Method and apparatus for adaptive neural interfaces

Abstract

A method of predictive analysis of one or more characteristics associated with multiple neurostimulators used by respective multiple users, the method comprising: processing at a first site first input data collected from a first plurality of neurostimulation devices and / or neurostimulation users, and generating first machine learning model data from the first input data, the first machine learning model data comprising first model parameters; processing at a second site second input data collected from a second plurality of neurostimulation devices and / or neurostimulation users, and generating second machine learning model data from the second input data, the second machine learning model data comprising second model parameters; processing the first model parameters and the second model parameters to generate third machine learning model data representative of at least one neurostimulation characteristic; and using the third machine learning model data to predict evolution of at least one neurostimulation characteristic.

Description

[0001] METHOD AND APPARATUS FOR ADAPTIVE NEURAL INTERFACES

[0002] TECHNICAL FIELD

[0003] The present disclosure relates to the field of neural interfaces, especially but not limited to electrical neurostimulation. Some aspects of the disclosure relate to controlling and / or analysing electrical neurostimulation, especially but not exclusively, for a spinal cord injury patient.

[0004] BACKGROUND

[0005] Spinal cord injuries (SCI) and degenerative diseases usually have a large impact on the lives of people having such medical condition. Spinal cord injuries and diseases may disrupt communication between the brain and the rest of the body, causing motor and sensory impairments. Partial or complete paralysis, sensory deficits and chronic pains are some of the problems that spinal cord illnesses may cause. Beyond motor and sensory impairments, spinal cord conditions can disrupt autonomic functions, which regulate processes like blood pressure, heart rate and temperature control. Unstable blood pressure can negatively affect a person’s feeling of wellbeing and good health, but can also cause serious morbidities, for example, in the case of autonomic dysreflexia.

[0006] Electrical stimulation has emerged as a promising technique for people with spinal cord conditions. Such electrical stimulation is typically known as Spinal Cord Stimulation (SCS); SCS also frequently denotes a Spinal Cord Stimulator. This technique involves delivering electrical impulses to the spinal cord. The stimulation can help improve the condition of neural pathways that have been affected in some way by the injury, for instance, the neural pathways get damaged or become inactive as a result of the injury. The technique can also facilitate triggering of neurons for controlling muscles or muscle groups, for regaining lost function.

[0007] It remains a technical challenge to deliver the full benefits of electrical neurostimulation to patients, and to maintain performance after a neurostimulator has been implanted.

[0008] DESCRIPTIONBroadly speaking, one aspect of the present disclosure arises from an appreciation by the inventors that neurostimulation, its effect, and treatment of neurological deficiency, can be highly individual. Each person has their own set of characteristics, both personally and medically, and what works on a person may not work in the same way or at all on a different person, even if they share some of these characteristics. Accordingly, to fully exploit the benefits of electrical stimulation, the stimuli applied to the spinal cord of a subject should be tailored, or personalized, or customized, to the particular situation that the person is in and the characteristics thereof.

[0009] Broadly speaking, another aspect of the disclosure arises from an appreciation by the inventors that, whether or not affected by individuals’ characteristics, there are many relationships between different anatomical, neurological and neurostimulation characteristics that are not yet fully understood and so are difficult for medical practitioners to predict for an individual person. Moreover, it is difficult to predict the effect and evolution of the therapy. This appreciation applies not only to neurostimulation, but also to neural sensing, optionally via a neural interface including a brain signal sensor, for example, a brain-computer interface (BCI).

[0010] Broadly speaking, one aspect of the invention provides a technique (also referred to as a process pipeline) for predictive analysis of at least one characteristic associated with neurostimulation and / or neural sensing. The technique is especially suitable for neurostimulation for treating neurological deficiency, for example, a spinal cord injury, but the technique is not limited only to this. Additionally or alternatively, the technique is especially suitable for spinal cord stimulation, but the technique is not limited to this. Additionally or alternatively, the technique is especially suitable for decoding brain signals sensed by a brain computer interface.

[0011] The technique may comprise:

[0012] processing at a first site first input data collected from a first plurality of neural interface devices (e.g. neurostimulation devices and / or neural sensing devices) and / or neural interface users (e.g. neurostimulation users and / or users of neural sensing), and generating first machine learning model data from the first input data, the first machine learning model data comprising first model parameters; and

[0013] processing at a second site second input data collected from a second plurality of neural interface devices (e.g. neurostimulation devices and / or neural sensingdevices) and / or neural interface users (e.g. neurostimulation users and / or users of neural sensing), and generating second machine learning model data from the second input data, the second machine learning model data comprising second model parameters.

