System for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient

Abstract

A system (100) for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient, comprising: a sensing unit (10) comprising a first sensor means (12) for detecting brain signals of the patient, and a second sensor means (14) for detecting behavioral signals of the patient; a stimulation unit (16) comprising a stimulation means (18) configured to deliver neuromodulation / neurostimulation to brain structures of ths patient, and a control unit (20), operatively connected to the sensing unit (10) and the stimulation unit (16), configured to control operation of the sensing unit (10) and the stimulation unit (16).

Description

[0001] System for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient

[0002] The present invention belongs to the technical field of neural decoding and delivery of neuromodulation / neurostimulation therapy to brain structures of a patient in need thereof.

[0003] As a non-limiting example, neuromodulation / neurostimulation therapy can be delivered to mitigate symptoms of neurological disorders such as Parkinson’s Disease (PD), and / or psychiatric disorders such as Obsessive-Compulsive Disorders (OCD).

[0004] PD is a neurodegenerative disease characterized by the loss of dopaminergic neurons within the substantia nigra of the brain.

[0005] Clinically, the disease manifests mainly with motor impairments, including rigidity, tremor, bradykinesia and postural instability, although non-motor symptoms can lead to significant morbidity.

[0006] Large epidemiologic studies revealed around 10 million people with PD worldwide, affecting all races and cultures [1],

[0007] The age of onset is usually over 60, but it is estimated that one in 10 people is diagnosed even before the age of 50, with slightly more men than women affected.

[0008] Neurological disorders such as PD are characterized by structural and functional abnormalities in multiple brain areas involving several distinct brain systems.

[0009] These abnormalities can be referred to as brain neuronal network deficiencies (BNND), which can broadly vary in causes, type of dysfunction, and clinical manifestation, e.g., existence of deleterious networks in the brain in case of neurological and psychiatric disorders, such as Alzheimer disease, Parkinson’s disease, Huntington’s chorea, depression, and / or schizophrenia. Network abnormalities are also described in patients diagnosed with, or with high-risk of, schizophrenia, attention deficit hyperactivity disorder, bipolar disorder, and / or OCD [2],

[0010] In PD, changes observed in the dynamic functional connectivity are suggestive of a reduction in functional segregation among brain networks and are correlated with the clinical severity of PD symptoms. Additionally, a higher variability in network global efficiency is observed, suggesting an abnormal global integration of brain networks [3],

[0011] Deleterious neural networks are formed based on intimate interactions among key pathogenic factors. These deleterious networks could be the common pathway of these brain disorders and, when triggered by aging, and genetic and environmental factors, may lead to a cascade of pathological alterations that characterize illnesses.

[0012] The brain is a highly complex system that is poorly understood, even more so in pathology.

[0013] Disorders induced by BNND may explain many neuropsychiatric and neurodegenerative diseases, but are presently far from being completely understood.

[0014] Treatments approaching these conditions from a brain network perspective offer a promising alternative to single-target pharmacological therapies that, so far, have largely proven insufficient [4], [5],

[0015] Following an initial diagnosis, PD subjects are typically treated with Levodopa (a precursor of dopamine), with dopamine agonists or monoamine oxidase-B inhibitors. These represent different classes of drugs, and are all characterized by different benefit-risk profiles.

[0016] Subjects using these medications over a long term may experience drug-induced side effects, such as motor fluctuations, dyskinesia, or psychiatric complications [6],

[0017] Usually, these side effects emerge three to five years after starting of medication therapy.

[0018] However, it must be noted that those complications may be also due to disease progression.

[0019] The term “dyskinesia” denotes involuntary, erratic, and writhing movements of the face, arms, legs, and / or trunk which, usually, occur one to two hours after a dose of medication is absorbed into the bloodstream of the subject, and is thus expressing its peak clinical effect.

[0020] Also, many patients are prone to develop drug resistance over time. Therefore, drug therapy alone becomes no longer effective in controlling the symptoms of PD.

[0021] Some neuromodulation therapies, approved by the USA Food and Drug Administration (FDA) and currently applied in clinical routines, have proven able to impact dysfunctional networks, providing successful therapeutic outcomes. At present, local brain stimulation has proven effective in reducing pathological symptoms in most PD patients.

[0022] For instance, Deep Brain Stimulation (DBS) has emerged as an option for PD patients to be an adjunct to medications and provide additional benefit.

[0023] DBS may target either the subthalamic nucleus (STN) or the internal globus pallidus (GPi).

[0024] DBS is a technological evolution of the irreversible surgical ablation of the basal ganglia, a more invasive treatment that was originally used to treat PD-affected patients.

[0025] Instead of permanently damaging some target brain areas, DBS utilizes electrodes that are implanted in the same areas to interfere with pathological signals and thereby improve movements.

[0026] Once a patient is selected as a candidate for DBS, imaging is performed to determine target site(s) for lead placement.

[0027] STN is chosen as the target for treatment of most PD patients. However, GPi may be a better option in some other cases, depending on the observed symptoms.

[0028] Off-period motor symptoms and tremor improve in both targets.

[0029] Advantages of targeting STN target include more medication reduction, less frequent IPG changes, and a more favorable economic profile.

[0030] On the other hand, advantages of targeting GPi include more robust dyskinesia suppression, easier programming, and greater flexibility in medication adjustment [7],

[0031] Once a target site is chosen, SOC surgical stereotactic techniques are used to implant lead(s) at the target site. Here, a frame or a robotic arm is used to control insertion angle and depth. Surgery is planned with a preoperative image with fiducial markers and dedicated image processing.

[0032] Current DBS surgery involves two different surgical steps, i.e., lead implantation and neurostimulation. The leads are placed through burr holes in the skull of the patient and connected to extensions that are tunneled down to a pacemaker-like implantable pulse generator (IPG) that is generally placed in the chest below the clavicle. All of the existing approved DBS systems for this patient population use this basic setup, and contemplate a fully-implanted system.

[0033] DBS with commercially-available leads is typically delivered by using a setup including 4 to 8 metal electrodes delivering energy.

[0034] The selection is based on localization within the target area and, most importantly, on the capacity to induce a therapeutic effect.

[0035] Usually, PD patients who are selected for DBS already attempted one or more pharmacological therapeutic approaches for disease control, which have proven not completely effective.

[0036] DBS was established to be an efficient therapy for the treatment of PD symptoms, in particular with documented improvements in motor control, functionality, and quality of life.

[0037] Moreover, DBS allows on average to reduce Levodopa treatment dose, leading to lower Levodopa-related adverse effects and higher compliance [8],

[0038] Although STN DBS and Levodopa independently lessened motor severity in PD patients to a similar magnitude, their combined effect has proven greater than either treatment alone, thereby suggesting therapeutic synergism [9],

[0039] For this reason, DBS efficacy is generally assessed in off-medication conditions to evaluate effects of DBS, and in combination with antiparkinsonian drugs to assess real-life patients benefits.

[0040] DBS benefits on motor symptoms are most commonly evaluated using both subjective assessment from users and objective assessment by clinicians, e.g., neurologists, using Unified Parkinson’s Disease Rating Scale (UPDRS) and, more particularly, the third part of the scale evaluating motor functions, rather than reporting on individual motor signs.

[0041] Measurement of DBS benefits in PD patients has been established by several class I studies, e.g., in a randomized double-blind design with sham control clinical trial, as discussed, e.g., in Vitek et al.

[0010] (the clinical trial being also known under the name of INTREPID).

[0042] In the context of this clinical trial, PD patients were implanted with a DBS system according to SOC, and were randomly assigned in a 3:1 ratio to receive either active therapeutic stimulation settings (active group) or subtherapeutic stimulation settings (control group) for a 3-month blinded period. During the blinded period, patients were not allowed to increase their antiparkinsonian medications per study protocol

[0010] ,

[0043] The primary efficacy outcome of the study was the difference in mean change from baseline visit to 3 months post-randomization between the active and control groups in the mean number of waking hours per day with good symptom control and no troublesome dyskinesias, as recorded by the user in a dedicated diary, with no increase in antiparkinsonian medications

[0010] .

[0044] The study reported that baseline duration for the active group was 7.78 hours, which increased to 12.37 hours after 12 weeks (3.74 hours improvement on average)

[0010] ,

[0045] Considering that patients were reporting to sleep 7.2 hours per day on average preoperatively, further considering that DBS therapy allows for about 1 more hour of sleep

[0011] , the total awake time after 12 weeks of treatment for active group was estimated around 15.8 hours on average, leading to a ceiling improvement from baseline of time with good symptoms control of about 8 hours

[0010] ,

[0046] Otherwise stated, the 3.74 hours change from baseline reported for active group after 12 weeks of DBS use was less than half of the maximum possible improvement.

