detection. The key to this process is that the corrected data is closer to the real physiological state, providing a more accurate basis for subsequent analysis and personalized management. Based on the cross-data fusion of multi-modal neural networks, different types of neurophysiological data (such as electroencephalogram, electrocardiogram, blood oxygen saturation, motion state, etc.) are effectively integrated, enabling data from different sources to complement each other, thus presenting a more comprehensive and multi-dimensional neurophysiological state. This cross-modal fusion technology can help doctors comprehensively understand the health status of children's nervous systems, not only limited to a single physiological indicator, but also considering the correlation between various indicators, greatly improving the accuracy and comprehensiveness of personalized management.Then, based on the generated multi-modal fusion data, personalized management recommendations can be made using reinforcement learning algorithms. According to the specific neurophysiological characteristics and health status of each pediatric patient, a personalized treatment management plan can be formulated. Reinforcement learning is a self-learning method based on feedback adjustment. When applied in the medical field, it can continuously optimize treatment strategies according to the real-time data of patients. For example, when the effects of certain interventions are significant, the system will automatically increase the application intensity of the corresponding treatment methods, while adjusting or reducing the use of less effective measures. The key to this dynamically adjusted personalized recommendation plan is that the personalized management plan is no longer a one-size-fits-all "standardized" model, but is refined based on specific data, which can more effectively respond to the unique needs of each pediatric patient, thereby improving the personalized management effect and reducing unnecessary waste of medical resources. Finally, after applying the personalized management plan to pediatric patients, various intelligent emergency devices are used to re-monitor the neural data of children, enabling real-time collection and feedback of the health status during the treatment process. When abnormal health signals appear in data collection and processing, the edge computing node can quickly analyze and give preliminary feedback, timely detecting potential health problems. Through the real-time processing of edge computing, the system can respond quickly without relying on a centralized cloud platform, which is particularly important for situations requiring emergency intervention. The key effect of this stage is that by performing local data processing with edge computing, network latency can be reduced, the system response speed can be improved, and delays during data transmission can be avoided. Based on these abnormal health feedbacks, the cloud platform will further adjust and optimize the personalized management recommendation plan, making the personalized management plan more targeted and adaptable. This optimization and adjustment mechanism has significant value in practical applications because it ensures the continuous improvement of the personalized treatment management plan and the dynamic adaptation of patient health management, contributing to improving the long-term health prognosis of pediatric patients and reducing personalized management errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Other features, objectives, and advantages of the present invention will become more apparent by reading the detailed description of the non-limiting embodiments with reference to the following drawings:

[0050] Figure 1 Schematic diagram of the modules of the intelligent management system for pediatric neural data based on artificial intelligence of the present invention;

[0051] Figure 2 is Figure 1 Schematic diagram of the functional process of the pediatric neural data acquisition module in

[0052] Figure 3 is Figure 1 Schematic diagram of the functional process of the pediatric multi-modal fusion module in DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] The technical system of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0054] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor systems and / or microcontroller systems.

[0055] It should be understood that although terms such as "first" and "second" may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly, the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed related items.

[0056] To achieve the above object, please refer to Figures 1 to 3 , the present invention provides an intelligent management system for pediatric neurological data based on artificial intelligence, and the system includes the following modules:

[0057] A pediatric neurological data acquisition module, configured to collect pediatric neurological physiological data corresponding to pediatric patients in real time through corresponding intelligent hardware devices, and transmit the data to the corresponding intelligent medical devices of pediatric patients in real time through wireless technology; upload the collected pediatric neurological physiological data to the cloud platform through the intelligent medical devices;

[0058] A pediatric multi-modal fusion module, configured to use the cloud platform and based on a convolutional neural network to correct abnormal noise points in the pediatric neurological physiological data to obtain corrected pediatric neurological abnormal data; perform cross-data fusion on each sub-data in the corrected pediatric neurological abnormal data based on a multi-modal neural network to generate pediatric neurological multi-modal fusion data;

[0059] A personalized management recommendation module, configured to perform personalized recommendation based on the pediatric neurological multi-modal fusion data and in combination with reinforcement learning to generate a corresponding personalized management recommendation plan for pediatric patients;

[0060] The recommended solution abnormal feedback optimization module is used to apply the personalized management recommended solution for pediatric patients to the corresponding pediatric patients, monitor and collect the corresponding pediatric nerve data after personalized management by using various intelligent emergency devices, and use the edge computing node to perform pediatric health abnormal feedback analysis on the corresponding pediatric nerve data to obtain the health abnormal signal feedback status of the corresponding pediatric patients; based on the health abnormal signal feedback status of the corresponding pediatric patients, the cloud platform is used to perform abnormal feedback optimization adjustment on the corresponding pediatric patient personalized management recommended solution to generate the corresponding pediatric personalized optimization adjustment management plan.

[0061] In the embodiment of the present invention, please refer to Figure 1 As shown, it is a module schematic diagram of the intelligent management system for pediatric nerve data based on artificial intelligence of the present invention. In this example, the intelligent management system for pediatric nerve data based on artificial intelligence includes the following modules:

[0062] S1: Pediatric nerve data acquisition module, which is used to collect the corresponding pediatric nerve physiological data of pediatric patients in real time through the corresponding intelligent hardware devices, and transmit it to the corresponding intelligent medical devices of pediatric patients in real time through wireless technology; upload the collected pediatric nerve physiological data to the cloud platform through the intelligent medical devices.

[0063] In the embodiment of the present invention, an electroencephalogram monitor is selected to collect electroencephalogram signals. This device is equipped with a 64-channel electrode cap and works at a sampling frequency of 1000 times per second, which can accurately capture brain electrical activities. The heart rate monitor measures the heart rate by the oscillometric method and collects data every 30 seconds. The finger clip type blood oxygen sensor supporting the portable multi-parameter monitor collects the respiratory blood oxygen saturation data once per second. The triaxial accelerometer is used as a motion sensor and is fixed on the patient's wrist to collect the nerve motion state data 50 times per second. These intelligent hardware devices transmit the collected pediatric nerve physiological data to the intelligent medical devices worn by pediatric patients in real time through Bluetooth or Wi-Fi wireless technology. For example, a customized intelligent bracelet has a built-in low-power Bluetooth module and a Wi-Fi module, which can stably receive data. The intelligent medical device runs an application program customized based on the Android system, sorts the received data into a specific format, and uploads it to the cloud platform built by the Elastic Compute Service (ECS) of Alibaba Cloud through the 5G network. For example, the intelligent bracelet packs various data collected within 1 minute every 5 minutes, including information such as time stamps and patient IDs, and sends it to the specified receiving interface of the cloud server through the 5G network.

[0064] S2: The pediatric multi-modal fusion module is used to correct abnormal noise points in pediatric neurophysiological data by using the cloud platform and based on a convolutional neural network to obtain corrected pediatric neuro-abnormal data; perform cross-data fusion on each sub-data in the corrected pediatric neuro-abnormal data based on a multi-modal neural network to generate pediatric neuro multi-modal fusion data;

[0065] In an embodiment of the present invention, on the Alibaba Cloud ECS instance of the cloud platform, a convolutional neural network model is built using the TensorFlow framework of Python, and pediatric neurophysiological data is read from cloud storage. The data format is CSV, including electroencephalogram, heart rate, blood oxygen saturation, and motion state data. The data is divided into multiple segments in chronological order, and each segment contains 1000 sampling points. Taking the electroencephalogram data segment as an example, it is converted into a two-dimensional array form as the input of the convolutional neural network. The model structure includes multiple convolutional layers and pooling layers. The convolutional layer uses a 3×3 convolutional kernel, the stride is 1, and the padding method is'same' to extract data features through multiple layers of convolution. The pooling layer selects max pooling, the pooling kernel size is 2×2, and the stride is 2 to reduce the dimension of the feature map. Through model training, abnormal noise points in the data are identified and corrected to obtain corrected pediatric neuro-abnormal data. For example, if the amplitude of a certain point in the electroencephalogram data deviates abnormally from the surrounding data, the model corrects this point to a reasonable value by learning the feature patterns of normal data. Then, a multi-modal neural network is built using the Keras framework. The network includes multiple input layers, corresponding to different sub-data types respectively. After each input layer, a fully connected layer is connected for preliminary feature extraction. Then, the outputs of each sub-network are fused through a concatenation operation, and the cross-data correlation features are further explored through multiple layers of fully connected layers to generate pediatric neuro multi-modal fusion data.

