Closed-loop neural interfaces for pain control

A closed-loop neural interface using machine learning to process signals from multiple brain regions accurately detects and treats chronic pain by integrating sensory and emotional components, offering real-time pain management with minimal side effects.

JP2024541295APending Publication Date: 2024-11-08NEW YORK UNIV
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
JP2024526938
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2021-11-08
Filing Date
2022-11-02
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Current treatments for chronic pain are limited by their inability to account for the precise timing of pain onset, leading to frequent delays, undertreatment, and overtreatment, and existing closed-loop neuromodulation systems face challenges in accurately decoding pain signals due to the complexity of pain processing in multiple brain circuits and the instability of neural recordings.

Method used

A closed-loop neural interface that utilizes machine learning to process neural signals from multiple brain regions, specifically the anterior cingulate cortex and primary somatosensory cortex, to detect pain through local field potentials, and initiates targeted stimulation, such as optical or electrical stimulation, in the prefrontal cortex to manage pain.

Benefits of technology

The system provides accurate, real-time pain detection and treatment with minimal side effects by integrating sensory and emotional aspects of pain, achieving high sensitivity and specificity over extended periods, suitable for both acute and chronic pain conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024541295000001_ABST
    Figure 2024541295000001_ABST
Patent Text Reader

Abstract

A system and method for treating pain includes a plurality of probes implantable in a plurality of brain regions of a patient to detect neural signals including local field potentials of the plurality of brain regions, a processing device that receives the neural signals from the plurality of brain regions of the patient's brain, processes the neural signals, and inputs the processed neural signals to a machine learning pain decoder model configured to indicate pain, and a stimulator that is implantable in a target region of the patient's brain to provide stimulation to the target region based on the indication of pain.
Need to check novelty before this filing date? Find Prior Art

Description

[Background technology]

[0001] [Government funds] This invention was made with government support under Grant No. NIH-R01GM115384 awarded by the National Institutes of Health. The U.S. Government has certain rights in this invention.

[0002] [Priority claim] This application claims priority to U.S. Provisional Application No. 63 / 263,738, entitled "Closed-Loop Neural Interface for Pain Control," filed November 8, 2021, the disclosure of which is incorporated herein by reference.

[0003] Chronic pain is one of the most common sensory disorders, and there is a critical need for non-opioid and non-addictive analgesics. Chronic pain is defined as a discrete pain that is triggered by noxious stimuli or occurs naturally. Current treatments for chronic pain are limited to either planned drug interventions or continuous spinal or peripheral nerve neuromodulation. These treatments do not take into account the precise timing of the onset of the discrete pain episode, resulting in frequent treatment delays, undertreatment, or overtreatment. Conceptually, a closed-loop neuromodulation approach that couples pain detection and treatment at the right time would be optimal for pain management by selectively targeting discrete nociceptive occurrences. However, decoding pain signals remains a very challenging task. Pain has both sensory and emotional aspects, so to decode pain, ideally both sensory and emotional signals must be combined in the brain. However, pain signals are encoded in multiple brain circuits. A second challenge is that most neuroimaging techniques are large in size and expensive, making them impractical for closed-loop use (e.g., MRI). While invasive neural recordings are amenable to mobile and closed-loop applications, they typically do not produce consistent spiking signals that facilitate sensory decoding. Overcoming these challenges is critical to successful and accurate pain decoding.

[0004] Meanwhile, in terms of treatment, deep brain stimulation (DBS) has been used to treat many neuropsychiatric disorders. However, continuous stimulation has various drawbacks, such as desensitization, which reduces efficacy over time, battery consumption, and side effects. Furthermore, a closed-loop DBS system for treating pain has not yet been realized. Summary of the Invention

[0005] The present disclosure relates to a computer-implemented method for detecting and treating chronic pain, the method including receiving neural signals from multiple brain regions via probes implanted in the multiple brain regions of a patient's brain, the neural signals including local field potentials (LFPs) of the multiple brain regions, processing the neural signals, inputting the processed neural signals to a machine learning pain decoder model, determining whether pain is indicated based on the processed neural signals, and if pain is indicated by the pain decoder model, triggering stimulation to a target region of the patient's brain based on the indication of pain.

[0006] In one embodiment, processing the neural signals includes calculating frequency dependent power characteristics of local field potentials in multiple brain regions.

[0007] In one embodiment, determining whether pain is indicated comprises identifying relative changes in neural activity in multiple brain regions.

[0008] In one embodiment, the multiple brain regions include the anterior cingulate cortex and the primary somatosensory cortex.

[0009] In one embodiment, stimulation of the target region of the patient's brain includes light stimulation and electrical stimulation.