[0014] The technique may optionally further comprise:

[0015] processing the first model parameters and the second model parameters to generate third machine learning model data representative of the at least one characteristic; and

[0016] using the third machine learning model data to predict evolution of the at least one characteristic.

[0017] Although only two processing sites are referred to, more than two sites may be used and can generate additional machine-learning model data. Processing at multiple sites facilitates scalability, and development of robust machine learning model data, in a decentralised way, from diverse input data.

[0018] This technique can enable historical data to be used to establish machine learning model data able to be used for predictive analysis of one or more characteristics associated with an individual’s neurostimulation and / or neural sensing.

[0019] The first and second pluralities may be distinct from one another.

[0020] The first and second sites may be geographically separated and / or distributed.

[0021] In some embodiments, the step of processing the first model parameters and the second model parameters comprises performing graph analysis, for example, to identify relationships between the model parameters useful for predictive analysis of the at least one neurostimulation characteristic.

[0022] In some embodiments, the technique may further comprise using at least one of the first, second and third machine learning model data to generate an adjustment based on the predicted evolution, wherein the adjustment comprises one or more of: an adjustment for an operating parameter of the neurostimulator; and / or a use adjustment for adoption by the subject person; and / or an operating parameter of a brain computer interface system. The evolution may be a deterioration compared to a target condition, and the adjustment generated for remediating at least partly the deterioration.In some embodiments, the operating parameter comprises one or more selected from: one or more durations of neurostimulation operation; one of signal parameters of the electrical pulses; one or more battery charging parameters; one or more electrode parameters for the neurostimulator; one or more feedback parameters used by the neurostimulator for controlling generation of electrical pulses; one or more parameters used for decoding brain signals sensed from the brain; one or more parameters used for controlling neurostimulation in response to sensed brain signals.

[0023] The characteristic may be a battery performance characteristic, and optionally the adjustment comprises an adjustment for battery performance.

[0024] Additionally or alternatively, the characteristic may be associated with maintenance servicing of the neurostimulator and / or with a technical deficiency of the neurostimulator. Additionally or alternatively, the characteristic may be associated with maintenance servicing of neural sensing apparatus and / or with a technical deficiency of neural sensing apparatus.

[0025] The technique may further comprise updating at least one of the first machine learning model data and the second machine learning model data based on: the third machine learning model data; and / or on the predicted evolution.

[0026] In some embodiments, the step of generating third machine learning model data may comprise generating plural candidate model data, and selecting a respective one of the candidate model data as the third machine learning model data. The selection may be based on validation of the candidates model to predict past evolution.

[0027] The first input data and / or the second input data may comprise data collected at different points in time, such that the data represents a time evolution over a period of time (for example, days, weeks, months and / or years).

[0028] In some embodiments, the technique may comprise analysing a predicted evolution of the at least one neurostimulation characteristic to determine a degree to which the predicted evolution concords with (e.g. past) time evolution of the first input data. The input data may comprise one or more of:

[0029] first information about the neurostimulator and about operation of the neurostimulator;second information about an implantation and / or implantation configuration of the neurostimulator;

[0030] third information about the subject person;

[0031] fourth information about an efficacy of neurostimulation for the subject person; fifth information about other collected sensor inputs, optionally from one or more of: blood pressure information measured by a blood pressure sensor; information from an inertial measurement unit; information from a brain signal sensor.

[0032] The technique may further comprise a step of generating stimulation parameters for downloading to a transcutaneous neurostimulator, and optionally a step of downloading the stimulation parameters to the transcutaneous neurostimulator.

[0033] Additionally or alternatively, the technique may further comprise a step of generating stimulation parameters for downloading to an implantable neurostimulator, and optionally a step of downloading the stimulation parameters to the implantable neurostimulator.

[0034] Additionally or alternatively, the technique may further comprise a step of generating decoder parameters for downloading to a brain computer interface system for decoding and responding to sensed brain signals, and optionally a step of downloading the decoder parameters to the brain computer interface system.

[0035] Additional aspects of the disclosure are defined in the claims.

