Neural feedback modulation method and system for rehabilitation training of patients with brain disorders

Abstract

The present application relates to the technical field of rehabilitation training, and particularly relates to a neural feedback modulation method and system for rehabilitation training of brain disorder patients, a generative adversarial network is used to provide a trained translation model based on real EEG sequences and real EMG sequences of healthy people, the translation model can provide accurate mapping relationship between the EEG sequences and the EMG sequences, the EEG sequences with motor intention of the brain disorder patients are translated into EMG sequences by using the translation model, and difference analysis is performed on the real EMG sequences of the patients, and then fine modulation of multi-modal neural feedback is realized, the availability of the rehabilitation system is improved, and the workload of medical staff is greatly reduced. And through fine-grained neural feedback modulation, the brain disorder patients can gradually recover the control of the limbs, and the self-confidence of the patients in rehabilitation training is improved.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation training technology, and in particular to a neurofeedback modulation method and system for rehabilitation training of patients with brain disorders. Background Technology

[0002] Brain disorders refer to diseases that cause motor dysfunction and other symptoms due to brain damage resulting from stroke, cerebral palsy, traumatic brain injury, etc. Among these, stroke is a leading cause of death and disability, characterized by high incidence, high disability rate, high mortality rate, and high recurrence rate. Although rehabilitation treatment is available, approximately 60%-80% of stroke patients still suffer from significant motor dysfunction, imposing a heavy care cost on families and society. Therefore, high-quality and efficient rehabilitation training is the most important means to solve the current problem and an inevitable choice to help patients regain independence and reintegrate into society.

[0003] Studies have shown that stroke patients primarily suffer from damage to neural circuits. Therefore, rehabilitation assessment and training should comprehensively consider information on brain function to assist physicians in providing adaptive rehabilitation training methods for patients. Thus, rehabilitation training programs that integrate motor and brain function are an important trend in future rehabilitation training. However, current traditional rehabilitation systems often face challenges in collecting training data due to the high cost of data acquisition equipment, long collection times, and privacy concerns. Consequently, they typically employ coarse-grained control models, failing to achieve optimal rehabilitation training effects and assessment levels. Summary of the Invention

[0004] To address the aforementioned problems, this invention discloses a neural feedback modulation method and system for rehabilitation training of patients with brain disorders, in order to solve the problem in the prior art that it is difficult to finely modulate the neural feedback signal from the motor intention to the execution of the action of patients with brain disorders during rehabilitation training.

[0005] On one hand, this invention discloses a neurofeedback modulation method for rehabilitation training of patients with brain disorders, comprising:

[0006] Simultaneously collect brain-myograph data from healthy individuals during simulated rehabilitation training for patients with brain disorders, in order to obtain real EEG (electroencephalography) sequences and real EMG (electromyography) sequences from healthy individuals;

[0007] Generative adversarial networks (GANs) are used to provide a trained translation model based on the real EEG and EMG sequences of the healthy individuals. The trained translation model provides a mapping relationship between the EEG and EMG sequences.

[0008] Real-time electroencephalogram (EEG) and electromyogram (EMG) data of patients with brain disorders are collected simultaneously to obtain their true EEG and EMG sequences. After mapping the true EEG sequences of patients with brain disorders to ideal EMG sequences using the translation model, the difference between the ideal EMG sequences and the synchronously collected true EMG sequences of patients with brain disorders is calculated. Based on the calculation results, the fine parameters of the neural feedback signals in the rehabilitation training of patients with brain disorders are modulated in real time.

[0009] In some embodiments, the generative adversarial network includes a generator network and a discriminator network; the generator network includes a generator model and a translation model, and the discriminator network includes a first discriminator model and a second discriminator model.

[0010] In some of these embodiments, during the process of providing a trained translation model based on the real EEG and real EMG sequences of the healthy individuals, the generative model is used to generate an EEG sequence based on random noise, and the translation model is used to obtain a translated EMG sequence that maps to the generated EEG sequence.