[0047] The results of the clinical trial discussed in Vitek et al.

[0010] are resumed in the diagram of Fig. 1.

[0048] LIPDRS III scores were assessed in on- and off-medication conditions at baseline and at 12 weeks for both active and control groups. Change from baseline comparison between groups was performed as secondary endpoint, and showed significant reduction of LIPDRS III score in on-stim / off-meds condition for active group, concluded to be clinically meaningful.

[0049] However, no significant reduction was observed in on-stim / on-meds condition.

[0050] This could be explained by the fact that individual motor scores for tremor (61 %), rigidity (51 %), bradykinesia (32%), and gait (39%) were all improved in the on-stim / off-meds, while only tremor and rigidity were significantly better in the on-stim / on-meds condition

[0010] ,

[0051] These results emphasize the importance of antiparkinsonian medication in symptoms management, and also confirm the role of DBS as an adjunctive therapy for PD treatment, supporting effects of medications on tremor and rigidity, as other symptoms are better managed by medication. From a safety perspective, DBS surgery is overall assumed as characterized by a reasonable risk profile.

[0052] The above-mentioned INTREPID clinical trial discussed in Vitek et al.

[0010] , for example, reported 74 serious adverse events among 177 involved participants. Infection has been the most commonly reported serious adverse event associated with device-hardware / procedure (8 events, representing 2.7% of subjects). There were three events (each) of device- hardware / procedure-related serious adverse events of peri-operative intracranial hemorrhage (representing 1 % of subjects) and seizure (representing 1 % of subjects). After 12 weeks of use, 3 more cerebral hemorrhages and 6 more device-related infections were reported as non- serious events.

[0053] Other meta-analysis and reviews on large cohorts of patients implanted with similar DBS systems (although not all suffering from PD) demonstrated similar outcomes

[0012] -

[0013] , as shown in Table 1.

[0054] It is noticeable that the rate of infections, surgical site complications, and device complications are lower in the most recent reports, showing improvements in the surgical procedure and devices over time.

[0055] Still, more than 10% device complication rate is reported in Rasiah et al.

[0013] , suggesting that there is room for further improvement of device safety and reliability.

[0056] It is indeed a higher revision rate when compared, for example, to established head-mounted implanted active stimulator such as cochlear implants, showing device failure rate below 5% in adults and children, with a cumulative revision rate for primary implants at all ages increased linearly by 1% per year

[0014] ,

[0057] The difference in device complication rate between the results shown in Table 1 and the outcome reported in the context of the INTREPID clinical trial

[0010] may also be reasonably explained in the light of the short follow-up period for the INTREPID clinical trial (only amounting in 12 weeks).

[0058] On the other hand, adverse events related to stimulation are much more frequent.

[0059] In fact, the INTREPID clinical trial

[0010] reported a high incidence of 53.6% stimulation-related adverse events, most frequently involving dyskinesia (16,5%), balance disorders (5%), and dysphagia (3.3%). On the other hand, the results reported in Appleby et al.

[0012] , referring to a clinical trial carried out on a large cohort of participants with longer follow-up time, showed a lower rate of 26.9% for stimulation-related adverse events, most frequently involving speech disturbance (4.8%) and delirium / confusion (4.1%).

[0060] Horn et. al.

[0015] evaluated connectivity between the stimulation site and other brain regions in PD patients treated with DBS by identifying the structural and functional connectivity profile of effective DBS to the STN.

[0061] This evaluation was made by estimating, for each patient, the volume of tissue activated by the DBS electrode from individual postoperative images, and then applying a combination of normative and PD pathologic averaged brain connectome to estimate brain networks activated from a given stimulated volume of tissue

[0015] ,

[0062] In particular, the approach described in Horn et. al.

[0015] included the following steps: acquiring pre- / postoperative imaging (Fig. 2A), localizing DBS electrodes in standard space (Fig. 2B), calculating the volume of tissue activated (VTA) based on stimulation parameters (Fig. 2C), and calculating functional (Fig. 2D) and structural (Fig. 2E) connectivity from the VTA to the rest of the brain using high-quality normative connectome data.

[0063] In the context of Fig. 2, positive correlations are denoted with (a), whereas negative correlations (anticorrelations) are denoted with (b).

[0064] As a result, the Authors observed that a specific pattern of structural and functional connectivity with STN DBS electrodes correlates with clinical outcome across patients in PD. Moreover, the authors could appreciate the prognostic property of this approach by successfully predicting outcomes in an independent cohort of DBS users

[0015] ,

[0065] These findings support the notion that the targets of therapeutic brain stimulation might be brain networks, and not individual brain regions, such as the STN or GPi.

[0066] Another study (described in Boutet et al.

[0016] ), implemented by using functional Magnetic Response Imaging (fMRI) in response to DBS in PD patients, supports this notion and the importance of brain network activation to deep nucleus stimulation.

[0067] The primary purpose of this study was to identify specific markers of optimal DBS using fMRI, to ease device fitting (in particular, to select the optimal electrode and voltage level). The activation of ipsilateral primary motor cortex, ipsilateral thalamus, and contralateral cerebellum was identified as a maker of optimal STN DBS, suggesting a large brain network activation to STN DBS, further optimized with control of activation of this network at different levels

[0016] ,

[0068] This evidence points toward the importance of monitoring brain network response at various relevant locations, to evaluate efficacy of stimulation and potentially adjust delivered therapy accordingly.

[0069] Presently, most brain stimulation therapy solutions, e.g., DBS for PD, Vagus nerve stimulation (VNS) for epilepsy, or spinal cord stimulation for pain, rely on delivery of constant stimulation.

[0070] Here, stimulation is delivered continuously (or, at best, according to a specific pre-established duty cycle), with no real-time adjustment of stimulation parameters and no specificity with respect to individual patient’s biomarkers.

[0071] In particular, therapy parameters are pre-set by a clinician, and then only refined over time in an office setting.

[0072] These known therapeutic approaches are unable to conform to changes in pathology, symptoms severity, and / or daily activities of the patient (e.g., routine activities such as sleeping, standing, walking, or the like).

[0073] That is, fine tuning of delivered therapy is still very sparse in time, and usually occurs in timeframes that are much longer than those needed to provide the patient with a feeling that is close to natural as much as possible.

[0074] Symptomatic activity usually varies over time, e.g., during the day or over multiple days.

[0075] This is explained in Powers et al.

[0017] , explaining the results of a study that was carried out using smartwatch-based motor symptoms detectors for a 6-month period on a large cohort of (non-implanted) PD patients.

[0076] Here, large fluctuations of dyskinesia and tremor and effects of medications were observed over a day in all patients with variability across patients, and within patients from one day to another, as schematically illustrated in Fig. 3

[0017] ,

[0077] Moreover, different activities (e.g., sleeping, standing, walking, running, cycling, eating, drinking, driving, or the like) are carried out by the patient during the day.

[0078] Open-loop therapeutic approaches are unable to properly consider changes in symptoms and / or side effects and / or activities that the patient may face over time, even on a daily basis. Open-loop approaches also have several drawbacks, e.g., in terms of tolerability-related issues, reduction of the therapeutic effect over time, and battery depletion.

[0079] More sophisticated, dynamic neurostimulation therapies (so-called “adaptive” or “closed-loop”) have been developed and demonstrated in literature and previous art for PD

[0018] , and pain

[0019] .

[0080] However, these known therapeutic approaches are still quite limited because they typically only modulate a single therapy parameter (usually, amplitude) in response to a single biomarker (e.g., beta-band power) describing a single symptom or an evoked response elicited by applied stimulation

[0081] However, patients experience a multitude of symptoms, evolving over time.

[0082] Symptoms may Vary depending, e.g., on the activity that is carried out by the patient, the time of the day, and time passed since the last medication intake.

[0083] Also, signal biomarkers (e.g., from brain sensors) may change over time based on activities that are carried out by the patient, making it difficult to discriminate changes due to pathologic symptoms from normal activity variation

[0020] ,

[0084] Other therapeutic approaches are based on reinforcement learning (RL).

[0085] Solutions of this kind are described, e.g., in EP3341073B1 and US20230191130A1.

[0086] However, these known approaches only contemplate characterization of certain predefined brain activities, and relating them to respective treatment effects.

[0087] So far, capability to fully observe the brain remains very limited. In the absence of a more complete observation of the brain or the definition of a context that constrains the interpretation space, discriminating pathologic symptoms from normal changes in the patient’s activities is challenging.

[0088] Thus, there is the need for an improved solution that enables to provide a more personalized therapeutic approach, duly considering the patient’s daily activities, symptoms, preferences, side effects, or the like.