[0036] Additional advantages and features of the present disclosure will become apparent from the detailed description that follows and will be particularly pointed out in the appended claims. Although certain aspects, features and advantages have been highlighted above and in the claims, protection is claimed for any novel feature described herein and / or illustrated in the drawings, whether or not emphasis has been placed thereon.

[0037] BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Non-limiting embodiments are now described by way of example, with reference to the accompanying drawings:Fig. 1 is a schematic diagram illustrated information flows for federated machine learning for predictive analytics of at least one characteristic associated with neurostimulation in an embodiment of the invention.

[0039] Fig. 2 is a schematic diagram illustrating information types that may form collected input data in Fig. 1

[0040] Fig. 3 is a schematic flow diagram illustrating steps for generating a machine-learning model for predictive analytics in an embodiment of the invention.

[0041] Fig. 4 is a schematic block diagram illustrating components associated with a neurostimulator and communication in the illustrated embodiment.

[0042] DETAILED DESCRIPTION

[0043] Examples in the present disclosure provide one or more solutions for predictive analytics in the field of SCS devices, especially applicable to SCI patients, more especially to SCI patents with cervical or thoracic injuries. However, the principles described herein may equally be applicable to other fields of neurostimulation therapy of a neurological deficiency or dysfunction, and / or to other neural interfaces, for example including neural sensing (e.g., a brain computer interface “BCI”). References to SCImay be extended to other fields of neurostimulation therapy unless the context prevents this. References to neurostimulation may be extended to neural sensing (in particular to neural sensing using a brain signal sensor) unless the context prevents this.

[0044] Predictive analytics leverages data mining, and machine learning to analyze current and historical data, enabling insightful predictions and enabling proactive actions. These models can enable personalized therapy for SCI patients, providing valuable insights and cost efficiencies through data collection.

[0045] Hitherto, no automated real-time predictive analytics pipeline has been available for electrical stimulation therapy in SCI. The invention addresses this gap by designing a processing pipeline that can be robust, optionally automated, and enabling handling everything from data collection and cleansing to modeling, validation, deployment, and real-time adjustment. This pipeline handles challenges like ensuring high-quality data, extracting robust signals, and eliminating noise from real-world neural and behavioural signals. The applications of this pipeline are vast and impactful. For instance, it canpredict patient-reported outcomes to specific therapies, forecast hospital readmission rates for SCI patients, and adjust stimulation parameters for controlling neurostimulaton based on real-time predictive models.

[0046] Moreover, the processing pipeline can be adapted for real-time predictions on edge devices, such as smartwatches or controllers, with scheduled model adjustments. Each application can collect different data sources (e.g. from clinical studies and via personal companion applications) and be customized for deployment on cloud or edge devices, allowing both flexibility and precision.

[0047] Referring to Fig. 1, predictive analysis is carried out to predict evolution of at least one characteristic, for example, a neurostimulation characteristic or a brain-computer interface characteristic as described above. The predictive analysis is based on machine learning techniques applied to input data collected from one or more pluralities of neurostimulation devices and / or neurostimulation users and / or brain computer interfaces for controlling neurostimulation. Machine learning model data is generated from the collected input data. As shown in Fig. 1, a federated learning technique may be implemented in some embodiments. Local data processing is carried out by respective local data processors at a plurality of different sites 10a, 10b, 10c, 10d (etc.), using respect input data collected at, or through, or transmitted to, the respective site. Each site generates its own respective machine learning model data using the input data, the machine learning model data comprising or represented by respective model parameters 12.

[0048] The model parameters 12 from the different sites 10a-d are processed collectively at a further “central” processing site 14, by a respective data processor, to generate aggregate or compound machine learning model data, representative of at least one (optionally plural) characteristic. The central processing site 14 may also be referred to as a “cloud”, and may be implemented by distributed processing devices with which the sites 10a-d can communicate. The compound model data 16 is distributed optionally back to the sites 10a-d, or to edge devices 18 of the data network. Edge devices 18 may include, for example, neurostimulation controllers, computing devices or terminals, used at the sites 10a-d or in other locations (e.g. at other rehabilitation clinics or at the homes of individual users) for administering neurostimulation and / or administering neurostimulation devices.Fig. 2 illustrates schematically the kinds of information that may be collected as input data. For example, one or more (or all) of:

[0049] - First information 40 about an individual’s neurostimulator and about operation of the neurostimulator. The first information may include any of: the neurostimulator model number; serial number; log information summarising neurostimulation operation and events; log information regarding battery charging; log information regarding errors recorded or intercepted during operation; log information regarding bioimpedance measurements or bioimpedance testing.