[0011] The first discrimination model is used to distinguish between the real EEG sequence of a healthy person and the generated EEG sequence, and inputs the difference between the real EEG sequence and the generated EEG sequence as a first reward value into the generation model. The second discrimination model is used to distinguish between the real EMG sequence of a healthy person and the translated EMG sequence, and inputs the difference between the real EMG sequence and the translated EMG sequence as a second reward value into the generation model.

[0012] In some embodiments, the differences between the ideal EMG sequence and the real EMG sequence of a patient with a brain disorder acquired simultaneously include at least one of variance, power spectrum variation, and correlation.

[0013] In some embodiments, the fine parameters of the neural feedback signal include at least one of the duration of the FES, the stimulation frequency, and the current intensity.

[0014] In some of these embodiments, a brain-motor acquisition module is used to simultaneously acquire brain-motor data from healthy individuals or patients with brain disorders.

[0015] The brain-myogra (BEM) acquisition module includes an EEG acquisition unit and an EMG acquisition unit; the EEG acquisition unit includes an EEG acquisition subunit and a wireless transmission subunit, and the EEG acquisition subunit is connected to an EEG cap; the EMG acquisition unit includes an EMG acquisition subunit and a wireless transmission subunit, and the EMG acquisition subunit is connected to EMG electrodes; both the EEG acquisition subunit and the EMG acquisition subunit transmit data to a generative adversarial network through the wireless transmission subunit.

[0016] On the other hand, the present invention also discloses a neurofeedback modulation system for rehabilitation training of patients with brain disorders, wherein the system includes: a first brain-myomyography acquisition module, a second brain-myomyography acquisition module, a generative adversarial network module, a computational processing module, and a modulation module;

[0017] The first brain-muscle electroacupuncture acquisition module is used to synchronously acquire brain-muscle electroacupuncture data of healthy individuals during rehabilitation training simulating patients with brain disorders, so as to obtain the real EEG sequence and real EMG sequence of healthy individuals;

[0018] The second brain-myograph acquisition module is used to simultaneously acquire real-time brain-myograph data of patients with brain disorders to obtain the real EEG and real EMG sequences of patients with brain disorders;

[0019] The generative adversarial network is communicatively connected to the first EEG-EMG acquisition module to provide a trained translation model based on real EEG and EMG sequences of healthy individuals. The trained translation model provides a mapping relationship between EEG and EMG sequences.

[0020] The computational processing module is communicatively connected to the generative adversarial network, the second EEG acquisition module, and the modulation module, respectively, so as to calculate the difference between the ideal EMG sequence and the synchronously acquired real EMG sequence of the brain disorder patient after mapping the real EEG sequence of the brain disorder patient to the ideal EMG sequence using the trained translation model, and transmit the calculation result to the modulation module.

[0021] The modulation module is used to modulate the fine parameters of the neural feedback signals in the rehabilitation training of patients with brain disorders in real time.

[0022] In some embodiments, the generative adversarial network includes a generator network and a discriminator network; the generator network includes a generator model and a translation model, and the discriminator network includes a first discriminator model and a second discriminator model.

[0023] In some embodiments, the generative model is used to generate an EEG sequence based on random noise, and the translation model is used to obtain a translated EMG sequence that maps to the generated EEG sequence.

[0024] The first discrimination model is used to distinguish between the real EEG sequence of a healthy person and the generated EEG sequence, and inputs the difference between the real EEG sequence and the generated EEG sequence as a first reward value into the generation model. The second discrimination model is used to distinguish between the real EMG sequence of a healthy person and the translated EMG sequence, and inputs the difference between the real EMG sequence and the translated EMG sequence as a second reward value into the generation model.