[0066] S3: The personalized management recommendation module is used to perform personalized recommendation based on the pediatric neuro multi-modal fusion data and combined with reinforcement learning to generate a corresponding personalized management recommendation plan for pediatric patients;

[0067] In the embodiments of the present invention, based on pediatric neurological multi-modal fusion data, on the cloud platform, the TensorFlow framework of Python is combined with a reinforcement learning algorithm (such as the Deep Q-Network DQN) for personalized recommendation, and a Deep Q-Network model is constructed. The model includes multiple fully connected layers. The number of neurons in the input layer is the same as the dimension of the pediatric neurological multi-modal fusion data. Two fully connected layers are set in the middle layer, and the number of neurons in each layer is 128 and 64 respectively. The activation function uses ReLU. The number of neurons in the output layer is the same as the number of possible treatment management plans. Each neuron outputs the Q value corresponding to a treatment management plan. Read the original treatment management plan data from the hospital's treatment management database (MySQL database), including information such as the symptoms, diagnosis results, treatment management plans adopted, and treatment effect feedback of different pediatric patients. Use the pediatric neurological multi-modal fusion data as the state vector to input into the model. The model selects the treatment management plan with the largest Q value as the recommended plan according to the current state. Design a reward function and adjust the reward value according to the treatment effect feedback. If the neurophysiological data indicators of the patient approach the normal range after treatment, give a positive reward; otherwise, give a negative reward. For example, after the patient receives a certain plan, the energy of the abnormal frequency band in the electroencephalogram decreases, and the reward value is set to +10. Through continuous training, the model generates a personalized management recommendation plan for the corresponding pediatric patient, specifying information such as the types and doses of drugs.

[0068] S4: The recommended plan abnormal feedback optimization module is used to apply the personalized management recommendation plan for pediatric patients to the corresponding pediatric patients and use various intelligent emergency devices to monitor and collect the corresponding pediatric neurological data after personalized management, and use the edge computing node to perform pediatric health abnormal feedback analysis on the corresponding pediatric neurological data to obtain the health abnormal signal feedback status corresponding to the pediatric patients; based on the health abnormal signal feedback status corresponding to the pediatric patients, optimize and adjust the personalized management recommendation plan for the corresponding pediatric patients through the cloud platform to generate the corresponding pediatric personalized optimization and adjustment management plan.

[0069] In an embodiment of the present invention, the generated personalized management recommendation plan for pediatric patients is sent to the terminal devices of medical staff, such as tablets, through the hospital's information management system (HIS). The medical staff implement treatment for the patients according to the plan, such as administering medications according to the recommended drug types and dosages. At the same time, using an electroencephalogram monitor, a heart rate monitor, a blood oxygen sensor supporting a portable multi-parameter monitor, and a triaxial accelerometer motion sensor, continuously monitor and collect the corresponding pediatric nerve data after personalized management. These devices transmit the data to a data collection terminal equipped with an edge computing module in the ward through Bluetooth or Wi-Fi. The edge computing node analyzes the data using the NumPy library and SciPy library of Python, calculates the fluctuation amplitude, frequency, mean, and standard deviation of each sub-signal, obtains the abnormal fluctuation value through a specific abnormal calculation formula, and then judges the health abnormal state, such as "mild electroencephalogram abnormality, tachycardia abnormality", etc., to generate the corresponding health abnormal signal feedback state for pediatric patients. The edge computing node transmits this state data to the cloud platform through the 5G network. On the cloud platform, using the big data analysis service (MaxCompute) of Alibaba Cloud and the pre-developed algorithm, adjust the personalized management recommendation plan for pediatric patients according to the health abnormal signal feedback state. For example, if there is tachycardia abnormality, combine medical knowledge and historical cases, adjust the drug dosage or change the drug type, generate the corresponding personalized optimization and adjustment management plan for pediatrics, and feedback it to the terminal of the medical staff through HIS for timely treatment adjustment.

[0070] Further, the pediatric nerve data acquisition module includes the following functions:

[0071] Real-time collect the electroencephalogram signal data corresponding to pediatric patients through an electroencephalogram monitor;

[0072] Real-time collect the heart rate signal data corresponding to pediatric patients through a heart rate monitor;

[0073] Real-time collect the respiratory blood oxygen saturation data corresponding to pediatric patients through a blood oxygen sensor;

[0074] Real-time collect the nerve movement state data corresponding to pediatric patients through a motion sensor;

[0075] Synthesize the electroencephalogram signal data, heart rate signal data, respiratory blood oxygen saturation data, and nerve movement state data corresponding to pediatric patients into a data set to obtain the corresponding pediatric neurophysiological data, and transmit it to the intelligent medical device worn by the corresponding pediatric patient in real time through the corresponding wireless technologies of Bluetooth or Wi-Fi;

[0076] Upload the collected pediatric neurophysiological data to the cloud platform through the intelligent medical device.

[0077] As an embodiment of the present invention, refer toFigure 2 As shown in Figure 1 the schematic diagram of the functional flow of the pediatric neurological data acquisition module in

[0078] S11: Real-time collect the electroencephalogram (EEG) signal data corresponding to pediatric patients through an EEG monitor;

[0079] In an embodiment of the present invention, by selecting a corresponding high-precision EEG monitor, this device is equipped with a 64-channel electrode cap and can accurately collect brain electrical activity signals. Before wearing the electrode cap on a pediatric patient, first clean the patient's scalp to reduce skin resistance and ensure signal acquisition quality. Wear the electrode cap tightly and accurately on the patient's head, and place each electrode at the standard positions of the international 10-20 system to ensure full coverage of the main functional areas of the cerebral cortex. The EEG monitor collects the EEG signal data of pediatric patients in real time at a sampling frequency of 1000 times per second. For example, in a pediatric ward, a child undergoing a neurological examination starts from entering the ward, and the EEG monitor continuously works and records its EEG signals without interruption. The collected data is temporarily stored in the high-speed cache built into the monitor in digital form, and a data file is automatically generated every 5 minutes. The file format is the internationally common EDF+ format, which is convenient for subsequent data processing and analysis, and finally obtains the EEG signal data corresponding to pediatric patients.

[0080] S12: Real-time collect the heart rate signal data corresponding to pediatric patients through a heart rate monitor;

[0081] In an embodiment of the present invention, by adopting a corresponding heart rate monitor, this device measures the heart rate using the oscillometric method and has high precision and stability. Wear the cuff of the heart rate monitor correctly on the upper arm of the pediatric patient, and keep the center of the cuff at the same horizontal height as the heart to ensure the accuracy of the measurement. The heart rate monitor automatically measures the heart rate signal data of pediatric patients every 30 seconds. Each time a measurement is made, the pressure sensor inside the instrument senses the pressure change in the cuff, and calculates the heart rate value through an algorithm. For example, in a pediatric intensive care unit, a postoperative child continuously wears this heart rate monitor. The instrument displays the heart rate data obtained from each measurement on the screen in real time, and at the same time transmits the data to the local data collection terminal in the ward using Bluetooth Low Energy (BLE) technology. The local data collection terminal sorts these heart rate data in chronological order and generates a CSV file containing all heart rate measurement values during this time period every hour, which is stored in the local hard disk and waits to be integrated with other data, and finally obtains the heart rate signal data corresponding to pediatric patients.

[0082] S13: Real-time collect the respiratory oxygen saturation data corresponding to pediatric patients through an oxygen sensor;

[0083] In an embodiment of the present invention, by using a blood oxygen sensor supporting a corresponding portable multi-parameter monitor, the sensor adopts a finger clip design and measures respiratory blood oxygen saturation through optoelectronic blood oxygen detection technology. The finger clip blood oxygen sensor is gently clipped on the finger of a pediatric patient, ensuring that the light-emitting diode and photodetector of the sensor are aligned on both sides of the finger to ensure that light can pass through the finger smoothly. The blood oxygen sensor collects the respiratory blood oxygen saturation data of the pediatric patient in real time at a frequency of once per second. For example, in a general pediatric ward, a child suffering from a respiratory disease wears this blood oxygen sensor from the time of admission. The sensor works continuously, transmits the collected blood oxygen saturation data to the monitor host in real time. The monitor host processes and displays the data, and at the same time sends the data to the data storage module of the corresponding intelligent medical device through the Wi-Fi network. The storage module classifies and stores the data according to patient information and time sequence, and backs up the data every 12 hours to prevent data loss, and finally obtains the corresponding respiratory blood oxygen saturation data of the pediatric patient.

[0084] S14: Collect the corresponding neuro-motor state data of the pediatric patient in real time through a motion sensor;

[0085] In an embodiment of the present invention, by selecting a corresponding three-axis accelerometer as the motion sensor, the sensor is small in size, low in power consumption and high in measurement accuracy. The motion sensor is fixed on the wrist of the pediatric patient through a special strap to ensure that the sensor can accurately sense the hand movement of the patient. The motion sensor collects the neuro-motor state data of the pediatric patient at a frequency of 50 times per second, including the components of acceleration in the x, y, and z axes. For example, in a pediatric rehabilitation treatment room, a child undergoing rehabilitation training wears this motion sensor. The sensor transmits the collected three-axis acceleration data to the corresponding intelligent medical device through the SPI interface, and finally obtains the corresponding neuro-motor state data of the pediatric patient.