[0050] - Second information 42 about an implantation configuration of the neurostimulator. The second information may include any of: the type of electrode or electrodes used by the neurostimulator; mapping information about the relative positioning and / or alignment of the electrode with respect to the segments of spinal cord.

[0051] - Third information 44 about the subject person. The third information may include any personal information that may have an impact on neurostimulation or efficacy. For example, the person’s age, sex, weight, height, fitness level, relative mobility, type of neurological deficiency, position of a spinal cord injury, date of spinal cord injury, severity (e.g. classification) of spinal cord injury, and details of any other medical problems impacting the person.

[0052] - Fourth information 46 about the efficacy of neurostimulation. For example, the fourth information may include the individual’s personal feedback on efficacy, and / or observations from clinician’s treating or monitoring the individual.

[0053] - Fifth information 48 about other collected sensor inputs. The fifth information may include any of: information about blood pressure measured by a blood pressure sensor; information from one or more inertial movement units (IMUs) for sensing body movement and / or posture; information from a brain signal sensor used to monitor brain activity (e.g. satisfaction, fatigue, irritation, etc) and / or to detect brain signals for commanding operation of the neurostimulator (for example, brain signals sensed from motor cortex regions of the brain). Referring to also to Fig. 1, multiple possibilities are envisaged for collecting such information. Certain information may be collected or uploaded from neurostimulators (or their “hubs” described later) 110 or edge devices 18 or 126 at the sites 10a-d forprocessing at the sites. Additionally or alternatively, certain information may be uploaded to the central data processor, for example, log data from the neurostimulator or information input through edge devices. Some information may be collected and / or processed only at the sites 10a-d. Additionally or alternatively, some information may be collected and / or processed only at the central processor 14.

[0054] Referring to Fig. 3, a flow or pipeline for processing information for predictive analytics may comprise one or more of the follow steps or functional modules:

[0055] Data and Workflow Analysis 50: An preparation step of comprehensively understanding the type of data available, the existing workflow, the target audience, and the specific actions that will be prompted by accurate predictions.

[0056] Problem Definition and Data Preparation 52: Clearly define the problem that is aimed to be addressed through predictive analytics. Gather the necessary initial data and preprocess it to ensure its quality and relevance. This includes cleansing the data to remove noise and inconsistencies. Evaluate multiple machine learning (ML) algorithms, such as logistic regression, gradient boosting machines, and random forests, to build a robust predictive model.

[0057] Federated Learning Integration 54: Implement federated learning to enhance model training using decentralized data sources. This approach allows the predictive model to learn from diverse datasets located at different institutions or edge devices without the need to centralize data, thereby preserving data privacy and security while improving the model’s generalizability and robustness.

[0058] Graph Theory Applications 56: Utilize graph theory to model and analyze the complex relationships and interactions within the data collected from SCI patients. Construct graph-based representations of neural networks, patient treatment histories, and other relevant data to uncover patterns and insights that traditional methods might overlook. Apply algorithms such as graph convolutional networks (GCNs) to improve the predictive accuracy and interpretability of the model.

[0059] Model Refinement and Validation 58: Refine the predictive process by selecting the best-performing model from the initial evaluation. Test this model with a separatedataset to validate its accuracy and reliability. This step ensures that the chosen model can generalize well to new, unseen data.

[0060] Real-World Deployment 60: Deploy the validated model in a real-world setting. Utilize the model to predict outcomes based on new input variables, enabling informed decision-making and personalized therapy adjustments in neuromodulation applications.

[0061] Real-Time Model Adjustment 62: Continuously adjust the ML model using real-time data sampling. In cloud computing environments, this adjustment is performed in realtime, while in edge computing environments, it is done on a scheduled basis. This dynamic adjustment ensures that the predictive model remains accurate and relevant over time.

[0062] By automating this pipeline and utilizing it on both edge and cloud infrastructures, the invention provides a scalable and efficient solution for predicting future data and improving patient outcomes in neuromodulation therapies. This method not only enhances the precision of eSCS therapy but also provide innovative applications across various neuromodulation scenarios. Integrating federated learning and graph theory concepts further strengthens the predictive capabilities and operational efficiency of the pipeline, setting a new standard in personalized healthcare through advanced Al technologies.