[0010] In one embodiment, the target region of the patient's brain includes the prefrontal cortex.

[0011] In one embodiment, the target region of the patient's brain includes one of the primary motor cortex, the anterior cingulate cortex, and / or the periaqueductal gray, and the thalamus.

[0012] In one embodiment, the method further includes training the pain decoder model using a state space model based on spectral features from low gamma (30-50 Hz), high gamma (50-100 Hz), and ultra-high frequency (300-500 Hz) bands.

[0013] The present disclosure further relates to a system for treating pain including a plurality of probes implantable in a plurality of brain regions of a patient to detect neural signals including local field potentials of the plurality of brain regions, a processing device that receives the neural signals from the plurality of brain regions of the patient's brain, processes the neural signals, and inputs the processed neural signals to a machine learning pain decoder model configured to indicate pain, and a stimulator that can be implanted in a target region of the patient's brain and provides stimulation to the target region based on the indication of pain.

[0014] In one embodiment, the processing unit is configured to process the neural signals by calculating frequency dependent power characteristics of local field potentials of multiple brain regions.

[0015] In one embodiment, the pain decoder model is trained to identify relative changes in neural activity in multiple brain regions.

[0016] In one embodiment, the pain decoder model is trained using a state space model based on spectral features from low gamma (30-50 Hz), high gamma (50-100 Hz), and ultra-high frequency (300-500 Hz) bands.

[0017] In one embodiment, the processing unit is configured to initiate activation of the stimulator in response to the indication of pain.

[0018] In one embodiment, the stimulator is configured to provide either light or electrical stimulation to the target area.

[0019] In one embodiment, the stimulator is configured to be implanted in one of the prefrontal cortex, the primary motor cortex, the anterior cingulate cortex, the periaqueductal gray, and the thalamus.

[0020] In one embodiment, the multiple probes are configured to be implanted in multiple brain regions including the anterior cingulate cortex and the primary somatosensory cortex.

[0021] In one embodiment, each of the plurality of probes comprises a silicon probe array.

[0022] In one embodiment, the system further comprises a graphical user interface that displays the LFP signal in real time and provides an option to change the threshold criteria.

[0023] The present disclosure further relates to a non-transitory computer readable storage medium including a set of instructions executable by a processor, which when executed by the processor causes the processor to perform operations including receiving neural signals from a plurality of brain regions of a patient, the neural signals including local field potentials (LFPs) of the plurality of brain regions, calculating frequency dependent power characteristics of the local field potentials of the plurality of brain regions, inputting the power characteristics into a machine learning pain decoder model to identify relative changes in neural activity in the plurality of brain regions to indicate pain, and initiating stimulation of a target brain region based on the indication of pain. [Brief description of the drawings]

[0024] [Figure 1] FIG. 1 shows a schematic diagram of a multi-region LFP-based brain-machine interface (BMI) system according to an exemplary embodiment of the present disclosure.

[0025] [Diagram 2]FIG. 2 shows a schematic diagram of an experiment based on a closed-loop BMI system according to the system of FIG.

[0026] [Diagram 3] Figure 3 shows a schematic diagram of placing optical fibers or stimulating electrodes in the PL-PFC and recording electrodes in the ACC and S1 according to the system in Figure 1.

[0027] [Figure 4] Figure 4 shows an illustration of simultaneous LFP signals in the ACC and S1, with pain-evoked event-related potentials (ERPs) indicated by black triangles.

[0028] [Diagram 5] Figure 5 shows a graph comparing ERP latencies between ACC and S1.

[0029] [Figure 6] FIG. 6 shows a schematic diagram of a model-based unsupervised learning approach designed to decode pain from multi-region LFP signals by the exemplary system of FIG.

[0030] [Figure 7] FIG. 7 shows a schematic diagram of an exemplary embodiment of a closed-loop multi-region LFP-based BMI.

[0031] [Figure 8] FIG. 8 illustrates an exemplary embodiment of a graphical user interface (GUI) of an exemplary BMI.

[0032] [Figure 9] FIG. 9 is a graph showing a comparison of false positive rates based on LFP decoding strategies.

[0033] [Figure 10] FIG. 10 is a graph showing a comparison of false positive rates based on model parameters set at 5-day intervals.