[0086] S15: Synthesize the electroencephalogram signal data, heart rate signal data, respiratory blood oxygen saturation data, and neuro-motor state data corresponding to the pediatric patient into a data set to obtain the corresponding pediatric neurophysiological data, and transmit it to the intelligent medical device worn by the pediatric patient in real time through the corresponding wireless technology such as Bluetooth or Wi-Fi;

[0087] In an embodiment of the present invention, a data integration server is set up at the nurse's station in the pediatric ward. The server is equipped with a high-performance CPU and a large-capacity memory, and runs the Windows Server operating system. The server establishes connections with the local data collection terminals corresponding to the electroencephalogram monitor, heart rate monitor, blood oxygen sensor, and motion sensor through Bluetooth and Wi-Fi modules respectively. When a new data file is generated by the local data collection terminal, the server automatically detects it and downloads the data file to the local hard disk of the server through the corresponding wireless connection. The server runs specially developed data integration software, which reads the downloaded electroencephalogram signal data (EDF+ format), heart rate signal data (CSV format), respiratory blood oxygen saturation data (stored in the HIS system and obtained through SQL query), and neuro-motor state data (in the local database). These data are aligned and integrated according to the time stamp. For example, for different types of data at the same moment, the software combines them into a record, and the record format is [time, electroencephalogram signal data, heart rate signal data, respiratory blood oxygen saturation data, neuro-motor state data]. The integrated data is stored in a new database table on the server to form pediatric neurophysiological data. At the same time, the server transmits these pediatric neurophysiological data to the intelligent medical devices worn by the pediatric patients, such as intelligent bracelets or smart watches, in real time through Wi-Fi. The devices inform the patients that the data has been updated through vibration or screen prompts.

[0088] S16: Upload the collected pediatric neurophysiological data to the cloud platform through the intelligent medical device.

[0089] In an embodiment of the present invention, the intelligent medical device worn by the pediatric patient adopts an operating system customized based on the Android system and is built with a 5G communication module. After receiving the pediatric neurophysiological data, the intelligent medical device runs a dedicated upload application program, which encrypts the data. The encryption algorithm uses AES-256 to ensure the security of data transmission. Then, the encrypted data is sent to the cloud platform through the 5G network. The cloud platform adopts Alibaba Cloud's Elastic Compute Service (ECS) and Object Storage Service (OSS). The intelligent medical device sends the data to a specific API interface of Alibaba Cloud, and the interface stores the received data in a dedicated storage bucket of OSS. The storage bucket is classified according to the patient ID for easy data management and query. At the same time, after the data is stored in OSS, it triggers Alibaba Cloud's Function Compute Service (FC). FC automatically decompresses and converts the format of the data, converts it into the Parquet format suitable for data analysis, and stores it in Alibaba Cloud's Table Store (OTS) to provide data support for subsequent artificial intelligence-based data analysis and processing.

[0090] Furthermore, the pediatric multi-modal fusion module includes the following functions:

[0091] Using the short-time Fourier transform through a cloud platform to perform data feature input conversion on the corresponding electroencephalogram, heart rate, blood oxygen saturation, and motion state data in pediatric neurophysiological data, so as to convert the corresponding data into a feature format suitable for deep learning input, and obtain pediatric neuro-temporal data features;

[0092] Obtain the labeled pediatric neuro-primitive data, including pediatric neuro-data corresponding to labeled normal and abnormal noise points;

[0093] Use the labeled pediatric neuro-primitive data to train the corresponding convolutional neural network model so that it can identify the corresponding normal and abnormal noise points in pediatric neuro-data, and input the pediatric neuro-temporal data features into the trained convolutional neural network model for abnormal noise point detection and division, so as to obtain pediatric neuro-normal data and pediatric neuro-abnormal noise data;

[0094] Based on the pediatric neuro-normal data and the pediatric neuro-abnormal noise data, perform abnormal noise correction at the corresponding abnormal noise positions in the pediatric neurophysiological data to obtain pediatric neuro-abnormal corrected data;

[0095] Based on a multi-modal neural network, perform cross-data fusion on each sub-data in the pediatric neuro-abnormal corrected data to generate pediatric neuro-multi-modal fusion data.

[0096] As an embodiment of the present invention, refer to Figure 3 shown, which is Figure 1 a schematic diagram of the functional flow of the pediatric multi-modal fusion module in

[0097] S21: Using the short-time Fourier transform through a cloud platform to perform data feature input conversion on the corresponding electroencephalogram, heart rate, blood oxygen saturation, and motion state data in pediatric neurophysiological data, so as to convert the corresponding data into a feature format suitable for deep learning input, and obtain pediatric neuro-temporal data features;

[0098] In an embodiment of the present invention, on an Elastic Compute Service (ECS) instance of Alibaba Cloud, the short-time Fourier transform operation is performed using the SciPy library in Python, and pediatric neurophysiological data is read from the Table Store (OTS) of Alibaba Cloud. This data includes electroencephalogram, heart rate, blood oxygen saturation, and motion state data. Taking the electroencephalogram data as an example, it is divided into multiple segments with a duration of 1 second in chronological order. For each segment, the window length is set to 0.2 seconds, the window type is a Hanning window, and the overlap rate is 50%. By calling the scipy.signal.stft() function to perform the short-time Fourier transform, the electroencephalogram data in the time domain is converted into frequency domain data, obtaining the amplitude and phase information of this segment at different frequencies. Such operations are performed on all electroencephalogram data segments, and the results are organized in the form of a two-dimensional array. Similarly, similar processing is performed on the heart rate, blood oxygen saturation, and motion state data. For example, after the short-time Fourier transform of the heart rate data, the heart rate change characteristics of each time segment at different frequencies are obtained. These converted data are combined according to the data type to form pediatric neuro-temporal data features, and the data format is [electroencephalogram frequency domain feature array, heart rate frequency domain feature array, blood oxygen saturation frequency domain feature array, motion state frequency domain feature array], making it suitable as the input of a deep learning model.

[0099] S22: Obtain the labeled pediatric neuron data, including pediatric nerve data corresponding to normal and abnormal noise points.

[0100] In an embodiment of the present invention, the labeled pediatric neuron data is obtained from the hospital's clinical data management system. This data is jointly annotated by professional pediatric neurologists and data annotators. During the annotation process, the doctor marks the normal and abnormal noise points in the electroencephalogram, heart rate, blood oxygen saturation, and motion state data according to clinical experience and medical knowledge. For example, in the electroencephalogram data, the high-frequency spike noise caused by equipment interference is marked as an abnormal noise point, while the normal electroencephalogram rhythm part is marked as normal data; in the heart rate data, the points with sudden heart rate changes that do not conform to physiological laws are marked as abnormal noise points, and the data within the stable normal heart rate range is marked as normal data. These labeled data are organized into a structured data table according to the patient ID and time sequence and stored in a relational database within the hospital, such as a MySQL database. The data table includes fields such as patient ID, data type (electroencephalogram, heart rate, etc.), time stamp, data value, and annotation result (normal or abnormal noise point), which is convenient for subsequent reading and used to train a convolutional neural network model.

[0101] S23: Train the corresponding convolutional neural network model using the labeled pediatric neuron data so that it can identify the corresponding normal and abnormal noise points in the pediatric nerve data, and input the pediatric nerve time series data features into the trained convolutional neural network model for abnormal noise point detection and division to obtain pediatric nerve normal data and pediatric nerve abnormal noise data;

[0102] In the embodiment of the present invention, a convolutional neural network model is built using the TensorFlow framework of Python. The model structure includes multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layer uses a 3×3 convolutional kernel, the stride is 1, and the padding method is'same' to extract features in the data through multiple convolutions. The pooling layer selects max pooling, the pooling kernel size is 2×2, and the stride is 2 to reduce the feature map dimension. The fully connected layer is set with 128 neurons, and the activation function is ReLU. Read the labeled pediatric neuron data from the internal MySQL database of the hospital, divide it into a training set, a validation set, and a test set according to the ratio of 80%, 10%, and 10%. Set the learning rate to 0.001, use the Adam optimizer, and train the model for 100 epochs. During the training process, the model calculates the prediction result through forward propagation, compares it with the true labeled result, calculates the loss value using the cross-entropy loss function, and then updates the network parameters through backpropagation. After training, input the previously obtained pediatric nerve time series data features into the trained convolutional neural network model for forward propagation. The model judges the normal and abnormal noise points in the input data according to the learned features, outputs the probability that each data point belongs to a normal or abnormal noise point, divides the data points with a probability greater than 0.5 into normal data, and divides the data points less than or equal to 0.5 into abnormal noise data, thereby obtaining pediatric nerve normal data and pediatric nerve abnormal noise data.