[0063] Example advantages that can be achieved include:

[0064] - Personalized Therapy: By accurately predicting patient-specific outcomes, the pipeline enables highly personalized therapy adjustments. This leads to more effective treatments toward the unique needs and conditions of each patient, ultimately improving therapeutic outcomes.

[0065] - Enhanced Decision-Making: The pipeline provides healthcare professionals with data-driven insights, allowing for more informed decision-making. Predictive models offer guidance on the best course of action, such as adjusting stimulation parameters or identifying high-risk patients, which can lead to better management of chronic conditions for SCI population.- Real-Time Adaptability: The continuous adjustment of the machine learning model using real-time data ensures that the predictive analytics remain accurate and relevant. This adaptability is crucial for maintaining the efficacy of neuromodulation therapies as patient conditions evolve.

[0066] - Scalability and Efficiency: Automating the predictive analytics pipeline and deploying it on both cloud and edge infrastructures provide a scalable solution that can handle large volumes of data efficiently. This automation reduces the manual workload for clinicians and streamlines the process of data analysis and model deployment.

[0067] - Improved Patient Outcomes: With predictive analytics guiding therapy, patients are likely to experience better outcomes, such as reduced pain, improved motor function, and enhanced quality of life. The ability to predict and preemptively address potential issues can also reduce hospital readmission rates and healthcare costs.

[0068] - Broad Applicability: The method is not limited to a single type of neuromodulation therapy. It can be adapted for various applications, including different types of electrical stimulation therapies and other neuromodulation techniques. This versatility makes the pipeline a valuable tool across multiple therapeutic areas.

[0069] - Data-Driven Insights: The pipeline leverages diverse data sources, including data from clinical studies (data aggregation in Empower EDC portal), preclinical studies, and real-world patient data. This comprehensive approach ensures that the predictive models are built on robust, high-quality data, leading to more reliable and actionable insights.

[0070] Fig. 4 illustrates a local system 100 of one or more edge devices and neurostimulator arrangements for use in the pipeline described above. The system may be a SCS system 100 according to some embodiments. The system includes a stimulator 130 which, in the present example, is an implantable unit, ie. a unit that is, or is intended to be, implanted within the body. The stimulator 130 may also be referred to as a pulse generator, for generating sequences of electrical pulse signals. At least one implantable electrode 132 (which may be in the form of an electrode array, for example, a paddle array or a segmented, cylindrical array) is connected to the stimulator 130 by one or more implantable leads. The electrode 132 is configured forplacement in the vicinity of the spinal cord, in order apply targeted electrical stimulation signals. However, in other embodiments, the stimulator 130 and / or the electrodes 132 may be external to the body

[0071] In Figure 4, the zone 10 generally depicts a region that is within the patient’s body, for implantable devices. Zone 12 generally depicts a region that is outside the patient’s body but within the vicinity of the patient, for devices that the patient may, for example, wear or carry as external devices. Such external devices may include a stimulation control apparatus 110 also referred to herein as a “hub” 110. The term “hub” is used interchangeably with stimulation control apparatus. The hub 110 may interact with the stimulator 130 by means of short-range wireless communication, for example, based on inductive coupling when placed sufficiently closely together. The hub 110 may serve as a power supply for charging the stimulator 130 by inductive coupling, and / or as a controller for controlling stimulation functions. Example functions may include any one or more of: (i) uploading stimulation programs to the stimulator 130, (ii) dynamically adjusting stimulation parameters, (iii) commanding starting, stopping and / or pausing of the stimulation, (iv) downloading information from the stimulator, such as functional status information (e.g. relating to the level of battery charge or battery life), and / or log information and / or other performance information relating to operation of the stimulator 130.

[0072] System 100 may further comprise one or more sensors 120, for providing input to the neurostimulator and / or hub. The sensor 120 may comprise one or more of: a blood pressure sensor for providing measurements of the individual’s blood pressure; and / or an inertial measurement unit (IMU) for providing information about the individual’s posture and / or body position and / or body movement; and / or a brain signal sensor for sensing signals from within the brain, such as signals sensed from a motor cortex region of the brain. The sensor 120 is illustrated as spanning zones 10 and 12. In some embodiments, the sensor 120 may be implantable (wholly or at least partly) in the body. The sensor 120 may optionally be included in the stimulator 130 and / or the electrode 132, or it may a discrete implantable device, for example, in the case of a brain signal sensor. In other embodiments, the sensor 120 may be or comprise an external device, for example, an externally wearable device. Examples can include an inflatable cuff for blood pressure measurements, or a sensor using other blood pressure sensing techniques, such as optical measurements, worn on the arm, handor finger, and / or attachable to the skin (for example, by a skin compatible adhesive); a wearable IMU; an external module of a brain signal sensor.