[0089] In the light of the above, it is an object of the present invention to provide a system for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient in a more personalized manner, thereby enhancing therapeutic outcome. According to the invention, a system for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient comprises: a sensing unit comprising a first sensor means for detecting brain signals of the patient, and a second sensor means for detecting behavioral signals of the patient; a stimulation unit comprising a stimulation means configured to deliver neuromodulation / neurostimulation to brain structures of the patient, and a control unit, operatively connected to the sensing unit and the stimulation unit, and configured to control operation of the sensing unit and the stimulation unit, wherein the control unit comprises a control means that is configured to operate a software means provided therein to: control the sensing unit to sense brain and behavioral signals of the patient; route the sensed signals to a supervisor policy, the supervisor policy being configured to: determine a current activity of the patient based on the sensed signals from the sensing unit, and select an expert policy among a plurality of expert policies, corresponding to the determined current activity of the patient, route the sensed signals to the selected expert policy, the expert policy being configured to: based on the sensed signals, determine a current status of the patient, and based on the determined status of the patient, select or define stimulation parameters for the patient, control the stimulation unit to deliver neuromodulation / neurostimulation to brain structures of the patient based on the stimulation parameters selected or defined by the expert policy, and control the sensing unit to sense updated brain and behavioral signals of the patient after delivery of neuromodulation / neurostimulation.

[0090] The present invention provides a system for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient in need thereof.

[0091] As a non-limiting example, the patient may be a PD subject.

[0092] According to a further, non-limiting example, the patient may be a subject affected by a psychiatric disorder, e.g., OCD. The system comprises a sensing unit.

[0093] The sensing unit comprises a first sensor means for detecting brain signals of the patient.

[0094] The sensing unit also comprises a second sensor means for detecting behavioral signals of the patient.

[0095] The system further comprises a stimulation unit.

[0096] The stimulation unit comprises a stimulation means.

[0097] In particular, the stimulation means is configured to deliver neuromodulation / neurostimulation to brain structures of the patient.

[0098] As a non-limiting example, said stimulation means may be configured to deliver DBS to a target brain tissue of the patient.

[0099] The system further comprises a control unit.

[0100] The control unit is operatively connected to the sensing unit and the stimulation unit.

[0101] In particular, the control unit is configured to control operation of the sensing unit and the stimulation unit.

[0102] The control unit comprises a control means.

[0103] In particular, said control means is configured to operate a software means to: control the sensing unit to sense brain and behavioral signals of the patient; route the sensed signals to a supervisor policy, the supervisor policy being configured to: determine a current activity of the patient based on the sensed signals from the sensing unit, and select an expert policy among a plurality of expert policies, corresponding to the determined current activity of the patient, route the sensed signals to the selected expert policy, the expert policy being configured to: based on the sensed signals, determine a current status of the patient, and based on the determined status of the patient, select or define stimulation parameters for the patient, control the stimulation unit to deliver neuromodulation / neurostimulation to brain structures of the patient based on the stimulation parameters selected or defined by the expert policy, and control the sensing unit to sense updated brain and behavioral signals of the patient after delivery of neuromodulation / neurostimulation.

[0104] As a non-limiting example, the plurality of expert policies may include a walk expert policy, a sleep expert policy, and a speak expert policy.

[0105] The definition of different / additional expert policies is also possible according to the invention.

[0106] As a non-limiting example, the patient’s status can be determined based on a determined physical and / or psychological status of the patient, pathological symptoms, side-effect increase / decrease, and medication intake. The invention is based on the basic idea that, by the provision of a plurality of expert policies for corresponding activities of the patient (e.g., sleeping, standing, walking, speaking, or the like) capable of selecting / defining stimulation parameters based on a current status of the patient, as well as a supervisor policy capable of detecting a current activity of the patient and selecting a corresponding expert policy, stimulation can be delivered in a more effective and personalized manner, thus improving therapeutic outcome.

[0107] The expert policy may select optimal stimulation parameters from a list, based on a current status of the patient.

[0108] Alternatively, the expert policy may be trained to automatically define optimal stimulation parameters, based on a current status of the patient.

[0109] By considering both brain and behavioral signals, determination of the patient’s current activity and / or status can be carried out with improved accuracy and reliability.

[0110] The system of the invention enables therapies to cover a broad range of pathologies that can be treated with neurostimulation, provides effective therapy to more patients, and longer lasting therapeutic benefit throughout the changes in symptoms and activities / needs that patients go through each day.

[0111] As a non-limiting example, the patient’s activity may comprise one or more among sleeping, standing, walking, running, cycling, speaking, swallowing, reading, swimming, cooking, skiing, sitting, or driving. The control means may be configured to implement an expert policy loop by operating the software means provided therein to iterate the process by routing the updated brain and behavioral signals to the selected expert policy.

[0112] The expert policy is configured to determine, based on the updated signals, if a change occurred in the patient’s status.

[0113] Then, upon determining that a change occurred in the patient’s status, the expert policy is further configured to update stimulation parameters.

[0114] Accordingly, stimulation parameters can be promptly adjusted based on a current status of the patient.

[0115] Therefore, delivered stimulation can be optimized.

[0116] Conveniently, the control means may be configured to operate said software means to iterate the expert policy loop every 100 milliseconds to 5 minutes seconds.

[0117] Preferably, the control means may be configured to operate said software means to iterate the expert policy loop every 100 milliseconds to 1 minute.

[0118] More preferably, the control means may be configured to operate said software means to iterate the expert policy loop every 100 milliseconds to 10 seconds.

[0119] Even more preferably, the control means is configured to operate said software means to iterate the expert policy loop every 10 seconds.

[0120] Accordingly, stimulation parameters can be updated by the selected expert policy with sufficient frequency to ensure a feeling for the patient that is close to natural.

[0121] The control means may be configured to implement a supervisor policy loop by operating the software means provided therein to iterate the process by routing the updated brain and behavioral signals to the supervisor policy.

[0122] The supervisor policy is configured to determine, based on the updated signals, if a change occurred in the patient’s activity (e.g., from sleeping to standing).

[0123] Then, upon determining that a change occurred in the patient’s activity, the supervisor policy is further configured to select a different expert policy among the plurality of expert policies, corresponding to the current activity of the patient. Patient’s activities such as sleeping, standing, walking, running, or the like may frequently change over the day.

[0124] By implementing the aforesaid supervisor policy loop, changes in the patient’s activity over time can be promptly detected, and a different expert policy can be selected by the supervisor policy upon detecting that the patient’s activity has changed.

[0125] This provides the patient with a feeling that is close to natural.

[0126] Conveniently, the control means may be configured to operate said software means to iterate the supervisor policy loop every 1 to 30 minutes.

[0127] Preferably, the control means may be configured to operate said software means to iterate the supervisor policy loop every 5 to 15 minutes.

[0128] More preferably, the control means may be configured to operate said software means to iterate the supervisor policy loop every 5 to 10 minutes.

[0129] Even more preferably, the control means is configured to operate said software means to iterate the supervisor policy loop every 10 minutes.

[0130] In principle, the supervisor and expert policy loops do not need to run sequentially.

[0131] For example, a sleep policy may be set to provide very little stimulation, so as not to disturb sleep cycles of the patient.

[0132] When the patient is sleeping (i.e., the sleep expert policy is selected and operated), the supervisor policy may be trained to check the patient’s activity, e.g., every 10 minutes.

[0133] In fact, once asleep, the patient’s activity does not change often. Therefore, triggering the supervisor loop to check for possible changes in the patient’s activity too often (e.g., in the range of a few seconds) is not necessary.

[0134] At the same time, the sleep expert policy may be trained to check the current patient’s status every 10 seconds, thereby ensuring optimization of delivered stimulation based on the current patient status during sleep.

[0135] At a certain moment, the patient may wake up and decide to walk.

[0136] Accordingly, the patient’s activity changes from “sleeping” to “walking”. Within a 10-minute period, the supervisor policy detects this change, and switches from the sleep expert policy to the walk expert policy.

[0137] Preventing unnecessary triggering of the supervisor policy loop (e.g., by operating said loop too frequently when not required) is also beneficial to reduce power consumption.

[0138] Optionally, when a change in the patient’s activity (e.g., from “sleeping” to “walking”) is detected by the supervisor policy based on behavioral signals from the second sensor means, the control means may be further configured to trigger an additional supervisor policy loop for obtaining a confirmation on the detected activity change.

[0139] Then, if the detected activity change is confirmed, the expert policy corresponding to the current activity (e.g., “walking”) is selected by the supervisor policy.

[0140] A hierarchy among sensors and / or threshold values can be defined for determining / confirming that a change occurred in the patient’s activity.

[0141] Conveniently the control means may comprise an implantable microprocessor configured to run a software means embedding the plurality of expert policies.