[0025] In some embodiments, the first and second EEG acquisition modules include an EEG acquisition unit and an EMG acquisition unit; the EEG acquisition unit includes an EEG acquisition subunit and a wireless transmission subunit, and the EEG acquisition subunit is connected to an EEG cap; the EMG acquisition unit includes an EMG acquisition subunit and a wireless transmission subunit, and the EMG acquisition subunit is connected to EMG electrodes; both the EEG acquisition subunit and the EMG acquisition subunit transmit data to a generative adversarial network through the wireless transmission subunit.

[0026] Compared with the prior art, the above invention has at least one of the following advantages or beneficial effects:

[0027] 1) A generative adversarial network (GAN) is used to provide a trained translation model based on real EEG and EMG sequences from healthy individuals. This translation model can provide an accurate mapping relationship between EEG and EMG sequences. Reinforcement-adversarial learning (incorporating reinforcement learning concepts) is performed on this GAN to augment the EEG data of healthy individuals. This allows the model to learn the data distribution from limited EEG data and synthesize a large amount of realistic augmented data with true neural properties. This overcomes the problems of limited training data and difficulty in obtaining data due to expensive acquisition equipment, long acquisition time, or privacy issues. Thus, the training performance of the rehabilitation system can be significantly improved by using a small amount of training data.

[0028] Second, by using a trained translation model to translate the EEG sequences of patients with motor intentions in brain disorders into EMG sequences, and performing differential analysis with the patients' actual EMG sequences, fine modulation of multimodal neurofeedback can be achieved. This can change the coarse-grained control mode of traditional rehabilitation medical systems, improve the usability of rehabilitation systems, and significantly reduce the workload of medical staff.

[0029] Third, through fine-grained neurofeedback modulation, it can help patients with brain disorders gradually regain control of their limbs, improve their confidence in rehabilitation training, and when patients with brain disorders use the product for rehabilitation, this method can be used to assess their level of motor rehabilitation. Attached Figure Description

[0030] The invention, its features, shape, and advantages will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. Like reference numerals denote like parts throughout the drawings. The drawings are not drawn to scale; their focus is on illustrating the gist of the invention.

[0031] Figure 1 This is a flowchart of a neurofeedback modulation method for rehabilitation training of patients with brain disorders, as described in an embodiment of the present invention.

[0032] Figure 2 for Figure 1 A schematic diagram of the flowchart corresponding to step S2 in the neurofeedback modulation method used in rehabilitation training for patients with brain disorders.

[0033] Figure 3 for Figure 1 A schematic diagram of the flowchart corresponding to step S3 in the neurofeedback modulation method used in rehabilitation training for patients with brain disorders. Detailed Implementation

[0034] This invention addresses the problem of finely modulating the neural feedback signal from a patient's motor intention to the execution of the action in clinical limb function rehabilitation medical systems. Based on reinforcement-adversarial learning, it achieves a dynamic neural feedback modulation mechanism with high matching degree to motor intention, aiming to construct an efficient rehabilitation training mechanism and improve the efficiency of rehabilitation training for patients with brain disorders.

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the present invention.

[0036] like Figure 1 As shown, this invention discloses a neurofeedback modulation method for rehabilitation training of patients with brain disorders. The patients with brain disorders referred to in this invention are those who suffer from motor dysfunction and other symptoms due to brain injury caused by stroke, cerebral palsy, traumatic brain injury, etc., and require rehabilitation training. Specifically, the neurofeedback modulation method includes:

[0037] Step S1: Simultaneously collect brain-myograph data from healthy individuals during simulated rehabilitation training for patients with brain disorders, in order to obtain real EEG (electroencephalography) sequences and real EMG (electromyography) sequences from healthy individuals. That is, the brain-myograph data contains neurofeedback information from EEG and EMG.