[0103] S24: Based on the pediatric nerve normal data and the pediatric nerve abnormal noise data, correct the abnormal noise points at the corresponding abnormal noise positions in the pediatric nerve physiological data to obtain pediatric nerve abnormal correction data;

[0104] In an embodiment of the present invention, for pediatric neurological abnormal noise data, the Pandas library and NumPy library of Python are used to correct the abnormal noise. For electroencephalogram (EEG) abnormal noise data, assuming that the abnormal noise is manifested as an abnormal increase in amplitude at a certain time point, by analyzing the normal EEG data for a period of time before and after this time point (such as 0.5 seconds before and after), the mean and standard deviation are calculated. If the amplitude of the abnormal noise exceeds the mean plus 3 times the standard deviation, the amplitude of the abnormal noise is corrected to the mean plus 2 times the standard deviation. For heart rate abnormal noise data, if the abnormal noise is manifested as a sudden decrease in heart rate and is lower than the lower limit of the normal heart rate range, by the method of linear interpolation, the heart rate value that should be at this time point is calculated based on the normal heart rate data points before and after, and the abnormal noise is corrected. For example, if the heart rate at the previous moment is 80 beats per minute and the heart rate at the next moment is 85 beats per minute, the heart rate value to be corrected at the abnormal noise moment is calculated by linear interpolation to be 82.5 beats per minute. Similar correction methods based on data characteristics and normal data before and after are also used for blood oxygen saturation and motion state abnormal noise data. The corrected abnormal noise data is combined with pediatric neurological normal data, and finally pediatric neurological abnormal corrected data is obtained.

[0105] S25: Based on a multi-modal neural network, cross-data fusion is performed on each sub-data within the pediatric neurological abnormal corrected data to generate pediatric neurological multi-modal fusion data.

[0106] In an embodiment of the present invention, a multi-modal neural network is built by using the Keras framework of Python. The network structure includes multiple input layers, corresponding to EEG, heart rate, blood oxygen saturation, and motion state data respectively. After each input layer, a fully connected layer is connected to perform preliminary feature extraction on data of different modalities. Then, the features preliminarily extracted are fused through a concatenation operation, and further cross-data correlation features are mined through multiple fully connected layers. For example, after the EEG feature and the heart rate feature are concatenated, they are fused through a fully connected layer containing 64 neurons. The previously obtained pediatric neurological abnormal corrected data is input into the corresponding input layer of the multi-modal neural network according to the data type. The learning rate is set to 0.0001, and the RMSprop optimizer is used to train the network for 50 epochs. During the training process, the network learns the relationship between data of different modalities and continuously optimizes the network parameters through forward propagation and backward propagation. After the training is completed, the pediatric neurological abnormal corrected data is input, and the network outputs the fused features. These features are organized into a new data format to generate pediatric neurological multi-modal fusion data, providing more comprehensive and effective data support for subsequent artificial intelligence-based pediatric neurological data analysis and processing.

[0107] Further, the abnormal noise correction at the corresponding abnormal noise position in the pediatric neurophysiological data based on the pediatric neural normal data and the pediatric neural abnormal noise data includes:

[0108] Perform statistical analysis on the data distribution of the pediatric neural abnormal noise data to obtain the mean value of the pediatric neural noise data distribution and the standard deviation of the pediatric neural noise data distribution;

[0109] In the embodiment of the present invention, the statistical analysis of the data distribution of the pediatric neural abnormal noise data is performed by using the Pandas library and the NumPy library of Python. Assume that the pediatric neural abnormal noise data is stored in a DataFrame data structure of Pandas, and the data is divided into different columns according to the data type (electroencephalogram, heart rate, blood oxygen saturation, and motion state data). Each row represents the data at a time point. Taking the electroencephalogram abnormal noise data column as an example, the mean value of the data in this column is calculated by calling the numpy.mean() function, that is, the mean value of the pediatric neural noise data distribution. For example, for the electroencephalogram abnormal noise data column eeg_noise_data, the operation to calculate the mean value is as follows: eeg_noise_mean = np.mean(eeg_noise_data). Similarly, the standard deviation of the data in this column is calculated by using the numpy.std() function, that is, the standard deviation of the pediatric neural noise data distribution. The calculation code is eeg_noise_std = np.std(eeg_noise_data). Such operations are performed on the abnormal noise data columns of heart rate, blood oxygen saturation, and motion state to obtain the respective mean values and standard deviations of the pediatric neural noise data distribution corresponding to them. These statistics reflect the central tendency and dispersion degree of the abnormal noise data in different data types.

[0110] Preferably, based on the mean value of the pediatric neural noise data distribution and the standard deviation of the pediatric neural noise data distribution, and combined with linear fitting interpolation, perform noise fitting correction on the corresponding normal data distribution in the pediatric neural normal data, and interpolate and fit the corresponding normal data distribution according to the mean value of the pediatric neural noise data distribution and the standard deviation of the pediatric neural noise data distribution to generate an abnormal correction value corresponding to the abnormal noise, and obtain the abnormal correction value of the pediatric neural noise;

[0111] In the embodiment of the present invention, based on the mean value of the pediatric neural noise data distribution and the standard deviation of the pediatric neural noise data distribution obtained previously, the linear fitting function scipy.stats.linregress() in the SciPy library of Python is used to perform noise fitting correction on the corresponding normal data distribution in the pediatric neural normal data. Taking the normal heart rate data as an example, assume that the normal heart rate data is stored in a one-dimensional NumPy array heart_rate_normal_data and arranged in chronological order. According to the distribution characteristics of the abnormal noise data, the range of linear fitting is determined. For example, if the abnormal noise data is concentrated in a certain time interval, a section of normal data before and after this interval is selected as the fitting sample. Let the fitting sample be heart_rate_fit_sample. The parameters of the linear regression equation of this sample are calculated through the scipy.stats.linregress() function to obtain the slope slope and the intercept intercept. Then, according to the mean value and standard deviation of the pediatric neural noise data distribution, interpolation fitting is performed within the normal data distribution range. For example, for a normal data point x near the abnormal noise, its predicted value y_pred is calculated using the linear regression equation y = slope × x + intercept. If the deviation between this point and the predicted value is large, it is adjusted according to the mean value and standard deviation of the noise data distribution. Assume that the mean value of the noise data distribution is noise_mean and the standard deviation is noise_std. If abs(x - y_pred) > 3 × noise_std, then this point is corrected to y_pred + noise_mean. In this way, noise fitting correction is performed on all normal data distributions, and abnormal correction values corresponding to abnormal noise are generated by interpolating and fitting the corresponding normal data distributions according to the mean value of the pediatric neural noise data distribution and the standard deviation of the pediatric neural noise data distribution. Finally, the abnormal correction value of the pediatric neural noise is obtained.

[0112] Preferably, based on the abnormal correction value of the pediatric neural noise, abnormal noise correction is performed at the position of the corresponding abnormal noise in the pediatric neural physiological data to obtain the abnormal correction data of the pediatric nerve.

[0113] In an embodiment of the present invention, by using the Pandas library of Python, the abnormal noise at the corresponding abnormal noise position in the pediatric neurophysiological data is corrected based on the previously obtained pediatric neuro noise abnormal correction value. It is assumed that the pediatric neurophysiological data is stored in a DataFrame of Pandas and the positions of the abnormal noise have been marked. For the abnormal noise in the electroencephalogram, find the row and column where the abnormal noise is located, and replace the data at the corresponding position with the correction value of the electroencephalogram part in the pediatric neuro noise abnormal correction value. For example, if the abnormal noise in the electroencephalogram is in the 100th row and the 2nd column, and the corresponding correction value of the electroencephalogram part in the pediatric neuro noise abnormal correction value is eeg_corrected_value, then perform the operation of data_frame.at[100, 'electroencephalogram'] = eeg_corrected_value. In the same way, replace the data at the abnormal noise positions in the heart rate, blood oxygen saturation, and motor state data. After completing the data correction at all abnormal noise positions, use the corrected DataFrame as the pediatric neuro abnormal correction data. This data removes the interference of abnormal noise and provides a more accurate data basis for subsequent pediatric neuro data analysis based on artificial intelligence.

[0114] Further, the cross-data fusion of each sub-data in the pediatric neuro abnormal correction data based on the multi-modal neural network includes:

[0115] Perform data feature analysis on each sub-data in the pediatric data abnormal correction data to implement time-frequency analysis of the electroencephalogram signal to obtain the corresponding electroencephalogram signal time-frequency features, perform periodic analysis of the heart rate signal to obtain the corresponding heart rate signal periodic features, perform data distribution statistics on the blood oxygen saturation and neural motor state data to obtain the corresponding data distribution features of blood oxygen and motor state, and obtain the sub-modal features corresponding to each pediatric sub-data type;

[0116] In the embodiments of the present invention, by using the SciPy library and Matplotlib library of Python to perform data feature analysis on each sub-data within the pediatric data anomaly correction data. For the electroencephalogram (EEG) signal, short-time Fourier transform (STFT) is used for time-frequency analysis. Assuming that the EEG data is stored in a one-dimensional NumPy array eeg_data, the window length is set to 0.2 seconds, the window type is Hanning window, and the overlap rate is 50%. By calling the scipy.signal.stft() function, the EEG data in the time domain is converted into frequency domain data, and the amplitude information at different times and frequencies is obtained. For example, f, t, Zxx = scipy.signal.stft(eeg_data, fs = 1000, window = 'hann', nperseg = 200, noverlap = 100), where f is the frequency array, t is the time array, and Zxx is the corresponding amplitude array. By analyzing the Zxx array, time-frequency characteristics such as the energy change of the EEG signal over time in different frequency bands (such as α, β, γ frequency bands) are obtained. For the heart rate signal, assuming that the heart rate data is stored in the heart_rate_data array, autocorrelation function is used for periodic analysis. The autocorrelation function of the heart rate data is calculated by the scipy.signal.correlate() function, and the peak position of the autocorrelation function is found. For example, autocorr = scipy.signal.correlate(heart_rate_data, heart_rate_data), so as to determine the period of the heart rate signal. Analyze the autocorr array to find the delay time corresponding to the peak, which is the period of the heart rate signal, and then obtain the periodic characteristics of the heart rate signal, such as the average period, the range of period change, etc. For the blood oxygen saturation and nerve movement state data, the Pandas library and NumPy library of Python are used to perform data distribution statistics. Assuming that the blood oxygen saturation data is stored in the spO2_data column and the nerve movement state data is stored in the motion_data column, both are in the Pandas DataFrame data structure. The mean is calculated by the numpy.mean() function, the standard deviation is calculated by the numpy.std() function, and different percentiles (such as 25%, 75%) are calculated by the numpy.percentile() function to obtain the data distribution characteristics corresponding to blood oxygen and movement state, such as the mean, standard deviation, skewness of the data distribution, etc., and finally the sub-modal characteristics corresponding to each pediatric sub-data type are obtained.