[0034] [Figure 11] FIG. 11 shows a flowchart of a method for detecting and treating pain using a closed-loop multi-region LFP-based BMI system according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0035] The present disclosure may be further understood with reference to the following description and the accompanying drawings. In an exemplary embodiment, a multi-region closed-loop neural interface for pain reception control is described, which uses a pain decoder on neural signals simultaneously obtained from at least two important regions for the treatment of pain to initiate (trigger) stimulation for treatment, for example, in the prefrontal cortex. In one embodiment, the two brain regions include the anterior cingulate cortex (ACC) and the primary somatosensory cortex (S1). The ACC is known to provide an emotional signal for pain, while S1 contains the sensory pain signal. PFC stimulation has been shown to result in pain relief by activating multiple downstream pain modulatory circuits. To decode pain occurrence, an exemplary embodiment records local field potentials (LFPs), which represent subthreshold local neural activity and are stable in chronic electrophysiological recordings, facilitating clinical application. An unsupervised machine learning pain decoder is model-based and trained to decode pain from multi-region LFP signals.

[0036] In an exemplary embodiment, the pain decoder specifically focuses on spectral features from low gamma (30-50 Hz), high gamma (50-100 Hz), and ultra-high frequency (300-500 Hz) bands. Frequency-dependent LFP power characteristics are calculated and input to the neural pain decoder in real time, for example, based on a state space model (SSM). In the presence of noxious stimulation, the SSM identifies the relative change from baseline in neural activity (Z-score) observed in the ACC or S1, and uses this change in neural activity as a surrogate for acute pain signals. To optimize the specificity of pain detection, in one embodiment, a cross-correlation function (CCF) may track in time the synchronized changes in local field potential (LFP) features for encoding pain in S1 and ACC. The use of simultaneous signals from cortical regions improves the accuracy of pain decoding by capturing both sensory and aversive components of pain. This CCF combines two SSM estimated Z-scores derived from LFP features of SCC and S1 separately, and optimizes detection performance by adjusting the relative weight of each region's contribution.For online closed-loop applications, the CCF-based decoder may automatically detect the onset of pain receptor signals to initiate electrical stimulation of the PFC to treat pain.

[0037] In an exemplary embodiment, a closed-loop neural interface for pain control is described, which automatically detects and treats pain with negligible delay. No human intervention is required. Pain detection is performed by simultaneous recording from multiple brain regions to increase sensitivity and specificity. LFP is used to decode pain, providing a stable signal and consistently high decoding accuracy. The exemplary system can also be applied to non-invasive EEG recordings. Deep brain stimulation is automatically initiated upon pain detection by the pain decoder and is only continued for a short period of time to minimize side effects and increase therapeutic efficacy. The system can effectively treat both evoked pain and spontaneous pain occurring at rest. The closed-loop system can also be adapted for other therapies such as ultrafast drug administration and electrical, optical, and ultrasound neuromodulation of the brain, spinal cord, and peripheral nervous system. One skilled in the art will appreciate that the exemplary embodiment can be adapted to decode pain predictions by targeting slightly different brain regions.

[0038] Closed-loop brain-machine interfaces (BMIs) link neural signals related to sensory or motor matters with neuromodulation to potentially restore or enhance brain function and treat neuropsychiatric disorders. BMIs have shown promising results in the treatment of epilepsy and upper or lower motor neuron diseases. However, their application to sensory disorders is limited due to the difficulties in detecting accurate sensory signals and providing fast and effective behavioral feedback.

[0039] Pain represents a unique challenge and opportunity for BMI design. Chronic pain is one of the most common sensory disorders and is defined as a discrete pain that can be triggered by noxious stimuli or occurs spontaneously. As mentioned above, current treatments for chronic pain are limited to either scheduled drug administration or continuous spinal neuromodulation, which do not take into account the precise timing of individual pain episodes, leading to frequent treatment delays, undertreatment, and overtreatment. Conceptually, BMI-based approaches are ideal for pain management by selectively targeting discrete nociceptive features, but decoding pain signals remains challenging. Unlike other sensory modalities, there is no single target for pain indications.

[0040] Among the distributed network of pain-processing brain regions, the primary somatosensory cortex (S1) is known to code the sensory discriminatory aspects of pain, such as the location, timing, and quality of pain. Meanwhile, the anterior cingulate cortex (ACC) has been shown to play an important role in aversive responses to pain through various types of studies. Because pain is processed in multiple brain regions, an attractive strategy for decoding pain is to integrate neural signals from multiple regions, with S1 and ACC as the most relevant targets. However, as will be appreciated by those skilled in the art, this exemplary system can also target other brain regions, such as the insular cortex, thalamus, and secondary somatosensory cortex (S2).

[0041] Current BMI applications primarily utilize neuronal spiking to generate accurate decoding signals. However, faithfully recording individual spikes over the long periods required for clinical management of chronic pain can be challenging. In contrast, local field potentials (LFPs), which represent subthreshold synaptic activity from localized neuronal collections, are relatively stable chronic electrophysiological recordings. Although the stability of the signal makes LFPs easier to apply clinically, they have only recently been used as an alternative to spikes for decoding in BMI applications.