[0073] The sensor 120 communicates with the stimulator 130 and / or the hub 110, to provide information for closed-loop feedback or other control, for example, brain computer interface control of the stimulator. Depending on the complexity of the brain signal sensor, additional decoding and / or interpretation of brain signals may be implemented, for example, in the hub 110 or an additional processing device (not shown) of a brain computer interface system. In the following description, the sensor response control is provided in the hub 110, but equivalent functional modules may additionally or alternatively be provided in the stimulator 130 or elsewhere.

[0074] The hub 110 comprises the following functional modules:

[0075] an input module 112 for processing input signals from the one or more sensors 120;

[0076] a processing module 114 for determining that electrical stimulation of a spinal cord of the person is to be applied, for example, based (e.g. at least partly) on the processed signals; and

[0077] an output control module 116 for providing one or more commands for controlling operation of the stimulator 130 upon determining that electrical stimulation is to be applied to the spinal cord of the person.

[0078] Optionally, the individual may control the stimulator 130, via the hub 110, by means of a portable device 124, for example, a smart-watch or telephone. The portable device 124 may communicate with the hub 110 by a wireless communication channel, for example, a Bluetooth® channel or other short range wireless communications protocol. The portable device 124 may run application software presenting a userinterface for starting, stopping or pausing stimulation, for changing a stimulation program, or (as explained below) for providing user-feedback regarding the effectiveness of the stimulation, for use by the predictive analytics pipeline. The portable device may also represent an edge device for the predictive analytics network.

[0079] The hub 110 further comprises an adaptation module 118 for adapting or adjusting the control implemented by the hub 110, based on externally received inputs. Theadaptation module 118 enables stimulation to be tuned, or customized, based, for example, on the results of predictive analytics, and may also represent an edge device thereof.

[0080] Zone 14 represents a data processing environment distinct from and / or remote from the user, typically corresponding to a respective one of the sites 10a-d associated with the user’s rehabilitation therapy, and / or the central data processor 14. A data processor 140 at zone 14 may communicate with the hub 110 by any suitable data connection, including a wired connection (e.g. when the hub 110 is plugged into a wired communication channel), and / or a wireless connection (e.g. wifi, Bluetooth, or cellular), or via a local intermediate device, such as the user’s mobile phone. The connection may be a direct connection, or an indirect connection, for example via the internet. The hub 110 may also optionally be configured to transmit to the data processor 140 one or more of the currently used feedback-control parameters for the hub 110, and / or log information relating to performance and / or operation of the stimulator 130.

[0081] Information received from a data processor 140 can provide the subject person, or a supervising clinician, with recommendations for adjusting neurostimulation. Additionally or alternatively, the data processor 140 can download adjustment changes to one or more neurostimulation programs or parameters to the hub 110. The adjustment may be or comprise one or more of: an adjustment for an operating parameter of the neurostimulator; and / or a use adjustment for adoption by the subject person; and / or an adjustment for a parameter of a brain computer interface or other brain signal sensor and / or decoder.

[0082] For example, where an evolution predicted by the predictive analytics is a deterioration compared to a target condition, the adjustment may be generated for remediating at least partly that deterioration. This may be expressed by a characteristic such as: one or more durations of neurostimulation operation; one or more signal parameters of the electrical pulses; one or more battery charging parameters; one or more electrode parameters for the neurostimulator; one or more sensor parameters used by the neurostimulator for controlling generation of electrical pulses. In one example, the characteristic is a battery performance characteristic, and the adjustment comprises an adjustment for battery performance. In another example, the characteristic isassociated with maintenance servicing of the neurostimulator and / or with a potential technical deficiency of the neurostimulator or the electrodes (e.g. in case it is predicted that the neurostimulator and / or electrode will be in need of replacement).