[0117] Preferably, each network branch corresponding to the multi-modal neural network is used to process the sub-modal features corresponding to the pediatric sub-data types, and the attention mechanism is used to weight the importance weights of the sub-modal features. At the same time, based on the importance weights, the sub-modal features corresponding to each pediatric sub-data type are fused across data to generate pediatric neural multi-modal fusion data.

[0118] In the embodiment of the present invention, a multi-modal neural network is built by using the Keras framework of Python. The network includes multiple branches, and each branch corresponds to a pediatric sub-data type (electroencephalogram, heart rate, blood oxygen saturation, neurological motor state). Taking the electroencephalogram branch as an example, a sub-network including multiple fully connected layers is constructed. The input is the time-frequency features of the previously obtained electroencephalogram signal. After passing through the first fully connected layer, the number of neurons is set to 64, and the activation function is ReLU to perform preliminary extraction and transformation of the features. The heart rate branch also constructs a fully connected layer sub-network, and the input is the periodic features of the heart rate signal and is processed similarly. The blood oxygen saturation and neurological motor state branches also construct sub-networks in the same way. The attention mechanism is used to weight the importance weights of the sub-modal features. After the output layer of each sub-network, an attention module is added. Taking the output eeg_output of the electroencephalogram sub-network as an example, the attention weight eeg_weight = Dense(1)(eeg_output) is calculated through a fully connected layer including a single neuron, and after passing through the Softmax activation function for normalization, a weight value between 0 and 1 is obtained, which reflects the importance degree of the sub-modal features of the electroencephalogram signal. The same operation is performed on the outputs of other sub-networks to obtain the weights of the sub-modal features of the heart rate signal, blood oxygen saturation, and neurological motor state. Finally, based on the importance weights, the sub-modal features corresponding to each pediatric sub-data type are fused across data. The outputs of each sub-network are multiplied by their corresponding weights and then concatenated, such as fused_features = concatenate([eeg_output*eeg_weight, heart_rate_output*heart_rate_weight, spO2_output*spO2_weight, motion_output*motion_weight]), and then further fused and feature extracted through multiple fully connected layers to finally generate pediatric neural multi-modal fusion data, providing a more comprehensive and effective data basis for subsequent pediatric neural data analysis based on artificial intelligence.

[0119] Furthermore, the personalized management recommendation module includes the following functions:

[0120] Obtain the feature anomaly correlations between the pediatric data sub-modalities of pediatric patients through the pediatric neural multi-modal fusion data;

[0121] In the embodiments of the present invention, by using the NumPy library and the Pandas library of Python to process pediatric neurological multimodal fusion data, the feature anomaly correlation between each pediatric data sub-modal of pediatric patients is obtained. Assuming that the pediatric neurological multimodal fusion data is stored in a Pandas DataFrame, and different sub-modal features are stored in different columns, such as the time-frequency feature column eeg_features of electroencephalogram signals, the periodic feature column heart_rate_features of heart rate signals, the data distribution feature column spO2_features of blood oxygen saturation, and the data distribution feature column motion_features of neurological motion states. First, the data is standardized to make different features have the same dimension. For example, for the time-frequency feature column of electroencephalogram signals, it is standardized by (eeg_features - eeg_features.mean()) / eeg_features.std(). Then, the Pearson correlation coefficient matrix between different sub-modal feature columns is calculated using the numpy.corrcoef() function. For example, to calculate the correlation coefficient between the time-frequency features of electroencephalogram signals and the periodic features of heart rate signals, the operation is as follows: corr_matrix = np.corrcoef(eeg_features, heart_rate_features), eeg_hr_corr = corr_matrix[0, 1]. In this way, the correlation coefficients between all sub-modal features are calculated, and the absolute value size and positive or negative of the correlation coefficients are analyzed to determine the feature anomaly correlation between each pediatric data sub-modal. For example, if the absolute value of the correlation coefficient between the energy change of a certain frequency band of electroencephalogram signals and the periodic change of heart rate signals is greater than 0.5 and is positively correlated, it indicates that there may be a strong association between these two sub-modal features in abnormal situations.

[0122] Preferably, based on the feature anomaly correlation between each pediatric data sub-modal of pediatric patients and combined with reinforcement learning and the corresponding pediatric data sub-modal modeling, a corresponding treatment management recommendation reinforcement learning model is generated. The corresponding pediatric data sub-modal is used as the state vector, and the corresponding pediatric treatment management plan is recommended according to the original treatment management database. At the same time, a reward function is designed according to the corresponding effective actions to continuously reinforce the learning of the corresponding treatment management plan, so as to generate a corresponding personalized management recommendation plan for pediatric patients.

[0123] In an embodiment of the present invention, a corresponding treatment management recommendation reinforcement learning model is generated by using the TensorFlow framework of Python in combination with a reinforcement learning algorithm (such as the deep Q-network DQN). First, the original treatment management plan data is read from the treatment management database of the hospital. The database uses a MySQL relational database for storage, and the data includes information such as the symptoms, diagnosis results, treatment management plans adopted, and treatment effect feedback of different pediatric patients. The pediatric data sub-modal features are used as the state vector. For example, the time-frequency features of electroencephalogram signals, the periodic features of heart rate signals, the distribution features of blood oxygen saturation data, and the distribution features of neuro-motor state data are spliced into a multi-dimensional vector as the state input. A deep Q-network model is constructed. The model includes multiple fully connected layers. The number of neurons in the input layer is the same as the dimension of the state vector. For example, if the dimension of the state vector is 100, then there are 100 neurons in the input layer. Two fully connected layers are set in the middle layer, with 128 and 64 neurons in each layer respectively. The activation function uses ReLU. The number of neurons in the output layer is the same as the number of treatment management plans. Each neuron outputs the Q value corresponding to a treatment management plan. For example, if there are 10 treatment management plans, then there are 10 neurons in the output layer. According to the original treatment management database, the treatment management plan corresponding to each state vector is labeled as the initial training data of the model. A reward function is designed. Based on the treatment effect, if the neurophysiological data indicators of the patient approach the normal range after treatment, such as the reduction of the abnormal frequency band energy of the electroencephalogram and the restoration of the normal cycle of the heart rate, a positive reward is given; otherwise, if the indicators deteriorate, a negative reward is given. For example, when a patient receives a certain treatment plan, if the energy of the α frequency band of the electroencephalogram decreases from an abnormal high value to near the normal range, the reward value is set to +10; if the heart rate cycle becomes more unstable, the reward value is set to -5. By continuously inputting the state vector into the model, the model selects the treatment management plan with the largest Q value as the recommended plan according to the current state, and performs backpropagation according to the feedback of the reward function to update the model parameters, and continuously strengthens the learning of the corresponding treatment management plan. After a large number of training iterations, a corresponding personalized management recommendation plan for pediatric patients is generated. The plan specifies the types and doses of drugs corresponding to pediatric patients under the clinical pediatric nerve data. For example, for a certain pediatric epilepsy patient, according to the electroencephalogram signal, heart rate, blood oxygen saturation, and neuro-motor state data characteristics of the patient, the model recommends using the corresponding anti-epileptic drug A, with a dose of 5 mg / kg, 3 times a day, and finally generates a corresponding personalized management recommendation plan for pediatric patients.

[0124] Further, the personalized management recommendation plan for pediatric patients includes the types and doses of drugs corresponding to pediatric patients under the clinical pediatric nerve data.

[0125] Further, the recommended plan abnormal feedback optimization module includes the following functions:

[0126] Personalize the management recommendation plan for pediatric patients for decision support and apply it to the corresponding pediatric patients. At the same time, use various intelligent emergency devices to monitor and collect the corresponding pediatric neurological data after personalized management, including electroencephalogram signals, heart rate signals, blood oxygen signals, and motion state signals corresponding to pediatric patients.