[0042] In terms of therapeutic targets, the prefrontal cortex (PFC) is a key center for top-down control of sensory experience, and multiple lines of evidence from human and animal studies indicate that hypoactivity in the PFC contributes to chronic pain symptoms. Importantly, stimulating excitatory neurons in the PFC has been demonstrated to rapidly suppress pain withdrawal and aversive responses to pain without significant side effects, supporting the PFC as a potential therapeutic avenue in BMI design for closed-loop pain control.

[0043] According to exemplary embodiments of the present disclosure, an LFP-based decoding strategy is developed using recordings from S1 and ACC, which is combined with optogenetic or electrical stimulation of the PFC to form a multi-region neural interface. As described in more detail below, exemplary multi-region LFP-based brain-machine interface (BMI) studies of the present disclosure have shown that the multi-region neural interface of the exemplary embodiments can provide analgesic effects for both acute and chronic pain conditions with high sensitivity and specificity over an extended period of time.

[0044] As shown in FIG. 1, a multi-region LFP-based BMI system 100 according to an exemplary embodiment has been designed and tested to detect and treat pain in real-time, as described in more detail below. The exemplary system 100 comprises a plurality of probes 102 for detecting local field potentials (LFPs), a processor 104 including a pain decoder model 106 configured to detect pain onset based on the LFPs, and a stimulator 108 for providing pain modulation in response to the detected pain onset. A graphical user interface (GUI), as shown in FIG. 8 and described in more detail below, illustrating the detected LFPs in the brain regions, the detection of pain, and / or the delivery of pain modulation, may be presented to a user via a display. The GUI may also include user selectable items for detecting LFPs and providing pain modulation.

[0045] In an exemplary embodiment, the probe 102 may be implanted in a desired brain region, such as the ACC or S1, to record LFPs. In an exemplary embodiment, LFPs in the ACC and S1 are detected simultaneously. However, one skilled in the art will appreciate that the device may have an option for the user to select a detection strategy (e.g., based on the ACC, S1, or a combination of both) if desired. Although the exemplary embodiment shows the probe 102 specifically implanted in the ACC and S1 regions, one skilled in the art will appreciate that the probe 102 may be inserted in any desired brain region of the brain that may generate pain signals. The probe 102 may include, for example, a silicon probe array, but may also include any of a variety of probe devices for detecting LFPs.

[0046] The LFPs detected by the probe 102 are processed via the processor 104 and sent to the pain decoder model 106 to detect the onset of pain. In an exemplary embodiment, the pain decoder model may be a machine learning model that can be trained based on a state space model, as described below. According to an exemplary embodiment, the detection of pain by the pain decoder model 106 activates neural feedback in the stimulator 108. The stimulator 108 may be placed in a desired brain region to provide stimulation for pain modulation. In an exemplary embodiment, the brain region to be treated may include the prelimbic prefrontal cortex (PL-PFC). The stimulator 108 may include optical fibers to provide optogenetic activation of the brain region and / or electrodes configured to provide electrical stimulation of the brain region. Testing of the system 100 confirmed that the system 100 can detect and treat pain in real time based on the LFPs, as described below.

[0047] The processor 104 may be configured to execute computer-executable instructions for operation of applications that provide functionality to the system 100. For example, the processor 104 may include instructions for processing and transmitting detected LFPs to the pain decoder model 106, which includes instructions for detecting an onset of pain based on the detected LFPs, and upon detection of an onset of pain, the processor 104 activates the stimulator 108. However, it is noted that the functionality described with respect to the processor 104 and / or the pain decoder model 106 may be represented as separate components integrated into the system 100, modules connected to the processor 104, or functions that can be implemented by multiple processors 104. For example, the system 100 may be configured as a networked computing system, with each computing system including one or more devices and / or providing one or more of the above-mentioned functions.

[0048] Based on the known roles of S1, ACC, and PFC in pain processing and modulation, a multi-region closed-loop neural interface for pain control was developed by using a pain decoder based on simultaneous neural signals from ACC and S1 to initiate therapeutic stimulation of the prelimbic PFC (PL-PFC) in freely behaving rats, as shown in Figures 2-3. As shown in the figure, a silicon probe array (probe 102) was implanted in the rat's anterior cingulate cortex (ACC) and primary somatosensory cortex (S1), and local field potentials (LFPs), which simultaneously represent subthreshold local neural activity, were recorded. LFPs are relatively stable in chronic electrophysiological recordings and are easy to apply clinically, so they have been increasingly used as an alternative to spike activity in population decoding.