[0083] Although some examples and embodiments may include a particular sequence of operations, the sequence may in some cases be altered without departing from the scope of the present disclosure. For example, some of the operations depicted may be performed in parallel or in a different sequence that does not materially affect the functions as described in the examples. In other examples, different components of an example device or system that implements an example method may perform functions at substantially the same time or in a specific sequence.

[0084] As used herein, the terms “module”, “computing device” and “processor” may refer to any one or more circuits or virtual circuits (e.g., a physical circuit emulated by logic executing on an actual processor) that manipulates data values according to control signals (e.g., commands, opcodes, machine code, control words, macroinstructions, etc.) and which produces corresponding output signals that are applied to operate a machine. A computing device or processor may, for example, include at least one of a Central Processing Unit (CPU), a Reduced Instruction Set Computing (RISC) Processor, a Complex Instruction Set Computing (CISC) Processor, a Graphics Processing Unit (GPU), a Digital Signal Processor (DSP), a Tensor Processing Unit (TPU), a Neural Processing Unit (NPU), a Vision Processing Unit (VPU), a Machine Learning Accelerator, an Artificial Intelligence Accelerator, an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), a Radio-Frequency Integrated Circuit (RFIC), a Neuromorphic Processor, a Quantum Processor, or any combination thereof. A computing device or processor may be a multi-core processor having two or more independent processors (sometimes referred to as “cores”) that may execute instructions contemporaneously. Multi-core processors may contain multiple computational cores on a single integrated circuit die, each of which can independently execute program instructions in parallel. Parallel processing on multi-core processors may be implemented via architectures like superscalar, VLIW, vector processing, or SIMD that allow each core to run separate instruction streams concurrently. A computing device or processor may be emulated in software, running on a physical processor, as a virtual processor or virtual circuit. The virtualprocessor may behave like an independent processor but is implemented in software rather than hardware.

[0085] The various operations of example methods described herein may be performed, at least partially, by one or more processors that are temporarily configured (e.g., by software) or permanently configured to perform the relevant operations. Whether temporarily or permanently configured, such processors may constitute processor-implemented modules / components that operate to perform one or more operations or functions. The modules / components referred to herein may, in some examples, comprise processor-implemented modules / components.

[0086] Similarly, the methods described herein may be at least partially processor-implemented. For example, at least some of the operations of a method may be performed by one or more processors or processor-implemented modules / components. The performance of certain of the operations may be distributed among the one or more processors, not only residing within a single machine, but deployed across a number of machines. In some examples, the processor or processors may be located in a single location (e.g., within a home environment, an office environment, or a server farm), while in other examples the processors may be distributed across a number of locations.

[0087] Examples may be implemented in digital electronic circuitry, or in computer hardware, firmware, or software, or in combinations of them. Examples may be implemented using a computer program product, e.g., a computer program tangibly embodied in an information carrier, e.g., in a machine-readable medium for execution by, or to control the operation of, data processing apparatus, e.g., a programmable processor, a computer, or multiple computers.

[0088] In this text, the term “includes”, “comprises” and derivations thereof (such as “including”, “comprising”, etc.) should not be understood in an excluding sense, that is, these terms should not be interpreted as excluding the possibility that what is described and defined may include further elements, steps, etc.

[0089] On the other hand, the disclosure is obviously not limited to the specific embodiment(s) described herein, but also encompasses any variations that may be considered by any person skilled in the art (for example, as regards the choice of materials, dimensions,components, configuration, etc.), within the general scope of the invention as defined in the claims.

Claims

CLAIMS1. A method of predictive analysis of at least one characteristic associated with multiple neurostimulators used by respective multiple users, the method comprising:processing at a first site first input data collected from a first plurality of neurostimulation devices and / or neurostimulation users, and generating first machine learning model data from the first input data, the first machine learning model data comprising first model parameters;processing at a second site second input data collected from a second plurality of neurostimulation devices and / or neurostimulation users, and generating second machine learning model data from the second input data, the second machine learning model data comprising second model parameters;processing the first model parameters and the second model parameters to generate third machine learning model data representative of the at least one characteristic; andusing the third machine learning model data to predict evolution of the at least one characteristic.

2. A method according to claim 1, wherein the characteristic is associated with neurostimulation therapy for one or more neurological deficiencies; and / or wherein the multiple neurostimulators are used by neurostimulation users for treating one or more neurological deficiencies; and / or wherein the characteristic is a characteristic of brain computer interface for controlling the neurostimulator.