[0038] Specifically, a brain-myograph (BMM) acquisition module is used to simultaneously acquire brain-myograph data from healthy individuals simulating rehabilitation training for patients with brain disorders. This BMM acquisition module includes an EEG acquisition unit and a myograph acquisition unit. The EEG acquisition unit includes an EEG acquisition subunit and a wireless transmission subunit, with the EEG acquisition subunit connected to an EEG cap. The myograph acquisition unit includes a myograph acquisition subunit and a wireless transmission subunit, with the myograph acquisition subunit connected to myograph electrodes. Both the EEG acquisition subunit and the myograph acquisition subunit transmit data to a generative adversarial network (GAN) via the wireless transmission subunit. Of course, in other embodiments, the aforementioned myograph acquisition module can also acquire data via wired transmission to the GAN, which does not affect the purpose of this invention.

[0039] Step S2, as follows Figure 2 As shown, a translation model trained on real EEG and EMG sequences from healthy individuals is provided using generative adversarial networks (a zero-sum game-based approach). And the trained translation model This trained translation model provides an accurate mapping between EEG and EMG sequences. It enables the mapping between EEG and EMG sequences of healthy individuals, and can obtain the characteristic patterns that should be exhibited on the EMG sequence based on the movement intention reflected in the EEG sequence.

[0040] It should be noted here that, in order to provide translation models This embodiment provides sufficient training data and incorporates reinforcement learning concepts into generative adversarial networks to generate more realistic data, thereby helping to... Achieve accurate mapping between EEG sequences and EMG sequences.

[0041] In one specific embodiment of the present invention, please continue to refer to... Figure 2 As shown, the aforementioned generative adversarial network includes a generator network and a discriminator network; the generator network includes a generator model. and translation model The discriminant network includes a first discriminant model. Second discriminant model A trained translation model is provided based on real EEG and EMG sequences from healthy individuals. During the process, generative model Used to generate EEG sequences from random noise, translation model The EMG sequence (i.e., electromyographic data) used to obtain the translation mapped to the generated EEG sequence is used in this translation model. It can also be viewed as another generative model); the first discriminative model This is used to distinguish between real EEG sequences from healthy individuals and generated EEG sequences, and the difference between the real and generated EEG sequences is used as the first reward value input into the generative model. The second discriminant model This is used to distinguish between real EMG sequences from healthy individuals and translated EMG sequences, and the difference between the real and translated EMG sequences is used as a second reward value input into the generative model. The generative adversarial network incorporates reinforcement learning concepts (specifically, a first reward value and a second reward value, i.e., spatiotemporally coupled brain-muscle reward), enabling the generative model to... Generates more realistic data. In summary, this generative adversarial network can simultaneously generate EEG sequences and corresponding EMG sequences. Subsequently, the generated brain-muscle signals and real brain-muscle signals are respectively fed into the first discriminant model. Second discriminant model The model differentiates between real and generated EEG sequences and incorporates reinforcement learning during training, using the difference between real and generated EEG sequences as a reward mechanism to guide the generative model. and translation model The model is then trained. After sufficient adversarial training, a generative model is generated. and translation model Compared with the first discriminant model Second discriminant model Eventually, equilibrium is reached. At this point, the generated EEG sequence has captured the distribution of the real data, and the translation model... While maintaining brain-muscle electrocoupling, it can achieve mapping to the ideal EMG based on the input EEG.

[0042] Step S3, as follows Figure 3 As shown, in step S2 above, a translation model capable of accurately mapping EEG sequences to ideal EMG sequences has been obtained. Based on this, in the rehabilitation training of patients with brain disorders (such as stroke patients), real-time electroencephalogram (EEG) and electromyography (EMG) data of patients with brain disorders are collected simultaneously to obtain the real EEG and EMG sequences of patients with brain disorders, and then a translation model is used. After mapping the real EEG sequences of patients with brain disorders to ideal EMG sequences, the differences between the ideal EMG sequences and the real EMG sequences of patients with brain disorders acquired simultaneously are calculated (i.e., sequence difference analysis is performed). Based on the calculation results, the fine parameters of the neural feedback signals in the rehabilitation training of patients with brain disorders are modulated in real time to match the patient's motor intentions and rehabilitation level.