[0127] In the embodiment of the present invention, the generated personalized management recommendation plan for pediatric patients is sent to the terminal devices of medical staff responsible for the patient, such as tablet computers or mobile nursing carts, through the hospital's information management system (HIS). According to the recommendation plan, medical staff formulate and implement a detailed treatment plan for pediatric patients. For example, for the recommended use of antiepileptic drug A with a dose of 5 mg / kg, three times a day, medical staff calculate the specific dosage per administration based on the patient's weight and administer the drug strictly according to the time interval. At the same time, use various intelligent emergency devices for data monitoring and collection. Use a high-precision electroencephalogram monitor of the same brand and model as before, wear an electrode cap for the patient according to the standard process, and continuously collect electroencephalogram signals at a sampling frequency of 1000 times per second; a heart rate monitor, correctly wear the cuff, and measure the heart rate signal every 30 seconds; a blood oxygen sensor supporting a portable multi-parameter monitor, clip it on the patient's finger, and collect blood oxygen signals once per second; a three-axis accelerometer motion sensor, fix it on the patient's wrist, and collect nerve motion state signals 50 times per second. These devices transmit the collected data to the data collection terminal in the ward in real time through Bluetooth or Wi-Fi wireless technology.

[0128] Preferably, upload the corresponding pediatric neurological data after personalized management to the edge computing node for pediatric health abnormality feedback analysis to obtain the feedback status of the health abnormality signal corresponding to the pediatric patient.

[0129] In the embodiments of the present invention, the data collection terminal in the ward is equipped with an edge computing module. The edge computing module runs an operating system based on Linux and installs specialized data processing and analysis software. When receiving the pediatric nerve data corresponding to various intelligent emergency devices after personalized management, the data processing software first preprocesses the data, such as operations like noise removal and filtering. For electroencephalogram (EEG) signals, a Butterworth low-pass filter is used to remove high-frequency noise; for heart rate signals, the moving average filtering method is used to eliminate abnormal fluctuations. Then, the pre-trained data analysis model is used to perform pediatric health anomaly feedback analysis on the data. For example, for heart rate signals, the model determines whether the heart rate is abnormal based on factors such as the normal heart rate range, the patient's age, and condition. If the heart rate continuously exceeds or is lower than the normal range by a certain threshold, it is determined as abnormal. Through the comprehensive analysis of EEG signals, heart rate signals, blood oxygen signals, and motion state signals, the health anomaly signal feedback status corresponding to the pediatric patient is obtained, such as "mild EEG abnormality, rapid heart rate, normal blood oxygen, and more active motion state than before", and this status information is organized into a specific data format and prepared to be uploaded to the cloud platform.

[0130] Preferably, the edge computing node is used to transmit the health anomaly signal feedback status corresponding to the pediatric patient to the cloud platform in real time with reduced corresponding transmission delay;

[0131] In the embodiments of the present invention, the edge computing node is connected to the data collection terminal in the ward through a high-speed Ethernet network. When receiving the health anomaly signal feedback status data of the pediatric patient sent by the data collection terminal, the edge computing node uses its built-in network acceleration module to compress and optimize the data to reduce the transmission delay. For example, the GZIP compression algorithm is used to compress the data to reduce the data volume. Then, the compressed data is transmitted to the cloud platform in real time through the 5G network. The edge computing node establishes a secure encrypted connection with the cloud platform and uses the SSL / TLS encryption protocol to ensure the security during the data transmission process. During the data transmission process, the edge computing node monitors the transmission status in real time. If situations such as network interruption or data loss occur, it automatically retransmits to ensure that the data reaches the cloud platform completely.

[0132] Preferably, on the cloud platform, the health anomaly signal feedback status corresponding to the pediatric patient is used to perform abnormal feedback optimization adjustment on the corresponding pediatric patient personalized management recommendation plan, so as to adjust the corresponding drug types and drug doses according to the abnormal fluctuation values of each EEG, heart rate, blood oxygen, and motion state sub-signals in the health anomaly signal feedback status, and generate the corresponding pediatric personalized optimization adjustment management plan.

[0133] In the embodiments of the present invention, in the cloud platform, the elastic computing service (ECS) and big data analysis service (MaxCompute) of Alibaba Cloud are used to process the feedback status of health abnormal signals corresponding to pediatric patients. After receiving the data transmitted by the edge computing node, first, the data cleaning function of MaxCompute is used to further clean and verify the data to ensure the accuracy and integrity of the data. Then, through the pre-developed data analysis algorithm, the abnormal fluctuation values corresponding to each electroencephalogram, heart rate, blood oxygen, and motion state sub-signal in the feedback status of health abnormal signals are analyzed. For example, if the energy of a certain frequency band in the electroencephalogram abnormally increases and the heart rate is too fast, combined with medical knowledge and historical case data, it is judged that the drug type or dosage may need to be adjusted. By calling the drug knowledge base and treatment plan database stored on the cloud platform, the corresponding drug type and drug dosage adjustment plan are calculated according to the abnormal fluctuation value. For example, the original drug A is replaced with drug B, and the dosage is adjusted to 6 mg / kg, twice a day, to generate the corresponding personalized optimization adjustment management plan for pediatrics. Finally, the optimization adjustment management plan is fed back to the terminal devices of medical staff through the hospital information management system (HIS) so that the medical staff can adjust the treatment plan in a timely manner.

[0134] Further, the uploading of the pediatric nerve data corresponding to the personalized management to the edge computing node for pediatric health abnormal feedback analysis includes:

[0135] Uploading the pediatric nerve data corresponding to the personalized management to the edge computing node;

[0136] In the embodiments of the present invention, a customized device with Bluetooth and Wi-Fi communication functions is used as the data collection terminal in the ward, and an operating system deeply optimized based on the Android system is run. When various intelligent emergency devices, such as an electroencephalogram monitor, a heart rate monitor, a blood oxygen sensor supporting a portable multi-parameter monitor, and a triaxial accelerometer motion sensor, transmit the pediatric nerve data corresponding to the personalized management to the data collection terminal through Bluetooth or Wi-Fi, the data collection terminal establishes a connection with the edge computing node at a transmission rate of 1000 Mbps through the built-in Ethernet module. Before data transmission, the data collection terminal preliminarily packages the data, and different types of pediatric nerve data, such as electroencephalogram signal data, heart rate signal data, blood oxygen signal data, and motion state signal data, are respectively encapsulated according to specific data formats, and a data identification header is added to indicate information such as data type, patient ID, and collection time. Then, the packaged data is sent to the edge computing node through the Ethernet to ensure the accuracy and integrity of data transmission.

[0137] Preferably, an edge computing node is used to perform statistical analysis on the fluctuation amplitude and frequency of each corresponding sub-signal in pediatric nerve data, so as to obtain the fluctuation amplitude and fluctuation frequency corresponding to the pediatric nerve sub-signal;

[0138] In the embodiment of the present invention, by running a data analysis program developed based on the Linux system on the edge computing node, this program uses the NumPy library and SciPy library of Python to perform statistical analysis on the fluctuation amplitude and frequency of each corresponding sub-signal in pediatric nerve data. Taking the electroencephalogram signal as an example, it is assumed that the received electroencephalogram signal data is stored in a one-dimensional NumPy array eeg_data, and the data sampling frequency is 1000 Hz. First, by calculating the difference between the maximum value and the minimum value in the array, the fluctuation amplitude of the electroencephalogram signal is obtained, that is, eeg_amplitude = np.max(eeg_data) - np.min(eeg_data). For frequency statistical analysis, the scipy.fftpack.fft() function is used to perform a fast Fourier transform on eeg_data to obtain the representation of the signal in the frequency domain. Then, by calculating the peak position of the energy distribution in the frequency domain, the main frequency components are determined. For example, frequencies = np.fft.fftfreq(len(eeg_data), 1 / 1000), and the frequency value corresponding to the maximum energy is found as the main fluctuation frequency eeg_frequency = frequencies[np.argmax(np.abs(np.fft.fft(eeg_data)))]. Similar methods are used to perform statistical analysis on the fluctuation amplitude and frequency of the heart rate signal, blood oxygen signal, and motion state signal, respectively, to obtain the corresponding fluctuation amplitude and fluctuation frequency.