[0142] In principle, expert policies need to be small in size and reactive in quasi-real-time.

[0143] Accordingly, embedding expert policies into microprocessor, in particular an implantable microprocessor, could be convenient.

[0144] The control means may further comprise an external control device configured to run software means embedding the supervisor policy.

[0145] For instance, said external control device can be a smartphone, a tablet computer, or the like.

[0146] The implantable microprocessor and the external control device are operatively connected to each other.

[0147] The separation between supervisor and expert policies, embedded in separate devices, enables to break up the therapy in several smaller pieces of therapy.

[0148] Each expert policy takes care of its smaller therapeutic region for a particular patient activity, thereby allowing to achieve a high level of therapeutic optimization and personalization. The expert policies and the supervisor policies may run asynchronously, which is particularly advantageous in those cases where the supervisor policy loop does not need to be operated so frequently (e.g., when the patient is asleep).

[0149] The supervisor policy and the expert policies may be trained to ensure that stimulation is delivered according to optimized and personalized stimulation parameters.

[0150] Accordingly, the control means is configured to implement a learning process for the supervisor policy and / or the expert policies.

[0151] Here, the system further comprises a reward input means.

[0152] The reward input means is operatively connected to the control means of the control unit.

[0153] The reward input means is configured to input a reward including a patient’s feedback on the applied neuromodulation / neurostimulation.

[0154] In particular, the control means may be configured to operate the software means provided therein to: determine an initial patient’s condition; operate the stimulation unit to deliver neuromodulation / neurostimulation to brain structures of the patient according to predefined baseline stimulation parameters set for a selected expert policy; control the sensing unit to sense brain and behavioral signals of the patient after delivery of neuromodulation / neurostimulation; acquire and compute a reward, provided through said reward input means after delivery of neuromodulation / neurostimulation; update rules for the supervisor policy and / or the selected expert policy based on the sensed brain and behavioral signals and the acquired reward, and iterate the learning process.

[0155] Rules for the supervisor policy and / or the expert policies are updated after each iteration.

[0156] Therefore, their capability of associating a detected brain state with a corresponding activity and / or status of the patient is improved. In particular, the supervisor policy becomes more effective in selecting the most appropriate expert policy among the plurality of expert policies based on the patient’s current activity.

[0157] Similarly, expert policies specialize in interpreting pathological signals in the context of the activity that the patient is performing.

[0158] In particular, the learning process increases the capacity of each expert policy to specialize in detecting what is pathological in the context of a particular daily activity, so that they are enabled to select / define optimized stimulation parameters based on the current patient’s status and preferences.

[0159] The patient is proactively engaged during the learning process.

[0160] Also, the patient’s feedback is crucial in order to properly update the rules for the supervisor policy and / or the expert policies.

[0161] Patients can voluntarily decide to spend more time in this learning phase to optimize the rules, so that a more effective and personalized therapy is rendered possible.

[0162] Patients may also decide to go back to the learning phase at a later stage, to fine tune previously learned rules for the supervisor policy and / or the expert policies.

[0163] During the training process, policies may be trained through reinforcement learning and supervised and unsupervised learning.

[0164] The supervisor policy can be trained independently from the expert policies using supervised and unsupervised learning.

[0165] Alternatively, the supervisor policy and the expert policies can be trained together, using reinforcement learning methods in a multi-actor version.

[0166] Expert policies are trained only while the patient is doing a particular activity.

[0167] In this way, they learn to contextualize interpretation of brain and behavioral signals to a particular patient activity context.

[0168] This simplifies task learning and leads to models that are smaller and simple in size and complexity.

[0169] This also enables models to be more data-efficient, and specialize in delivering optimal stimulation in the context of the specific task, as reflected in the reward. For example, while sleeping, a patient might care to sleep better rather than not presenting certain symptoms that are not really disrupting sleep.

[0170] In this case, the patient might prioritize quality of sleep over other parameters.

[0171] It is thus determined that, during sleep, delivery of reduced stimulation could be preferable.

[0172] By contrast, while walking, the patient might need gait-locked high intensity dynamic stimulation.

[0173] It is thus determined that, while the patient is walking, appropriate stimulation is required.

[0174] To implement the learning process, a task is designed for each specific activity.

[0175] For example, to define / update rules for the sleep expert policy, a patient may indicate (e.g., through a user input device such as a smartphone, tablet computer, or the like) that he / she is going to sleep, and introduce a score rating sleeping quality the next morning. Such quality sleeping could be subjectively determined by the patient and objectively determined using behavioral and brain signals.

[0176] For a walking activity, sensor means such as a camera may be used to capture a video of the patient while walking, and subsequently feed rewards to the system, e.g., in terms of gait speed, cadence, amplitude, and / or other relevant gait features.

[0177] Both recordings and stimulation need to happen synchronously if a metric is derived from video and / or life sensing measurements, or a metric for a reward reflecting the effect of delivered stimulation during a predefined period needs to be provided.

[0178] The patient’s feedback included in the reward may refer to one or more among a physical and / or psychological status of the patient, patient’s symptoms, and side-effect increase / decrease.

[0179] Rewards may also refer to system performance metrics such as energy consumption, memory footprint of the state space representing the patient status or policies themselves, and refreshing rates for each of the policies impacting in the system behavior.

[0180] Accordingly, rules for the supervisor policy and / or the expert policies can be trained not only to improve therapeutic outcome for the patient, but also in order to optimize system performance metrics, such as battery usage and / or refresh rate for the supervisor policy and / or the expert policies (e.g., to set a proper time lapse between a supervisor or expert policy loop and the next one).

[0181] For example, the supervisor policy may learn that, during a sleeping phase, checking if the patient is still asleep every 10 minutes is sufficient, while the sleep expert policy may learn that updates of stimulation parameters every 10 seconds is enough to ensure good sleep quality.

[0182] Rules for the supervisor policy may also be set so that the supervisor policy loop is triggered more frequently when the patient performs activities (e.g., walking or running) demanding a closer follow-up.

[0183] During both training and stimulation, safety for the patient is ensured by the provision of limits for each stimulation parameter, e.g., set by a clinician during programming and / or follow-up optimization sessions.

[0184] The reward input means may include a user input device.

[0185] Conveniently, said user input device may comprise a display means of the touch-screen type, allowing the patient to easily and intuitively provide a feedback regarding the applied therapy.

[0186] As a non-limiting example, the user input device can be a smartphone, tablet computer, or the like.

[0187] Additionally or alternatively, the user input device may include a sensor means.

[0188] Conveniently, said sensor means may include one or more among a camera, an implantable or non-implantable accelerometer, and a microphone.

[0189] For example, a camera may be used to record a video of the patient during walking, so that a reward is fed to the system accordingly.

[0190] Transition from a current expert policy (e.g., a sleep expert policy) to a different one (e.g., a walk expert policy) shall not be abrupt as this may cause inconvenience for the patient.

[0191] Therefore, the control means may be configured to operate the software means provided therein to detect a transition phase from a current expert policy (e.g., a sleep expert policy) to a different expert policy (e.g., standing or walking).

[0192] During the transition phase, the software means are operated by the control means to control the stimulation unit to deliver neuromodulation / neurostimulation according to stimulation parameters defined for a generalized policy. Accordingly, smooth transitioning from an expert policy to a different one can be accomplished.

[0193] In principle, several expert policies are set for corresponding patient’s activities.

[0194] Nevertheless, it might be the case that a detected activity of the patient does not correspond to any of the plurality of expert policies.

[0195] Even in such a case, however, it is important that stimulation therapy is delivered (although in a less-personalized fashion) without disruptions that may cause inconvenience for the patient.

[0196] Accordingly, upon detecting that the current patient’s activity does not correspond to any among the plurality of expert policies, the software means of the control means may be operated to control the stimulation unit to deliver neuromodulation / neurostimulation according to stimulation parameters set for a predefined basal rule.

[0197] Otherwise stated, the predefined basal rule configures a fallback position that ensures continuity in the delivered therapy (although in a less-personalized fashion) in the absence of an expert policy corresponding to the activity that is currently carried out by the patient.

[0198] Conveniently, the first sensor means may comprise a high-density implantable neural interface.

[0199] Preferably, said implantable neural interface comprises at least two electrodes per cm2.

[0200] More preferably, said implantable neural interface comprises at least four electrodes per cm2.

[0201] Preferably, the second sensor means may comprise one or more among an implantable or non-implantable accelerometer, a microphone, a temperature sensor, and / or other wearable sensor means.

[0202] Conveniently, the stimulation means may comprise a high-density implantable neural interface.

[0203] Preferably, said implantable neural interface comprises at least two electrodes per cm2.