[0043] In a preferred embodiment of the present invention, the differences between the ideal EMG sequence and the real EMG sequence of a patient with brain disorder acquired simultaneously include at least one of variance, power spectrum variation, and correlation.

[0044] In a preferred embodiment of the present invention, the fine parameters of the above-mentioned neural feedback signal include at least one of the duration of FES, stimulation frequency and current intensity.

[0045] In a preferred embodiment of the present invention, a brain-myograph (BMM) acquisition module is used to simultaneously acquire real-time BMM data from patients with brain disorders to obtain their actual EEG and EMG sequences. The BMM acquisition module includes an EEG acquisition unit and an EMG acquisition unit; the EEG acquisition unit includes an EEG acquisition subunit and a wireless transmission subunit, and the EEG acquisition subunit is connected to an EEG cap; the EMG acquisition unit includes an EMG acquisition subunit and a wireless transmission subunit, and the EMG acquisition subunit is connected to EMG electrodes; both the EEG acquisition subunit and the EMG acquisition subunit transmit data to a generative adversarial network via the wireless transmission subunit.

[0046] Furthermore, this invention also discloses a neurofeedback modulation system for rehabilitation training of patients with brain disorders. Specifically, the system includes: a first brain-muscle electromyography (BEM) acquisition module, a second BEM acquisition module, a generative adversarial network (GAN) module, a computational processing module, and a modulation module. The first BEM acquisition module is used to synchronously acquire BEM data from healthy individuals simulating rehabilitation training of patients with brain disorders to obtain real EEG and EMG sequences from healthy individuals. The second BEM acquisition module is used to synchronously acquire real-time BEM data from patients with brain disorders to obtain real EEG and EMG sequences from patients with brain disorders. The GAN is communicatively connected to the first BEM acquisition module to provide a trained translation model based on the real EEG and EMG sequences from healthy individuals. The trained translation model provides a mapping relationship between the EEG and EMG sequences. This computational processing module is communicatively connected to a generative adversarial network, a second EEG acquisition module, and a modulation module. After mapping the real EEG sequences of patients with brain disorders to ideal EMG sequences using a trained translation model, it calculates the differences between the ideal EMG sequences and the synchronously acquired real EMG sequences of the patients with brain disorders, and transmits the calculation results to the modulation module. This modulation module is used to modulate the fine parameters of the neural feedback signals during rehabilitation training for patients with brain disorders in real time.

[0047] In a specific embodiment of the present invention, the aforementioned generative adversarial network includes a generator network and a discriminator network; the generator network includes a generator model and a translation model, and the discriminator network includes a first discriminator model and a second discriminator model. The generator model is used to generate an EEG sequence based on random noise, and the translation model is used to obtain a translated EMG sequence mapped to the generated EEG sequence based on the generated EEG sequence; the first discriminator model is used to distinguish between the real EEG sequence of a healthy person and the generated EEG sequence, and inputs the difference between the real EEG sequence and the generated EEG sequence as a first reward value into the generator model; the second discriminator model is used to distinguish between the real EMG sequence of a healthy person and the translated EMG sequence, and inputs the difference between the real EMG sequence and the translated EMG sequence as a second reward value into the generator model.

[0048] In a specific embodiment of the present invention, the first brain-myogra acquisition module and the second brain-myogra acquisition module include an EEG acquisition unit and an EMG acquisition unit; the EEG acquisition unit includes an EEG acquisition subunit and a wireless transmission subunit, and the EEG acquisition subunit is connected to an EEG cap; the EMG acquisition unit includes an EMG acquisition subunit and a wireless transmission subunit, and the EMG acquisition subunit is connected to EMG electrodes; both the EEG acquisition subunit and the EMG acquisition subunit transmit data to a generative adversarial network through the wireless transmission subunit.