[0049] The detected LFP signals are processed and sent to an automatic decoder (pain decoder model 106) based on a state space model (SSM) to detect the occurrence of pain signals. The detected pain occurrence triggers neural feedback via optogenetic or electrical activation of the prelimbic prefrontal cortex (PL-PFC) to modulate pain, for example, via optical fibers and / or electrodes (stimulator 108) placed in the PL-PFC. From these signals, pain-eliciting event-related potentials (ERPs) from the ACC and S1 could be read out, as shown in Figure 4, indicating that these two regions contain pain-receptive signals. As will be explained in more detail below, the ERPs were extracted for each latency, and it was found that the latency of the ERP peak in the ACC was slightly longer on average than that of the S1. This suggests that pain information reached the S1 before the ACC, as shown in Figure 5. These results support the use of LFP signals from the S1 and ACC to decode pain.

[0050] The unsupervised machine learning pain decoder is model-based and trained to decode pain from multi-domain LFP signals. Previous studies have shown that the spectral characteristics of low gamma (30-50Hz), high gamma (50-100Hz), and ultra-high frequency (300-500Hz) bands are particularly relevant for cortical pain processing. Ultra-high frequency power can be considered as a proxy for multi-unit activity. The frequency-dependent LFP power characteristics are calculated and fed into the neural pain decoder in real time, for example, based on a state space model (SSM). As shown in Figure 6, the raw LFP signal is processed and the ACC channels {y ACC 1,k ,y ACC 2,k ,y ACC 3,k} and S1 channel {y S1 1,k ,y S1 2,k ,y S1 3,kThe power characteristics of the ACC LFP characteristics {Y ACC k} and the LFP characteristics of S1 {Y S1 k}, each latent variable {z ACC k , z S1 k} independently. SSM is represented by a graphical model with a Markov structure, where each node represents a random variable and the arrows indicate the statistical dependency between two random variables. A strategy for continuous detection of the occurrence of pain signals is used. First, the latent variables {z ACC k , z S1 k}, a Z-score is derived (horizontal dashed lines indicate confidence intervals for 95% statistical significance). A moving average cross-correlation function (CCF) is then used to calculate the correlation between the two Z-score series. The area beyond the statistically significant range (horizontal dashed lines) is calculated to determine the change point. When pain onset is detected, the decoder automatically initiates optogenetic or DBS stimulation to activate the PL-PFC.

[0051] In the presence of noxious stimuli, the SSM identifies the relative change from baseline in neural activity (Z-score) observed in the ACC or S1 and uses this change in neural activity as a surrogate for acute pain signals. To optimize the specificity of pain detection, a cross-correlation function (CCF) was designed to track temporally consistent changes in pain-encoding LFP features in the S1 and ACC. Using simultaneous signals from these cortical regions, both the sensory and aversive components of pain can be captured. This CCF combines two SSM-estimated Z-scores derived from the LFP features of the ACC and S1 separately, optimizing the detection performance by adjusting the relative weights of each region's contribution. In the online BMI experiment, the CCF-based decoder is used to automatically detect the onset of pain signals and initiate optogenetic or electrical stimulation of the PL-PFC to control pain.

[0052] Fig. 7 shows a schematic diagram of a BMI design based on the above exemplary system 100. Fig. 8 shows an example of a GUI for the BMI, which provides the user with the option to select an LFP channel and allows real-time visualization of the LFP signal. As shown in the illustrated GUI, the user can select a detection strategy (based on ACC, S1, or a combination of both (CCF method)) and can choose to change the significance threshold criteria.

[0053] The decoding strategy was tested in a series of acute pain assays. First, rats were presented with a noxious mechanical stimulus (pinprick, PP) or a non-noxious mechanical stimulus (von Frey filament, vF, with 6 g intensity) in the hind paw, and LFPs were recorded from the contralateral regions of the ACC and S1. As expected, rats showed higher paw withdrawal rates to PP stimuli than to vF stimuli. In online experiments, the present SSM decoder based on multisite recordings successfully detected the onset of noxious PP stimuli, but not non-noxious vF stimuli. In offline experiments, the detection accuracy using LFPs recorded from the ACC, S1, or a combination of both is also compared. In the latter case, the CCF method is used to adjust the weights of the Z-scores estimated from the ACC or S1. Based on LFPs from the S1, ACC, or a combination of S1 and ACC (CCF method), the detection rate of noxious stimuli (PP) was found to be significantly higher than that of non-noxious stimuli (6 g vF).