[0204] More preferably, said implantable neural interface comprises at least four electrodes per cm2.

[0205] The neural interface may be configured to be implanted cortically.

[0206] Alternatively, the neural interface may be configured to be implanted subcortically. Conveniently, the neural interface may be configured to deliver neuromodulation / neurostimulation and sense brain signals of the patient.

[0207] That is, the same neural interface can be used for both delivering neuromodulation / neurostimulation and sensing brain signals.

[0208] The neural interface may comprise a support.

[0209] For example, the support may at least partially be and / or comprise a thin film and / or be made of a nanostructure and / or a microstructure.

[0210] In particular, the term thin film is to be understood as a stack of layers of materials ranging from fractions of a nanometer to several micrometers in thickness.

[0211] Thin film layers can be obtained applying a variety of thin-film deposition / preparation techniques including sputtering, evaporation, chemical vapor deposition, molecular beam epitaxy, atomic layer deposition, spin coating, 3D printing, or the like.

[0212] The stack of layers exhibits a variety of functionalities, such as conductive layers, insulating layers, passivating layers, or the like.

[0213] The support, e.g., a thin film, may include and / or carry an electrode array comprising a plurality of electrodes.

[0214] The electrodes can be made of reduced graphene oxide (rGO).

[0215] Also, each electrode of said plurality of electrodes may be characterized by an area in the micrometer range.

[0216] Conveniently, the area of each electrode may be comprised between 1.0 and 0.0005 mm2.

[0217] Conveniently, the electrode array may comprise at least 20 electrodes.

[0218] For example, the electrode array may include 40 electrodes.

[0219] Alternatively, the electrode array may include 50 electrodes.

[0220] Alternatively, the electrode array may include from 128 to 1024 electrodes.

[0221] As mentioned, the provision of micrometer-range rGO electrodes allows achieving precise stimulation while allowing decoding of neural signals with high signal-to-noise ratio.

[0222] Conveniently, the electrode array may comprise a plurality of electrodes having different areas in the micrometer range.

[0223] For example, one or more electrodes having a smaller area can be used for sensing smaller areas targeting single- and multi-unit activity, optimizing SNR and enlarging bandwidth. On the other hand, one or more electrodes having a larger area can be used for a more extended activation area targeting larger groups of neurons and measure, e.g., LFPs.

[0224] Also, electrodes having different areas can be selected depending on characteristics of the stimulation target, spatial selectivity, charge injection, current density, required compliance voltage, and / or impedance.

[0225] Conveniently, the thin film can be made of a flexible material.

[0226] Preferably, said flexible material may include metal or a metal alloy.

[0227] Additionally or alternatively, said flexible material may include a polymer material.

[0228] Conveniently, said polymer material may include one among Polylmide, Parylene, or Polydimethylsiloxane (PDMS).

[0229] The neural interface may allow sensing a large variety of brain signals over time.

[0230] For instance, said brain signals may include brain oscillations (e.g., beta, gamma, HFO).

[0231] Additionally or alternatively, said brain signals may include evoked responses to subcortical stimulation.

[0232] These recordings are characterized by a high resolution, both spatially and temporally.

[0233] Spatial high resolution can be obtained through the provision of a quite large number of rGO electrodes, arranged to form a high-density neural interface allowing recording / sensing activity of different locations nearby the implantation site.

[0234] On the other hand, temporal high resolution is mostly enabled by the properties of graphene.

[0235] In particular, rGO electrodes in the micrometer scale having low impedance allow for a high signal-to-noise ratio, enabling recording / sensing of local neural activity at both low and high frequencies (with supporting sensing electronics).

[0236] A method of delivery of closed-loop neuromodulation / neurostimulation therapy to brain structures of a patient may be implemented by using the system described above.

[0237] An exemplary method carried out by using the above-described system may comprise: controlling the sensing unit to sense brain and behavioral signals of the patient; routing the sensed signals to a supervisor policy to: determine a current activity of the patient based on the sensed signals from the sensing unit, and select an expert policy among a plurality of expert policies, corresponding to the determined current activity of the patient, routing the sensed signals to the selected expert policy to: based on the sensed signals, determine a current status of the patient, and based on the determined status of the patient, select or define stimulation parameters for the patient, controlling the stimulation unit to deliver neuromodulation / neurostimulation to brain structures of the patient based on the stimulation parameters selected or defined by the expert policy, and controlling the sensing unit to sense updated brain and behavioral signals of the patient after delivery of neuromodulation / neurostimulation.

[0238] Conveniently, said exemplary method may be further adapted to implement a learning process, comprising: determining an initial patient’s condition; operating the stimulation unit to deliver neuromodulation / neurostimulation to brain structures of the patient according to predefined baseline stimulation parameters set for a selected expert policy; controlling the sensing unit to sense brain and behavioral signals of the patient after delivery of neuromodulation / neurostimulation; acquiring and compute a reward, provided through said reward input means after delivery of neuromodulation / neurostimulation; updating rules for the supervisor policy and / or the selected expert policy based on the sensed brain and behavioral signals and the acquired reward, and iterating the learning process.

[0239] Also, a computer-readable non-volatile storage medium can be provided, comprising computer-readable instructions that, when executed by a processing means, causes said processing means to implement and / or process and / or perform the method steps defined above.

[0240] Further details and advantages of the present invention shall now be disclosed in connection with the drawings, where. Fig. 1 is a diagram showing the results of the clinical trial carried out on a plurality of PD patients, discussed in Vitek et al.

[0010] , Reproduced from Vitek et al.

[0010] ;

[0241] Fig. 2 is a diagram illustrating a method of identifying DBS connectivity. Processing steps include: A. acquiring pre- / postoperative imaging; B. localizing DBS electrodes in standard space; C. calculating the volume of tissue activated (VTA) based on stimulation parameters; D. calculating functional and E. structural connectivity from the VTA to the rest of the brain using high-quality normative connectome data. The processing stream using the connectome datasets defined in healthy subjects is shown and was used in all primary analyses. Reproduced from Horn et. al. (2017)

[0015] ;

[0242] Fig. 3 is a diagram showing smartwatch symptom profiles (% per 15-min window) aggregated across multiple days according to medication schedules for a given subject. Upper plot: Higher tremor levels were observed in the morning and mid-afternoon, which matched patient-reported "off1times. Lower plot: A reduction in the amount of tremor was seen during the patient’s mid-afternoon “off" time on the new medication schedule (day 20 onward). Reproduced from Powers et al. (2021)

[0017] ;

[0243] Fig. 4 is a diagram illustrating a system for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient according to an embodiment of the invention;

[0244] Fig. 5 is a diagram showing delivery of neuromodulation / neurostimulation therapy to brain structures of a patient using the system of Fig. 4;

[0245] Fig. 6 is a diagram showing a learning process for a supervisor policy and / or a selected expert policy, in the system of Fig. 4;

[0246] Fig. 7 is a diagram showing how delivered neuromodulation / neurostimulation therapy can be personalized in the system of Fig. 4.

[0247] Fig. 4 schematically illustrates a system 100 for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient in need thereof.

[0248] As a non-limiting example, the patient may be a PD subject.

[0249] According to a further, non-limiting example, the patient may be a subject affected by a psychiatric disorder, e.g., OCD.

[0250] The system 100 comprises a sensing unit 10.

[0251] The sensing unit 10 comprises a first sensor means 12 for detecting brain signals of the patient. In the present embodiment, said first sensor means 12 comprises a high-density implantable neural interface.

[0252] Conveniently, the implantable neural interface comprises at least two, more preferably at least four, electrode contacts per cm2.

[0253] The sensing unit 10 further comprises a second sensor means 14 for detecting behavioral signals of the patient.

[0254] In the present embodiment, the second sensor means 14 comprises one or more among an implantable or non-implantable accelerometer, a microphone, a temperature sensor, and / or other wearable sensor means.

[0255] Different configurations for the first sensor means 12 and / or the second sensor means 14 are also possible according to the invention.

[0256] The system 100 further comprises a stimulation unit 16.

[0257] The stimulation unit 16 comprises a stimulation means 18 configured to deliver neuromodulation / neurostimulation to brain structures of the patient.

[0258] As a non-limiting example, the stimulation means 18 may be configured to deliver DBS.

[0259] Other configurations for the stimulation means 18 are also possible according to the invention.

[0260] In the present embodiment, the stimulation means 18 comprises a high-density implantable neural interface.

[0261] Conveniently, the implantable neural interface comprises at least two, more preferably at least four, electrode contacts per cm2.

[0262] The neural interface is configured to be implanted cortically or subcortically.

[0263] The neural interface may be configured to deliver neuromodulation / neurostimulation and sense brain signals of the patient.