3. A method according to claim 2, wherein the one or more neurological deficiencies comprise a spinal cord injury and / or wherein the neurostimulation comprises spinal cord stimulation.

4. A method according to any preceding claim, wherein the first and second pluralities are distinct from one another.

5. A method according to any preceding claim, wherein the first and second sites are geographically separated and / or distributed.

6. A method according to any preceding claim, wherein the step of processing the first model parameters and the second model parameters comprises performing graph analysis.

7. A method according to claim 6, wherein the step of performing graph analysis comprises identifying relationships between the model parameters useful for predictive analysis of the at least one neurostimulation characteristic.

8. A method according to any preceding claim, further comprising using at least one of the first, second and third machine learning model data to generate an adjustment based on the predicted evolution, wherein the adjustment comprises one or more of: an adjustment for an operating parameter of the neurostimulator; and / or a use adjustment for adoption by the subject person; and / or an adjustment for an operating parameter of a brain computer interface system.

9. A method according to claim 8, wherein the evolution is a deterioration compared to a target condition, and wherein the adjustment is generated for remediating at least partly the deterioration.

10. A method according to claim 8 or 9, wherein the operating parameter comprises one or more selected from: one or more durations of neurostimulation operation; one of signal parameters of the electrical pulses; one or more battery charging parameters; one or more electrode parameters for the neurostimulator; one or more feedbackparameters used by the neurostimulator for controlling generation of electrical pulses; one or more parameters used for decoding brain signals sensed from the brain; one or more parameters used for controlling neurostimulation in response to sensed brain signals.

11. A method according to claim 8, 9 or 10, wherein the characteristic is a battery performance characteristic, optionally wherein the adjustment comprises an adjustment for battery performance.

12. A method according to any preceding claim, wherein the characteristic is associated with maintenance servicing of the neurostimulator and / or with a technical deficiency of the neurostimulator.

13. A method according to any preceding claim, further comprising updating at least one of the first machine learning model data and the second machine learning model data based on: the third machine learning model data; and / or on the predicted evolution.

14. A method according to any preceding claim, wherein the step of generating third machine learning model data comprises generating plural candidate model data, and selecting a respective one of the candidate model data as the third machine learning model data.

15. A method according to any preceding claim, wherein the first input data comprises data collected at different points in time, such that the data represents a time evolution.2116. A method according to claim 15, further comprising analysing a predicted evolution of the at least one characteristic to determine a degree to which the predicted evolution concords with time evolution of the first input data.

17. A method according to any preceding claim, wherein the input data comprises one or more of:first information about the neurostimulator and about operation of the neurostimulator;second information about an implantation configuration of the neurostimulator; third information about the subject person;fourth information about an efficacy of neurostimulation for the subject person; fifth information about other collected sensor inputs, optionally from one or more of: blood pressure information measured by a blood pressure sensor; information from an inertial measurement unit; information from a brain signal sensor.

19. A method according to any preceding claim, further comprising a step of generating stimulation parameters for downloading to a transcutaneous neurostimulator, and optionally a step of downloading the stimulation parameters to the transcutaneous neurostimulator.

20. A method according to any preceding claim, further comprising a step of generating stimulation parameters for downloading to an implantable neurostimulator, and optionally a step of downloading the stimulation parameters to the implantable neurostimulator.

21. A method according to any preceding claim, further comprising a step of generating decoder parameters for downloading to a brain computer interface system for decoding and responding to sensed brain signals, and optionally a step of downloading the decoder parameters to the brain computer interface system.2222. A method of predictive analysis of at least one characteristic associated with multiple neural interface devices used by respective multiple users, the method comprising:processing at a first site first input data collected from a first plurality of neural interface devices and / or neural interface device users, and generating first machine learning model data from the first input data, the first machine learning model data comprising first model parameters;processing at a second site second input data collected from a second plurality of neural interface devices and / or neural interface device users, and generating second machine learning model data from the second input data, the second machine learning model data comprising second model parameters;processing the first model parameters and the second model parameters to generate third machine learning model data representative of the at least one characteristic; andusing the third machine learning model data to predict evolution of the at least one characteristic.

23. A system configured to perform a method as defined in any preceding claim.

24. Computer software code which when executed by processor circuitry causes the processor circuitry to implement a method as defined in any of claims 1 to 22.

Images