[0139] Preferably, the signal distribution mean value and signal distribution standard deviation corresponding to a period of time are obtained through each corresponding sub-signal in the pediatric nerve data, and based on the signal distribution mean value and signal distribution standard deviation corresponding to a period of time, an abnormal fluctuation calculation is performed on the fluctuation amplitude and fluctuation frequency corresponding to the pediatric nerve sub-signal by using the pediatric signal abnormality calculation formula, so as to obtain the abnormal fluctuation value corresponding to the pediatric nerve sub-signal;

[0140] In an embodiment of the present invention, on the edge computing node, the data of each sub-signal corresponding in pediatric nerve data within a specific time period (assumed to be 1 minute, corresponding to 60,000 sampling points, sampling frequency of 1000 Hz) is processed by using the NumPy library of Python to obtain the mean value of signal distribution and the standard deviation of signal distribution. For the electroencephalogram signal, assuming that the electroencephalogram signal data within 1 minute is stored in the eeg_data_60s array, the mean value of signal distribution eeg_mean is calculated by np.mean(eeg_data_60s), and the standard deviation of signal distribution eeg_std is calculated by np.std(eeg_data_60s). The pediatric signal abnormality calculation formula is set as In the formula, Δε is the abnormal fluctuation value, T is the time period interval, t is the time variable parameter, X(t) is the specific value corresponding to the pediatric nerve sub-signal at time t, μ is the mean value of signal distribution corresponding within the time period T, σ is the standard deviation of signal distribution corresponding within the time period T, ω is the fluctuation amplitude, f is the fluctuation frequency. Such abnormal fluctuation calculations are performed on the heart rate signal, blood oxygen signal, and motion state signal. Finally, the abnormal fluctuation values corresponding to each pediatric nerve sub-signal are obtained. In addition, the pediatric signal abnormality calculation formula can also use any abnormal detection algorithm in the field to replace the process of abnormal fluctuation calculation. For example, the abnormal fluctuation value = (actual fluctuation amplitude - mean fluctuation amplitude) / standard deviation fluctuation amplitude + (actual fluctuation frequency - mean fluctuation frequency) / standard deviation fluctuation frequency, where the mean fluctuation amplitude and mean fluctuation frequency are obtained by statistics of a large amount of normal pediatric nerve data within the same 1-minute time period, and the standard deviation fluctuation amplitude and standard deviation fluctuation frequency are also obtained in the same way. Taking the electroencephalogram signal as an example, assuming that the mean fluctuation amplitude is mean_amplitude, the standard deviation fluctuation amplitude is std_amplitude, the mean fluctuation frequency is mean_frequency, and the standard deviation fluctuation frequency is std_frequency, using the actual fluctuation amplitude eeg_amplitude and actual fluctuation frequency eeg_frequency calculated above, the abnormal fluctuation value of the electroencephalogram signal eeg_abnormal_value = (eeg_amplitude - mean_amplitude) / std_amplitude + (eeg_frequency - mean_frequency) / std_frequency, which is not limited to this pediatric signal abnormality calculation formula.

[0141] Preferably, based on the abnormal fluctuation values corresponding to the pediatric nerve sub-signals, pediatric health abnormality feedback analysis is performed on each sub-signal corresponding in the pediatric nerve data within the same time period to obtain the health abnormality signal feedback state corresponding to the pediatric patient.

[0142] In an embodiment of the present invention, on an edge computing node, using a pre-written data analysis script, pediatric health abnormality feedback analysis is performed on each sub-signal corresponding to pediatric nerve data within the same time period based on the abnormal fluctuation value corresponding to the pediatric nerve sub-signal. The script sets the health abnormality status corresponding to different abnormal fluctuation value ranges. For example, for an electroencephalogram (EEG) signal, if the abnormal fluctuation value is greater than 3, it is determined as "significant EEG abnormality"; if it is between 1 and 3, it is determined as "mild EEG abnormality". For a heart rate signal, if the abnormal fluctuation value is less than -2, it is determined as "bradycardia abnormality"; if it is greater than 2, it is determined as "tachycardia abnormality". For a blood oxygen signal, assuming the normal blood oxygen saturation range is 95%-100%, by analyzing the data of the blood oxygen signal within a specific time period (such as 1 minute), calculating its mean value and standard deviation, if the blood oxygen saturation mean value is lower than 95% and the abnormal fluctuation value (obtained according to the previous abnormal calculation formula) is greater than 1, it is determined as "low blood oxygen saturation abnormality"; if the mean value is higher than 100% and the abnormal fluctuation value is greater than 1, it is determined as "high blood oxygen saturation abnormality". For a motion state signal, taking the data collected by an accelerometer as an example, assuming that within 1 minute, the normal motion state signal has a certain range of fluctuation amplitudes in each axis. By statistically analyzing historical normal data, the mean fluctuation amplitude and standard deviation fluctuation amplitude of each axis are obtained, and the actual fluctuation amplitude of each axis of the current motion state signal is calculated. According to the abnormal calculation formula, the abnormal fluctuation value is obtained. If on a certain axis (such as the x-axis), the abnormal fluctuation value is greater than 2, it is determined as "abnormal motion state in the x-axis direction". Combining the determination results of each axis, if there are abnormal fluctuation values greater than 2 in multiple axes, it is determined as "overall abnormal motion state". Combining the abnormal determination results of the EEG signal, heart rate signal, blood oxygen signal, and motion state signal, the health abnormality signal feedback status corresponding to the pediatric patient is obtained. Such as "mild EEG abnormality, tachycardia abnormality, normal blood oxygen, basically normal motion state", and this health abnormality signal feedback status is organized into a specific data format, including information such as patient ID, abnormal determination results of each sub-signal, and corresponding abnormal fluctuation values, and is prepared to be transmitted to the subsequent processing link.

[0143] Further, the specific formula for calculating the pediatric signal abnormality is as follows:

[0144]

[0145] In the formula, Δε is the abnormal fluctuation value, T is the time period interval, t is the time variable parameter, X(t) is the specific value corresponding to the pediatric nerve sub-signal at time t, μ is the signal distribution mean value corresponding to the time period T, σ is the signal distribution standard deviation corresponding to the time period T, ω is the fluctuation amplitude, and f is the fluctuation frequency.

[0146] The present invention obtains a pediatric signal abnormality calculation formula by using a specific mathematical model and verifying it, which is used to calculate the abnormal fluctuation of the fluctuation amplitude and fluctuation frequency corresponding to the pediatric nerve sub-signal. The formula fully considers the abnormal fluctuation value Δε, the time interval T, the time variable parameter t, the specific value X(t) corresponding to the pediatric nerve sub-signal at time t, the signal distribution mean μ corresponding to the time period T, the signal distribution standard deviation σ corresponding to the time period T, the fluctuation amplitude ω, and the fluctuation frequency f. According to the mutual correlation between the abnormal fluctuation value Δε and the above parameters, a functional relationship is formed:

[0147]

[0148] This formula can realize the abnormal fluctuation calculation process of the fluctuation amplitude and fluctuation frequency corresponding to the pediatric nerve sub-signal. At the same time, through in-depth analysis of the fluctuation amplitude and frequency of the pediatric nerve sub-signal, it can effectively capture the abnormal fluctuation of the signal. Using the mean and standard deviation of the signal distribution as a benchmark, the degree of deviation from the normal fluctuation pattern is calculated. This method can accurately identify the abnormal fluctuation of the patient's nerve signal. The integral part of the formula enables the calculation to dynamically evaluate the abnormal fluctuation value of the signal according to time changes. Considering the fluctuation changes of the signal in different time periods, this can monitor the changes of the nerve signals of pediatric patients in real time and identify abnormal fluctuations in time, providing dynamic data support for clinical diagnosis. The formula involves multiple variables (such as fluctuation amplitude, fluctuation frequency, signal value, etc.), and introduces the comprehensive influence of fluctuation amplitude and frequency through the exponential function. This method can more accurately simulate and predict the actual fluctuation of the signal, thereby helping to identify those subtle changes that are ignored by traditional methods. By calculating the multiple factors involved in the formula, the abnormal fluctuation value not only takes into account the distribution of the signal, but also integrates the changes in multiple dimensions such as fluctuation amplitude, frequency and time, making abnormal detection more comprehensive and detailed. This comprehensive abnormal fluctuation calculation helps to improve the analysis ability of complex neural signals, especially in pediatric neuropathological data, and can more effectively distinguish normal fluctuations from pathological fluctuations. The formula can respond highly sensitively to the slight fluctuations of pediatric neural signals, so it can identify potential health problems in advance. For example, abnormalities in children's nervous system often manifest as subtle fluctuations. The formula can keenly detect these changes, so as to provide timely feedback on possible health risks and assist doctors in making early interventions. Since the formula can perform comprehensive analysis based on different signal data (such as frequency, amplitude, standard deviation, etc.), the abnormal fluctuation value of pediatric neural data can be combined with other clinical data to form a multi-dimensional health monitoring system, which provides data support for personalized diagnosis and treatment and health management, and enhances doctors' comprehensive understanding of patients' health status.

[0149] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be embraced within the present invention.

[0150] The above description is only a specific implementation manner of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. An intelligent management system for pediatric neurological data based on artificial intelligence, characterized in that, It includes the following modules: The pediatric neural data acquisition module is used to collect pediatric neurophysiological data of pediatric patients in real time through corresponding intelligent hardware devices, and transmit it to the corresponding intelligent medical device of pediatric patients in real time through wireless technology; the collected pediatric neurophysiological data is uploaded to the cloud platform through the intelligent medical device; The pediatric multimodal fusion module is used to use the cloud platform and based on convolutional neural networks to correct abnormal noise in pediatric neurophysiological data to obtain pediatric neural abnormal correction data; Cross-data fusion is performed on each sub-data in the pediatric neural abnormal correction data based on a multimodal neural network to generate pediatric neural multimodal fusion data; The personalized management recommendation module is used to perform personalized recommendation based on pediatric neural multimodal fusion data and combined with reinforcement learning to generate a corresponding personalized management recommendation plan for pediatric patients; The recommendation plan abnormal feedback optimization module is used to apply the personalized management recommendation plan for pediatric patients to the corresponding pediatric patients and use various intelligent emergency devices to monitor and collect the corresponding pediatric neural data after personalized management, and use edge computing nodes to perform pediatric health abnormal feedback analysis on the corresponding pediatric neural data to obtain the health abnormal signal feedback status of the corresponding pediatric patients; based on the health abnormal signal feedback status of the corresponding pediatric patients, the cloud platform performs abnormal feedback optimization adjustment on the corresponding personalized management recommendation plan for pediatric patients to generate a corresponding pediatric personalized optimization adjustment management plan.