[0264] That is, the same neural interface can be used for both stimulation and brain signals recording.

[0265] The system 100 further comprises a control unit 20.

[0266] The control unit 20 is operatively connected to the sensing unit 10 and the stimulation unit 16.

[0267] The control unit is configured to control operation of the sensing unit 10 and the stimulation unit 16.

[0268] The control unit 20 comprises a control means 22. Fig. 5 shows how the control means 22 of the system 100 are configured to operate software means provided therein to deliver neuromodulation / neurostimulation to a patient.

[0269] In particular, the control means 22 is configured to operate software means provided therein to: control the sensing unit 10 to sense brain and behavioral signals s_t of the patient; route the sensed signals s_t to a supervisor policy, the supervisor policy being configured to: determine a current activity of the patient based on the sensed signals s_t from the sensing unit 10, and select an expert policy e_t among a plurality of expert policies e_t, corresponding to the determined current activity of the patient, route the sensed signals to the selected expert policy e_t, the expert policy e_t being configured to: based on the sensed signals s_t, determine a current status of the patient, and based on the determined status of the patient, select or define stimulation parameters a_t for the patient, control the stimulation unit 16 to deliver neuromodulation / neurostimulation to brain structures of the patient based on the stimulation parameters a_t selected or defined by the expert policy e_t, and control the sensing unit 10 to sense updated brain and behavioral signals s_t+1 of the patient after delivery of neuromodulation / neurostimulation.

[0270] Three expert policies, namely a walk expert policy, a sleep expert policy, and a speak expert policy, are shown in Fig. 5 for simplification.

[0271] Nevertheless, a higher number of expert policies may be defined for corresponding activities of the patient.

[0272] By the provision of a sufficient number of expert policies, covering the primary daily activities of the patient, stimulation therapy can be delivered with increased precision and effectiveness.

[0273] As a non-limiting example, the patient activity may include one or more among sleeping, standing, walking, running, cycling, speaking, swallowing, reading, swimming, cooking, skiing, sitting, or driving. As a non-limiting example, the patient’s status can be determined based on a determined physical and / or psychological status of the patient, pathological symptoms, side-effect increase / decrease, and medication intake.

[0274] As shown in Fig. 5, the control means 22 may be adapted to implement two different loops during operation of the system 100, namely a supervisor policy loop and an expert policy loop.

[0275] Conveniently, in the present embodiment, the supervisor policy loop and the expert policy loop do not need to run sequentially.

[0276] In the present embodiment, the control means 22 is configured to implement an expert policy loop by operating the software means provided therein to: iterate the process t+1 by routing the updated brain and behavioral signals s_t+1 to the selected expert policy e_t, wherein the expert policy e_t is configured to: determine, based on the updated signals s_t+1 , if a change occurred in the patient’s status, and upon determining that a change occurred in the patient’s status, update stimulation parameters a_t for the patient.

[0277] Here, updated brain and behavioral signals s_t+1 are directly routed to the selected expert policy (e.g., a walk expert policy, as shown in Fig. 5), without being routed to the supervisor policy.

[0278] In the present embodiment, the control means 22 may be configured to operate said software means to iterate the expert policy loop every 100 milliseconds to 5 minutes.

[0279] Preferably, the control means 22 may be configured to operate said software means to iterate the expert policy loop every 100 milliseconds to 1 minute.

[0280] More preferably, the control means 22 may be configured to operate said software means to iterate the expert policy loop every 100 milliseconds to 10 seconds.

[0281] Even more preferably, the expert policy loop is iterated every 10 seconds.

[0282] Accordingly, stimulation parameters can be promptly refreshed in line with the current status of the patient.

[0283] The control means 22 is further configured to implement a supervisor policy loop by operating the software means provided therein to: iterate the process t+1 by routing the updated brain and behavioral signals s_t+1 to the supervisor policy, wherein the supervisor policy is configured to: determine, based on the updated signals s_t+1 , if a change occurred in the patient’s activity, and upon determining that a change occurred in the patient’s activity, select a different expert policy (e_t) among the plurality of expert policies e_t, corresponding to the current activity of the patient.

[0284] Here, updated brain and behavioral signals s_t+1 are routed to the supervisor policy, so that the supervisor policy can check if a change occurred in the patient’s activity (e.g., from sleeping to walking, or vice versa).

[0285] If so, the supervisor policy selects a different expert policy, corresponding to the current activity of the patient.

[0286] In the present embodiment, the control means 22 may be configured to operate said software means to iterate the supervisor policy loop every 1 to 30 minutes.

[0287] Preferably, the control means 22 may be configured to operate said software means to iterate the supervisor policy loop every 5 to 15 minutes.

[0288] More preferably, the control means 22 may be configured to operate said software means to iterate the supervisor policy loop every 5 to 10 minutes.

[0289] Even more preferably, the supervisor policy loop is iterated every 10 minutes.

[0290] For example, when the patient is sleeping, checking whether the patient is still asleep every 10 minutes might be sufficient, while the sleep expert policy may more frequently check the patient’s current status, e.g., every 10 seconds.

[0291] More frequent iterations of the supervisor policy can be provided in case of activities (e.g., walking or running) demanding a closer follow-up.

[0292] Not shown is that, in the present embodiment the control means 22 comprises an implantable microprocessor configured to run a software means embedding the plurality of expert policies e_t.

[0293] Also not shown is that, in the present embodiment, the control means 22 further comprises an external control device configured to run software means embedding the supervisor policy.

[0294] For instance, said external control device can be a smartphone, a tablet computer, or the like. The implantable microprocessor and the external control device are operatively connected to each other.

[0295] In the present embodiment, the control means 22 is further configured to control the software means provided therein to implement a learning process for the supervisor policy or the expert policies, as shown in Figs. 6-7.

[0296] Here, the system 100 further comprises a reward input means 24, operatively connected to the control means 22 of the control unit 20, allowing the patient to input a reward r_t comprising a feedback on the applied neuromodulation / neurostimulation a_t.

[0297] To implement the learning process, the control means 22 is configured to control the software means to: determine an initial patient’s condition St; operate the stimulation unit 16 to deliver neuromodulation / neurostimulation a_t to brain structures of the patient according to predefined baseline stimulation parameters set for a selected expert policy e_t; control the sensing unit 10 to sense brain and behavioral signals s_t of the patient after delivery of neuromodulation / neurostimulation a_t; acquire and compute a reward r_t, provided through said reward input means 24 after delivery of neuromodulation / neurostimulation a_t; update rules for the supervisor policy and / or the selected expert policy based on the sensed brain and behavioral signals s_t and the acquired reward r_t, and iterate the learning process.

[0298] For instance, an algorithm may be set to evaluate symptoms and / or side effects based on the patient’s feedback, and compute a reward r_t accordingly (Fig. 6).

[0299] The computed reward is used to train the supervisor policy and / or the expert policies.

[0300] Accordingly, stimulation can be delivered in a highly-personalized fashion, thus improving therapeutic outcome.

[0301] The patient may decide to go back the learning phase at any time to fine-tune policy rules.

[0302] The diagram on Fig. 7 schematically shows how stimulation therapy is personalized in the system 100 of the invention.

[0303] For example, as shown in Fig. 7, behavioral signals are acquired together with the patient’s feedback. Simultaneously, patient’s symptoms and side effects are evaluated to compute reward.

[0304] An algorithm learns to associate the best stimulation parameters to the best rewards (e.g., denoting a reduction in symptoms and / or side effects).

[0305] As a result, the supervisor policy is enabled to detect a patient’s activity (e.g., walking, sleeping, or speaking) with enhanced precision and reliability, and select the most appropriate expert policy accordingly. Similarly, the selected expert policy is enabled to select / define stimulation parameter that are most appropriate according to the patient’s current status.

[0306] Not shown is that that reward input means 24 may comprise a user input device.

[0307] Conveniently, said user input device may comprise a display means of the touch-screen type.

[0308] Accordingly, the patient is enabled to provide his / her feedback in a simple and intuitive manner, e.g., by filling-in a questionnaire that is displayed on the screen.

[0309] As a non-limiting example, the patient’s feedback may refer to one or more among a physical and / or psychological status of the patient, patient’s symptoms, and side-effect increase / decrease.

[0310] Also not shown is that, additionally or alternatively, the user input device may comprise a sensor means.

[0311] Conveniently, said sensor means may comprise one or more among a camera, an implantable or non-implantable accelerometer, and a microphone.

[0312] For instance, a camera may be used to capture a video of the patient, e.g., while walking or running.

[0313] As mentioned, transition from a current expert policy (e.g., a sleep expert policy) to a different one (e.g., a walk expert policy) shall not be abrupt as this may cause inconvenience for the patient.