[0054] In identifying pain episodes, the decoding accuracy based on only S1 or CCF was higher than that based on only ACC. More importantly, the CCF-based decoding strategy was significantly better than the S1-only or ACC-only strategies in avoiding false positives (detections in the absence of noxious stimuli). Such high true positive rates and low false positive rates are of great importance for the practical implementation of BMI systems, and demonstrate the importance of using multiple regions to optimize sensitivity and specificity in pain decoding.

[0055] In the online BMI operation experiment, the decoder is trained with several calibration trials, after which the decoder is run continuously to automatically detect the onset of pain signals. The false positive rate obtained by the CCF-based decoding strategy is compared to that obtained by a single-domain decoding strategy. It is found that the multi-domain decoding strategy significantly reduces false positives compared to detection from either of the two domains alone, further demonstrating the importance of an exemplary multi-domain decoding approach.

[0056] Signal stability is important in therapeutic BMI applications. The reason for using LFP signals is their relative stability over time. To verify this, the reliability of the LFP signals used in the example pain decoder was tested. As shown in FIG. 9, the LFP-based strategy of the example embodiment was found to indeed maintain a high level of decoding accuracy over a 3-month period. FIG. 9 shows a comparison of false positive rates based on the LFP decoding strategy using ACC, S1, and combined (ACC+S1) signals over two sessions (the second session was performed 3 months after the first session). A t-test is used here, where n=5, P=0.6183 (ACC), P=0.7292 (S1), and P=0.8133 (ACC+S1). Furthermore, as shown in FIG. 10, when model parameters derived from an unsupervised learning algorithm trained on day 1 were used, the same model was found to be able to detect acute pain with high accuracy on day 5, suggesting that frequent training and calibration of the model parameters may not be required. The first three trials were used to train the parameters of the SSM on day 1. The same parameters were then used to detect pain on the following 5 days. One-way ANOVA with repeated measures was performed, p=0.6888, n=5. The fidelity of such a signal and model in decoding pain is attractive for real-world applications involving chronic neural recordings.

[0057] After establishing the accuracy, specificity, and reliability of the example CCF-based pain decoder, the example decoder was combined with optogenetic stimulation of pyramidal neurons in the PL-PFC (using a CaMKII promoter expressing channelrhodopsin (ChR2)) to form an analgesic BMI. To evaluate how this BM can suppress acute mechanically induced pain, a conditioned place preference (CPP) assay was used. In the preconditioning phase, animals spent 10 min moving freely between the two chambers. During conditioning, each chamber was paired with a peripheral stimulus (noxious or non-noxious) in combination with the BMI or various optogenetic neurostimulation control protocols. In the test phase, the peripheral and neurostimuli were removed and the rats were allowed to move freely between the chambers. If the BMI treats pain, the rats should prefer the chamber associated with the BMI. The CPP score was calculated by subtracting the time the rats spent in the chamber associated with the BMI during the preconditioning phase from the time they spent in the test phase to quantitatively measure the effect of the BMI on reducing pain avoidance behavior.

[0058] First, we compared noxious PP stimulation combining BMI-induced and optogenetic stimulation with the PL-PFC (BMI+PP) with PP combined with random optogenetic stimulation of the PL-PFC matched in duration and intensity (random nerve stimulation+PP). Rats preferred the room associated with BMI, suggesting that acute mechanical pain was reduced. This experiment was then repeated in rats expressing yellow fluorescent protein (YFP) rather than ChR2, and it was found that rats administered YFP did not experience pain relief. Comparison of CPP scores highlighted the efficacy of BMI in producing analgesia. As a positive control experiment, we compared manual activation of the PL^PFC immediately after delivery of PP to the leg (manual PL-PFC stimulation+PP) with random PL-PFC stimulation combined with PP (random nerve stimulation+PP). Here, we observed a preference for manual PL-PFC activation in ChR2 rats, but no such preference in YFP rats.

[0059] This result further demonstrated that the BMI functions similarly to precise manual control of the PL-PFC. To confirm this finding, we compared BMI control of the PL-PFC in the presence of PP with manual control of the PL-PFC in the presence of PP and found that rats could not distinguish between these two treatments. Finally, to show that the effect of BMI-induced PL-PFC activation is pain specific, we examined rats' preference for BMI in the presence of non-noxious vF stimulation (BMI + 6g vF vs. random PL-PFC activation + 6g vF) and found that rats showed no preference for either chamber. These results support the specificity of the BMI to achieve behavioral pain control without significant side effects.