[0049] It is not difficult to see that this embodiment is a system embodiment corresponding to the embodiment of the neurofeedback modulation method in the rehabilitation training of patients with brain disorders described above. This embodiment can be implemented in conjunction with the embodiment of the neurofeedback modulation method in the rehabilitation training of patients with brain disorders described above. The relevant technical details mentioned in the embodiment of the neurofeedback modulation method in the rehabilitation training of patients with brain disorders described above are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the embodiment of the neurofeedback modulation method in the rehabilitation training of patients with brain disorders described above.

[0050] In summary, the neurofeedback modulation method and system disclosed in this invention for rehabilitation training of patients with brain disorders focuses on the fine-grained modulation of neurofeedback from motor intention to motor execution during rehabilitation training. Based on limited brain-motor synchronization data, data augmentation is performed. Through training with a large amount of data, the neurofeedback modulation mechanism between brain and muscle electromyography in healthy individuals is explored. This mechanism is then used to achieve fine-grained neurofeedback modulation of the patient's motor execution, thereby improving the rehabilitation training effect and assessment level.

[0051] Those skilled in the art should understand that variations can be implemented by combining existing technology with the above embodiments, which will not be elaborated here. Such variations do not affect the essence of the present invention, and will not be elaborated here either.

[0052] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and the devices and structures not described in detail should be understood as being implemented in a conventional manner in the art. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the present invention. This does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention's technical solutions still fall within the protection scope of the present invention.

Claims

1. A neurofeedback modulation system for rehabilitation training of patients with brain disorders, characterized in that, The system includes: a first brain-EMG acquisition module, a second brain-EMG acquisition module, a generative adversarial network module, a computational processing module, and a modulation module; The first brain-muscle electroacupuncture acquisition module is used to synchronously acquire brain-muscle electroacupuncture data of healthy individuals during rehabilitation training simulating patients with brain disorders, so as to obtain the real EEG sequence and real EMG sequence of healthy individuals; The second brain-myograph acquisition module is used to simultaneously acquire real-time brain-myograph data of patients with brain disorders to obtain the real EEG and real EMG sequences of patients with brain disorders; The generative adversarial network is communicatively connected to the first EEG-EMG acquisition module to provide a trained translation model based on real EEG and EMG sequences of healthy individuals. The trained translation model provides a mapping relationship between EEG and EMG sequences. The generative adversarial network (GAN) comprises a generator network and a discriminator network. The generator network includes a generator model and a translation model, and the discriminator network includes a first discriminator model and a second discriminator model. The generator model generates an EEG sequence based on random noise, and the translation model obtains a translated EMG sequence that maps to the generated EEG sequence. The first discriminator model distinguishes between the real EEG sequence of a healthy person and the generated EEG sequence, and inputs the difference between the real and generated EEG sequences as a first reward value into the generator model. The second discriminator model distinguishes between the real EMG sequence of a healthy person and the translated EMG sequence, and inputs the difference between the real and translated EMG sequences as a second reward value into the generator model. The computational processing module is communicatively connected to the generative adversarial network, the second EEG acquisition module, and the modulation module, respectively, so as to calculate the difference between the ideal EMG sequence and the synchronously acquired real EMG sequence of the brain disorder patient after mapping the real EEG sequence of the brain disorder patient to the ideal EMG sequence using the trained translation model, and transmit the calculation result to the modulation module. The modulation module is used to modulate the fine parameters of the neural feedback signals in the rehabilitation training of patients with brain disorders in real time.

2. The neurofeedback modulation system for rehabilitation training of patients with brain disorders as described in claim 1, characterized in that, The first and second EEG acquisition modules include an EEG acquisition unit and an EMG acquisition unit; the EEG acquisition unit includes an EEG acquisition subunit and a wireless transmission subunit, and the EEG acquisition subunit is connected to an EEG cap; the EMG acquisition unit includes an EMG acquisition subunit and a wireless transmission subunit, and the EMG acquisition subunit is connected to EMG electrodes; both the EEG acquisition subunit and the EMG acquisition subunit transmit data to a generative adversarial network through the wireless transmission subunit.

Images