[0314] Accordingly, in the present embodiment, the control means 22 are configured to operate the software means provided therein to detect a transition phase from a current expert policy (e.g., a sleep expert policy) to a different expert policy (e.g., standing or walking).

[0315] During the transition phase, the software means are operated by the control means 22 to control the stimulation unit 16 to deliver neuromodulation / neurostimulation according to stimulation parameters defined for a generalized policy.

[0316] Accordingly, smooth transitioning from an expert policy to a different one can be accomplished. It might be the case that none among the defined expert policies corresponds to the current activity of the patient.

[0317] In such a case, it is nevertheless important that stimulation therapy is delivered (although in a less-personalized fashion) without disruptions that may cause inconvenience for the patient.

[0318] Accordingly, upon detecting that the current patient’s activity does not correspond to any among the plurality of expert policies, the software means of the control means 22 may be operated to control the stimulation unit 16 to deliver neuromodulation / neurostimulation according to stimulation parameters set for a predefined basal rule.

[0319] Otherwise stated, the predefined basal rule configures a fallback position that ensures continuity in the delivered therapy (although in a less-personalized fashion) in the absence of an expert policy corresponding to the activity that is currently carried out by the patient.

[0320] In a comparative example, the system 100 can be used to implement a method of delivery of closed-loop neuromodulation / neurostimulation therapy to brain structures of a patient.

[0321] In particular, said exemplary method may comprise: controlling the sensing unit 10 to sense brain and behavioral signals s_t of the patient; routing the sensed signals s_t to a supervisor policy to: determine a current activity of the patient based on the sensed signals s_t from the sensing unit 10, and select an expert policy e_t among a plurality of expert policies e_t, corresponding to the determined current activity of the patient, routing the sensed signals to the selected expert policy e_t to: based on the sensed signals s_t, determine a current status of the patient, and based on the determined status of the patient, select or define stimulation parameters a_t for the patient, controlling the stimulation unit 16 to deliver neuromodulation / neurostimulation to brain structures of the patient based on the stimulation parameters a_t selected or defined by the expert policy e_t, and control the sensing unit 10 to sense updated brain and behavioral signals s_t+1 of the patient after delivery of neuromodulation / neurostimulation.

[0322] Conveniently, said exemplary method may be further adapted to implement a learning process, comprising: determining an initial patient’s condition St; operating the stimulation unit 16 to deliver neuromodulation / neurostimulation a_t to brain structures of the patient according to predefined baseline stimulation parameters set for a selected expert policy e_t; controlling the sensing unit 10 to sense brain and behavioral signals s_t of the patient after delivery of neuromodulation / neurostimulation a_t; acquiring and compute a reward r_t, provided through said reward input means 24 after delivery of neuromodulation / neurostimulation a_t; updating rules for the supervisor policy and / or the selected expert policy based on the sensed brain and behavioral signals s_t and the acquired reward r_t, and iterate the learning process.

[0323] According to a further comparative example, a computer-readable non-volatile storage medium can be provided, comprising computer-readable instructions that, when executed by a processing means, causes said processing means to implement and / or process and / or perform the method steps defined above.

[0324] Bibliography

[0325] [1] GBD 2016 Parkinson’s Disease Collaborators, (2018). Global, regional, and national burden of Parkinson’s disease, 1990-2016: a systematic analysis for the Global Burden of Disease Study 2016. The Lancet. Neurology, 17(11), 939-953. https: / / doi.org / 10.1016 / S1474- 4422(18)30295-3.

[0326] [2] Bullmore, E., & Sporns, O. (2012). The economy of brain network organization. Nature Reviews Neuroscience, 13(5), 336-349. https: / / doi.org / 10.1038 / nrn3214.

[0327] [3] Kim, J., Criaud, M., Cho, S. S., Diez-Cirarda, M., Mihaescu, A., Coakeley, S., Ghadery, C., Valli, M., Jacobs, M. F., Houle, S., & Strafella, A. P. (2017). Abnormal intrinsic brain functional network dynamics in Parkinson’s disease. Brain. A Journal of Neurology, 140(11), 2955-2967. https: / / doi.org / 10.1093 / brain / awx233.

[0328] [4] Horn, A., Wenzel, G., Irmen, F., Huebl, J., Li, N., Neumann, W.-J., Krause, P., Bohner, G.,

[0329] Scheel, M., & Kuhn, A. A. (2019). Deep brain stimulation induced normalization of the human functional connectome in Parkinson’s disease. Brain, 142(10), 3129-3143. https: / / doi.Org / 10.1093 / brain / awz239.

[0330] [5] Husain, M. (2019). Targeting network dysfunction in neurodegenerative diseases. Brain, 142(12), 3661-3662. https: / / doi.org / 10.1093 / brain / awz347.

[0331] [6] Chou, K. L. (2008). Adverse events from the treatment of Parkinson’s disease. Neurologic Clinics, 26(3 Suppl), S65-83, vi. https: / / doi.Org / 10.1016 / j.ncl.2008.05.003.

[0332] [7] Williams, N. R., Foote, K. D., & Okun, M. S. (2014). STN vs. GPi Deep Brain Stimulation: Translating the Rematch into Clinical Practice. Movement Disorders Clinical Practice, 1(1), 24-35. https: / / doi.org / 10.1002 / mdc3.12004.

[0333] [8] Bratsos, S. P., Karponis, D., & Saleh, S. N. (2018). Efficacy and Safety of Deep Brain Stimulation in the Treatment of Parkinson’s Disease: A Systematic Review and Meta-analysis of Randomized Controlled Trials. Cureus. https: / / doi.org / 10.7759 / cureus.3474.

[0334] [9] Vizcarra, J. A., Situ-Kcomt, M., Artusi, C. A., Duker, A. P., Lopiano, L., Okun, M. S., Espay, A. J., & Merola, A. (2019). Subthalamic deep brain stimulation and levodopa in Parkinson’s disease: a meta-analysis of combined effects. Journal of Neurology, 266(2), 289-297. https : / / do i . org / 10.1007 / s00415-018-8936-2.

[0335]

[0010] Vitek, J. L., Jain, R., Chen, L., Trdster, A. I., Schrock, L. E., House, P. A., Giroux, M. L., Hebb, A. O., Farris, S. M., Whiting, D. M., Leichliter, T. A., Ostrem, J. L., San Luciano, M., Galifianakis, N., Verhagen Metman, L., Sani, S., Karl, J. A., Siddiqui, M. S., Tatter, S. B., ... Starr, P. A. (2020). Subthalamic nucleus deep brain stimulation with a multiple independent constant current-controlled device in Parkinson’s disease (INTREPID): a multicentre, doubleblind, randomised, sham-controlled study. The Lancet. Neurology, 19(6), 491-501. https: / / doi.Org / 10.1016 / S1474-4422(20)30108-3.

[0336]

[0011] Lyons, K. E., & Pahwa, R. (2006). Effects of bilateral subthalamic nucleus stimulation on sleep, daytime sleepiness, and early morning dystonia in patients with Parkinson disease. Journal of Neurosurgery, 104(4), 502-505. https: / / doi.Org / 10.3171 / jns.2006.104.4.502.

[0337]

[0012] Appleby, B. S., Duggan, P. S., Regenberg, A., & Rabins, P. V. (2007). Psychiatric and neuropsychiatric adverse events associated with deep brain stimulation: A meta-analysis of ten years’ experience. Movement Disorders : Official Journal of the Movement Disorder Society, 22(12), 1722-1728. https: / / doi.org / 10.1002 / mds.21551.

[0338]

[0013] Rasiah, N. P., Maheshwary, R., Kwon, C.-S., Bloomstein, J. D., & Girgis, F. (2023). Complications of Deep Brain Stimulation for Parkinson Disease and Relationship between Micro-electrode tracks and hemorrhage: Systematic Review and Meta-Analysis. World Neurosurgery, 171, e8-e23. https: / / doi.Org / 10.1016 / j.wneu.2022.10.034.

[0339]

[0014] Wang, J. T., Wang, A. Y., Psarros, C., & Da Cruz, M. (2014). Rates of revision and device failure in cochlear implant surgery: a 30-year experience. The Laryngoscope, 124(10), 2393- 2399. https: / / d0i.0rg / l 0.1002 / lary.24649.

[0340]

[0015] Horn, A., Reich, M., Vorwerk, J., Li, N., Wenzel, G., Fang, Q., Schmitz-Hubsch, T., Nickl, R., Kupsch, A., Volkmann, J., Kuhn, A. A., & Fox, M. D. (2017). Connectivity Predicts deep brain stimulation outcome in Parkinson disease. Annals of Neurology, 82(1), 67-78. https: / / d0i.0rg / l 0.1002 / ana.24974.