[0060] Exemplary multi-region BMI for acute heat pain using the Hargreaves test. Infrared (IR) stimuli were applied to the hind paws of rats at two different intensities (IR70 for noxious, IR10 for non-noxious). It was found that rats withdrew their paws 100% of the time for IR70 stimulation, whereas rats withdrew their paws less than 10% of the time for IR10 stimulation. The multi-region LFP-based pain decoder of the exemplary embodiment successfully detected the onset of acute heat pain before paw withdrawal due to noxious stimulation, as opposed to non-noxious heat stimulation. Similar to mechanical pain decoding, the multi-region CCF-based pain decoder had a significantly lower false positive rate for non-noxious stimulation (<10%), while having a higher detection rate for noxious stimulation (~80%) than single-region decoding. The effectiveness of this multi-region neural interface for heat pain relief was then tested. In the presence of neural stimulation driven by the exemplary multi-region BMI, the latency to paw withdrawal due to analgesia was significantly increased. The exemplary BMI was found to achieve a similar reduction in analgesic paw withdrawal as manually controlled homeostatic PL-PFC activation. As expected, control rats expressing YFP showed no pain relief, supporting the specificity of BMI control for acute heat pain.

[0061] We next tested this closed-loop multi-region neural interface to determine whether the exemplary BMI could also treat chronic pain using the complete Freund's adjuvant (CFA) model, a well-known inflammatory pain model. CFA was injected into the rat's leg, contralateral to the leg in which the recording electrodes were implanted. Rats administered CFA exhibited persistent mechanical allodynia lasting for 14 days, with higher rates of leg withdrawal in response to 6g vF (allodynic) stimulation than to 0.4g vF (nonallodynic) stimulation. The exemplary LFP-based decoding strategy reliably detected the onset of allodynic attacks. Again, the multi-region decoding strategy (using the CCF method) significantly reduced the false positive rate while maintaining a relatively high decoding sensitivity.

[0062] We next used the CPP assay to test the anti-aversive effect of BMI in the CFA model. Allodynic 6g vF stimulation was paired with BMI in one chamber and with random PL-PFC activation matched in duration and intensity in the contralateral chamber. Rats expressing ChR2 showed a preference for the chamber associated with BMI, in contrast to YFP rats who showed no preference for the chamber. To confirm that this anti-aversive effect of BMI exhibited specificity to pain, we repeated the same experiment using non-allodynic 0.4g vF stimulation. In this case, neither ChR2 nor YFP rats showed any preference for BMI treatment.

[0063] In addition to hypersensitivity to eliciting stimuli, the main pathological feature of chronic pain is tonic pain, i.e. spontaneously occurring pain events (pain episodes). Currently, there is no test that can reliably identify the onset of these attacks, making treatment plan decisions very difficult, leading to delayed treatment, undertreatment, or overtreatment. To test this exemplary strategy, a classical CPP design was used to reveal persistent pain in rats administered CFA. In this assay, one of the chambers was paired with the present multi-region BMI and the other with random PL-PFC optogenetic stimulation. No peripheral stimulation was given, but the rats were conditioned to elicit tonic pain attacks over a long period of time. An exemplary multi-region decoder was trained using noxious stimulation (PP). The trained decoder was now able to automatically detect persistent pain events even in the absence of peripheral stimulation.

[0064] During conditioning, one room is equipped with the BMI of the exemplary embodiment, which uses automated tonic pain detection to initiate (trigger) optogenetic PL-PFC activation, and the other room is equipped with random PL-PFC stimulation that is matched in duration and intensity. After this conditioning, rats administered CFA were found to prefer the room associated with the BMI, indicating that this treatment is more likely to target the onset of tonic pain as opposed to random PL-PFC stimulation. This preference was not shown in control rats administered YFP. CPP scores further quantified the effectiveness of the BMI in reducing aversive responses to chronic tonic pain. These results strongly suggest that this exemplary multi-region neural interface can effectively treat spontaneous pain in a timely manner. Finally, the anti-allodynic effect of the BMI was tested. Application of the exemplary BMI significantly reduced mechanical allodynia in rats administered CFA, further validating the analgesic effect of the exemplary neural interface.

[0065] The use of optogenetics provides cell-type specific stimulation but is currently unavailable for clinical applications. To increase the translational value of the exemplary BMI, optogenetic stimulation of the PL-PFC was replaced with electrical deep brain stimulation (DBS) that can be safely used in humans. Electrical stimulation of the PL-PFC was combined with the exemplary multi-region LFP-based decoder to generate closed-loop BMI-triggered DBS. First, a CPP assay was performed to evaluate the efficacy of this BMI-triggered DBS in treating acute mechanical pain. When repeated noxious stimulation (PP) was given, rats were found to prefer the room associated with the BMI over the room paired with randomly timed DBS, indicating that the exemplary BMI-triggered DBS provided mechanical pain relief.