2. The intelligent pediatric nerve data management system based on artificial intelligence according to claim 1, wherein The pediatric neural data acquisition module includes the following functions: Real-time collection of electroencephalogram signal data of pediatric patients through an electroencephalogram monitor; Real-time collection of heart rate signal data of pediatric patients through a heart rate monitor; Real-time collection of respiratory oxygen saturation data of pediatric patients through a blood oxygen sensor; Real-time collection of neural movement state data of pediatric patients through a motion sensor; Synthesize the electroencephalogram signal data, heart rate signal data, respiratory oxygen saturation data, and neural movement state data of pediatric patients into a data set to obtain corresponding pediatric neurophysiological data, and transmit it to the corresponding intelligent medical device worn by pediatric patients in real time through wireless technologies such as Bluetooth or Wi-Fi; Upload the collected pediatric neurophysiological data to the cloud platform through the intelligent medical device.

3. The intelligent pediatric nerve data management system based on artificial intelligence according to claim 2, characterized in that, The pediatric multimodal fusion module includes the following functions: Use the cloud platform to perform data feature input conversion on the corresponding electroencephalogram, heart rate, blood oxygen saturation, and motion state data in the pediatric neurophysiological data through short-time Fourier transform to convert the corresponding data into a feature format suitable for deep learning input, and obtain pediatric neural time series data features; Obtain the labeled pediatric neural original data, including pediatric neural data labeled with normal and abnormal noise; Train a corresponding convolutional neural network model using the labeled pediatric neuron data so that it can identify the corresponding normal and abnormal noise points in the pediatric neuron data, and input the pediatric neuron time-series data features into the trained convolutional neural network model for abnormal noise point detection and classification to obtain pediatric neuron normal data and pediatric neuron abnormal noise data; Based on the pediatric neuron normal data and the pediatric neuron abnormal noise data, perform abnormal noise correction at the corresponding abnormal noise positions in the pediatric neuron physiological data to obtain pediatric neuron abnormal correction data; Based on the multi-modal neural network, perform cross-data fusion on each sub-data in the pediatric neuron abnormal correction data to generate pediatric neuron multi-modal fusion data.

4. The intelligent pediatric nerve data management system based on artificial intelligence according to claim 3, characterized in that The performing abnormal noise correction at the corresponding abnormal noise positions in the pediatric neuron physiological data based on the pediatric neuron normal data and the pediatric neuron abnormal noise data includes: Perform statistical analysis on the data distribution of the pediatric neuron abnormal noise data to obtain the mean value of the pediatric neuron noise data distribution and the standard deviation of the pediatric neuron noise data distribution; Based on the mean value of the pediatric neuron noise data distribution and the standard deviation of the pediatric neuron noise data distribution, and combined with linear fitting interpolation, perform noise fitting correction on the corresponding normal data distribution in the pediatric neuron normal data, so as to generate the abnormal correction value corresponding to the abnormal noise by interpolating and fitting the corresponding normal data distribution according to the mean value of the pediatric neuron noise data distribution and the standard deviation of the pediatric neuron noise data distribution, and obtain the pediatric neuron noise abnormal correction value; Based on the pediatric neuron noise abnormal correction value, perform abnormal noise correction at the corresponding abnormal noise positions in the pediatric neuron physiological data to obtain pediatric neuron abnormal correction data.

5. The intelligent pediatric nerve data management system based on artificial intelligence according to claim 3, characterized in that, The performing cross-data fusion on each sub-data in the pediatric neuron abnormal correction data based on the multi-modal neural network includes: Perform data feature analysis on each sub-data in the pediatric data abnormal correction data to realize time-frequency analysis of the electroencephalogram signal to obtain the corresponding time-frequency characteristics of the electroencephalogram signal, perform periodic analysis of the heart rate signal to obtain the corresponding periodic characteristics of the heart rate signal, perform statistical analysis on the data distribution of blood oxygen saturation and nerve movement state to obtain the corresponding data distribution characteristics of blood oxygen and movement state, and obtain the sub-modal characteristics corresponding to each pediatric sub-data type; Use each network branch of the multi-modal neural network to process the sub-modal characteristics corresponding to the pediatric sub-data type, and use the attention mechanism to weight the importance weights of each sub-modal characteristic. At the same time, based on the importance weights, perform cross-data fusion on the sub-modal characteristics corresponding to each pediatric sub-data type to generate pediatric neuron multi-modal fusion data.

6. The intelligent pediatric nerve data management system based on artificial intelligence according to claim 1, characterized in that The personalized management recommendation module includes the following functions: Obtain the feature abnormal correlation between each pediatric data sub-modal of the pediatric patient through the pediatric neuron multi-modal fusion data; Based on the feature anomaly correlation between various pediatric data sub-modalities of pediatric patients, combined with reinforcement learning and corresponding pediatric data sub-modalities modeling, a corresponding treatment management recommendation reinforcement learning model is generated. The corresponding pediatric data sub-modalities are used as state vectors, and corresponding pediatric treatment management plans are recommended according to the original treatment management database. At the same time, a reward function is designed according to the corresponding effect actions to continuously reinforce the learning of the corresponding treatment management plan, so as to generate a corresponding personalized management recommendation plan for pediatric patients.

7. The intelligent management system for pediatric neural data based on artificial intelligence according to claim 6, characterized in that The personalized management recommendation plan for pediatric patients includes the types and doses of drugs corresponding to the pediatric neurological data of pediatric patients in the clinic.

8. The intelligent pediatric nerve data management system based on artificial intelligence according to claim 1, characterized in that, The recommended plan anomaly feedback optimization module includes the following functions: Provide decision support for the personalized management recommendation plan for pediatric patients and apply it to the corresponding pediatric patients. At the same time, use various intelligent emergency devices to monitor and collect the corresponding pediatric neurological data after personalized management, including the electroencephalogram signal, heart rate signal, blood oxygen signal, and motion state signal corresponding to pediatric patients; Upload the corresponding pediatric neurological data after personalized management to the edge computing node for pediatric health anomaly feedback analysis to obtain the health anomaly signal feedback status corresponding to pediatric patients; Use the edge computing node to transmit the health anomaly signal feedback status corresponding to pediatric patients to the cloud platform in real time to reduce the corresponding transmission delay; On the cloud platform, use the health anomaly signal feedback status corresponding to pediatric patients to perform anomaly feedback optimization adjustment on the corresponding personalized management recommendation plan for pediatric patients, and adjust the corresponding types and doses of drugs according to the abnormal fluctuation values of each electroencephalogram, heart rate, blood oxygen, and motion state sub-signals in the health anomaly signal feedback status, so as to generate a corresponding personalized optimization adjustment management plan for pediatrics.

9. The intelligent pediatric nerve data management system based on artificial intelligence according to claim 8, characterized in that The uploading of the corresponding pediatric neurological data after personalized management to the edge computing node for pediatric health anomaly feedback analysis includes: Upload the corresponding pediatric neurological data after personalized management to the edge computing node; Use the edge computing node to perform statistical analysis on the fluctuation amplitude and frequency of each sub-signal in the pediatric neurological data to obtain the fluctuation amplitude and fluctuation frequency corresponding to the pediatric neurological sub-signal; Obtain the signal distribution mean and signal distribution standard deviation corresponding to each sub-signal in a period through the corresponding sub-signals in the pediatric neurological data, and use the pediatric signal anomaly calculation formula based on the signal distribution mean and signal distribution standard deviation corresponding to the period to perform abnormal fluctuation calculation on the fluctuation amplitude and fluctuation frequency corresponding to the pediatric neurological sub-signal, so as to obtain the abnormal fluctuation value corresponding to the pediatric neurological sub-signal; Based on the abnormal fluctuation value corresponding to the pediatric neurological sub-signal, perform pediatric health anomaly feedback analysis on each sub-signal in the pediatric neurological data in the same period to obtain the health anomaly signal feedback status corresponding to pediatric patients.

10. The intelligent pediatric neural data management system based on artificial intelligence according to claim 9, characterized in that, The specific formula for pediatric signal anomaly calculation is: Where Δε is the abnormal fluctuation value, T is the time period interval, t is the time variable parameter, X(t) is the specific value corresponding to the pediatric nerve sub-signal at time t, μ is the mean value of the signal distribution corresponding to the time period T, σ is the standard deviation of the signal distribution corresponding to the time period T, ω is the fluctuation amplitude, and f is the fluctuation frequency.

Images