[0341]

[0016] Boutet, A., Madhavan, R., Elias, G. J. B., Joel, S. E., Gramer, R., Ranjan, M., Paramanandam, V., Xu, D., Germann, J., Loh, A., Kalia, S. K., Hodaie, M., Li, B., Prasad, S., Coblentz, A., Munhoz, R. P., Ashe, J., Kucharczyk, W., Fasano, A., & Lozano, A. M. (2021). Predicting optimal deep brain stimulation parameters for Parkinson’s disease using functional MRI and machine learning. Nature Communications, 12(1), 3043.

[0342]

[0017] Powers, R., Etezadi-Amoli, M., Arnold, E. M., Kianian, S., Mance, I., Gibiansky, M., Trietsch, D., Alvarado, A. S., Kretlow, J. D., Herrington, T. M., Brillman, S., Huang, N., Lin, P. T., Pham, H. A., & Ullal, A. V. (2021). Smartwatch inertial sensors continuously monitor real- world motor fluctuations in Parkinson’s disease. Science Translational Medicine, 13(579). https: / / doi.Org / 10.1126 / scitranslmed.abd7865.

[0018] Neumann, W.J., et al., Toward Electrophysiology-Based Intelligent Adaptive Deep Brain Stimulation for Movement Disorders. Neurotherapeutics, 2019. 16(1): p. 105-118.

[0343]

[0019] Mekhail, N.A., et al. , ECAP-controlled closed-loop versus open-loop SCS forthe treatment of chronic pain: 36-month results of the EVOKE blinded randomized clinical trial. Reg Anesth Pain Med, 2024. 49(5): p. 346-354.

[0344]

[0020] Thenaisie, Y., et al., Principles of gait encoding in the subthalamic nucleus of people with Parkinson's disease. Sci Transl Med, 2022. 14(661): p. eabo1800.

[0345] Table 1 - Adverse events following DBS surgery in two studies (Appleby et al.

[0012] and Rasiah et al.

[0013] )

[0346] References

[0347] System

[0348] Sensing unit

[0349] First sensor means

[0350] Second sensor means

[0351] Stimulation unit

[0352] Stimulation means

[0353] Control unit

[0354] Control means

[0355] Reward input means

Claims

Claims1. System (100) for delivery of neuromodulation / neurostimulation therapy to brain structures of a patient, comprising: a sensing unit (10) comprising a first sensor means (12) for detecting brain signals of the patient, and a second sensor means (14) for detecting behavioral signals of the patient; a stimulation unit (16) comprising a stimulation means (18) configured to deliver neuromodulation / neurostimulation to brain structures of the patient, and a control unit (20), operatively connected to the sensing unit (10) and the stimulation unit (16), configured to control operation of the sensing unit (10) and the stimulation unit (16), wherein the control unit (20) comprises a control means (22) that is configured to operate a software means provided therein to: control the sensing unit (10) to sense brain and behavioral signals (s_t) of the patient; route the sensed signals (s_t) to a supervisor policy, the supervisor policy being configured to: determine a current activity of the patient based on the sensed signals (s_t) from the sensing unit (10), and select an expert policy (e_t) among a plurality of expert policies (e_t), corresponding to the determined current activity of the patient, route the sensed signals to the selected expert policy (e_t), the expert policy (e_t) being configured to: based on the sensed signals (s_t), determine a current status of the patient, and based on the determined status of the patient, select or define stimulation parameters (a_t) for the patient,control the stimulation unit (16) to deliver neuromodulation / neurostimulation to brain structures of the patient based on the stimulation parameters (a_t) selected or defined by the expert policy (e_t), and control the sensing unit (10) to sense updated brain and behavioral signals (s_t+1) of the patient after delivery of neuromodulation / neurostimulation.

2. The system (100) according to claim 1, characterized in that the control means (22) is configured to implement an expert policy loop by operating the software means provided therein to: iterate the process (t+1) by routing the updated brain and behavioral signals (s_t+1) to the selected expert policy (e_t), wherein the expert policy (e_t) is configured to: determine, based on the updated signals (s_t+1 ), if a change occurred in the patient’s status, and upon determining that a change occurred in the patient’s status, update stimulation parameters (a_t) for the patient.

3. The system (100) according to claim 2, characterized in that the control means (22) is configured to operate said software means to iterate the expert policy loop every 100 milliseconds to 5 minutes seconds, preferably, 100 milliseconds to 1 minute, more preferably 100 milliseconds to 10 seconds, even more preferably every 10 seconds.

4. The system (100) according to claim 1,characterized in that the control means (22) is configured to implement a supervisor policy loop by operating the software means provided therein to: iterate the process (t+1) by routing the updated brain and behavioral signals (s_t+1) to the supervisor policy, wherein the supervisor policy is configured to: determine, based on the updated signals (s_t+1 ), if a change occurred in the patient’s activity, and upon determining that a change occurred in the patient’s activity, select a different expert policy (e_t) among the plurality of expert policies (e_t), corresponding to the current activity of the patient.

5. The system (100) according to claim 4, characterized in that the control means (22) is configured to operate said software means to iterate the supervisor policy loop every 1 to 30 minutes, preferably every 5 to 15 minutes, more preferably every 5 to 10 minutes, even more preferably every 10 minutes.

6. The system (100) according to any of the preceding claims, characterized in that the control means (22) at least comprises: an implantable microprocessor configured to run a software means embedding the plurality of expert policies (e_t), and an external control device configured to run software means embedding the supervisor policy,wherein the implantable microprocessor and the external control device are operatively connected to each other.

7. The system (100) according to any of the preceding claims, characterized in that the system (100) further comprises a reward input means (24), operatively connected to said control means (22) of the control unit (20), for inputting a reward (r_t) including a patient’s feedback on the applied neuromodulation / neurostimulation (a_t), wherein the control means (22) is configured to implement a learning process by operating the software means provided therein to: determine an initial patient’s condition (St); operate the stimulation unit (16) to deliver neuromodulation / neurostimulation (a_t) to brain structures of the patient according to predefined baseline stimulation parameters set for a selected expert policy (e_t); control the sensing unit (10) to sense brain and behavioral signals (s_t) of the patient after delivery of neuromodulation / neurostimulation (a_t); acquire and compute a reward (r_t), provided through said reward input means (24) after delivery of neuromodulation / neurostimulation (a_t); update rules for the supervisor policy and / or the selected expert policy based on the sensed brain and behavioral signals (s_t) and the acquired reward (r_t), and iterate the learning process.

8. The system (100) according to any one of the preceding claims, characterized in that the control means (22) is configured to operate the software means provided therein todetect a transition phase from a current expert policy to a different expert policy, and during the transition phase, control the stimulation unit (16) to deliver neuromodulation / neurostimulation according to stimulation parameters defined for a generalized policy, to smoothen transition from the current expert policy to the different expert policy.

9. The system (100) according to any of the preceding claims, characterized in that the control means (22) is configured to operate the software means provided therein to: determine that a patient’s current activity does not correspond to any among the plurality of expert policies, and upon determining that the patient’s current activity does not correspond to any expert policy, control the stimulation unit (16) to deliver neuromodulation / neurostimulation according to stimulation parameters set for a predefined basal rule.

10. The system (100) according to any of the preceding claims, characterized in that the first sensor means (12) comprises a high-density implantable neural interface, preferably wherein said implantable neural interface comprises at least two, more preferably at least four, electrode contacts per cm2, and / or the second sensor means (14) comprises one or more among an implantable or nonimplantable accelerometer, a microphone, a temperature sensor, and / or other wearable sensor means.11 . The system (100) according to any of the preceding claims, characterized in that the stimulation means (18) comprises a high-density implantable neural interface, preferably wherein said implantable neural interface comprises at least two, more preferably at least four, electrode contacts per cm2.

12. The system (100) according to claim 10 and / or 11 , characterized in that said neural interface is configured to be implanted cortically or subcortically, preferably wherein said neural interface is configured to deliver neuromodulation / neurostimulation and sense brain signals of the patient.

13. The system (100) according to any of the preceding claims, characterized in that said reward input means (24) comprises a user input device and / or a sensor means, preferably wherein said user input device comprises a display means of the touchscreen type, and / or said sensor means include one or more among a camera, an implantable or non-implantable accelerometer, and a microphone.

14. The system (100) according to claim 7, characterized in that the patient’s feedback included in the reward (r_t) refers to one or more among a physical and / or psychological status of the patient, patient’s symptoms, and side-effect increase / decrease.

15. The system (100) according to any of the preceding claims, characterized in that the patient’s activity comprises one or more among sleeping, standing, walking, running, cycling, speaking, swallowing, reading, swimming, cooking, skiing, sitting or driving.

Images