[0066] Furthermore, the BMI reduced acute heat pain in the Hargreaves test. Next, the efficacy of BMI-triggered DBS in chronic pain treatment was evaluated. It was found that the exemplary system significantly reduced mechanical allodynia in rats administered CFA. CPP was performed under the condition of allodynic stimulation (6g vF). When this allodynic stimulation was administered, rats administered CFA preferred the room paired with BMI over the room paired with randomly timed DBS, indicating that the neural interface reduced pain aversion. Finally, a CPP assay for spontaneous pain was performed. The exemplary multi-region decoder was trained using allodynic 6g vF stimulation, after which the decoder was able to automatically detect the onset of persistent pain and initiate therapeutic DBS during the conditioning phase. After conditioning, rats administered CFA preferred the room paired with BMI-triggered DBS over the randomly administered DBS. Similarly, when BMI-induced DBS conditioning was compared with no DBS, rats administered CFA were found to prefer the room associated with BMI. These results indicate that BMI-triggered DBS can suppress persistent pain. To confirm that this BMI-triggered DBS does not produce significant side effects, we investigated the stimulating effect on locomotion and found no side effects.

[0067] As described above, the study of the multi-region LFP-based neural interface of system 100 was designed as a new method of pain medication delivery. The exemplary system 100 utilizes recordings from multiple brain regions to increase the specificity of pain decoding. The system 100 is stable over time and compatible with current human electroencephalography (EEG) or electrocorticography (ECoG) devices. The interface has been demonstrated to provide near-instantaneous pain relief. The use of the exemplary neural interface with optogenetic stimulation of PL-PFC pyramidal neurons supports cell type specificity and allows for the investigation of a variety of mechanisms, but success with DBS further expands the potential for clinical application of closed-loop pain control.

[0068] There is no single region of the brain that specifically processes pain information. Instead, different brain regions process different aspects of pain. To solve the challenge of accurate pain detection, this study utilized strategies that accommodate this unique multidimensional nature of the pain experience. In particular, the exemplary neural interface design decodes pain based on neural signals recorded simultaneously from multiple brain regions. Ascending pain signals from the periphery are known to terminate in the ACC and S1. The ACC is well known for processing the emotional components of pain across a range of types, and neural activity in this region has previously been used to decode pain intensity and timing. Meanwhile, S1 provides important sensory information about pain somatosensorily. Previous studies have further demonstrated that the flow of information between these two brain regions integrates sensory and emotional information to produce a holistic pain experience. Testing of the exemplary system 100 has demonstrated the success of the multi-region neural interface in treating acute and chronic pain, demonstrating the specificity of decoding based on simultaneous signals from S1 and ACC. At the same time, at a mechanistic level, the results of the above tests also provide evidence that these two regions together play an important role in pain embodiment.

[0069] Another important advancement of the described study is the real-time decoding of pain using LFPs. Spikes provide specific signals at the individual nerve level, but are less stable over time in freely moving animals and humans. In contrast, LFPs offer an alternative solution for neural readout. In the currently described study, LFPs were reliably recorded over a period of 3 months. This signal stability allows the use of LFPs for BMI applications under chronic recording conditions, which is of great importance in the management of chronic pain and similar neuropsychiatric disorders. Remarkably, the exemplary decoding model (pain decoder model 106) remains stable for 5 days after training. The robustness of the current model likely results from a combination of signal stability and the use of multi-region decoding, which holds promise for future clinical applications.

[0070] Many studies have shown that the prelimbic PFC provides pain inhibition through top-down projections and projections to other cortical regions. In an exemplary embodiment, this region is used as the treatment site of the neural interface because it is one of the few neural structures that can control both the sensory and emotional components of pain, especially in the context of nociceptive input. The establishment of an exemplary BMI that inhibits both sensory avoidance and pain aversion indicates that this choice is correct. Because of the functional homology between the prelimbic PFC of rodents and the dorsolateral PFC of primates, adapting the exemplary neural interface to the dorsolateral PFC may provide treatment based on the desires of chronic pain patients.

[0071] Those skilled in the art will appreciate that false positives still occur in the above studies, likely caused by the possibility of non-specific neuronal firing in S1 and ACC and / or non-stationarity of neural signals in freely behaving rats. However, it has been shown that integrating neural activity in two different brain regions that play complementary roles in pain processing can minimize false positives and improve specificity. This approach to decoding pain using signals from multiple brain regions supports the multidimensional nature of pain experience. Each cortical region, such as S1 and ACC, may process unique aspects of pain in addition to other behavioral functions. However, since multiple brain regions need to be activated / deactivated simultaneously during pain development, a decoder based on activity across multiple nodes of the pain network is more likely to improve specificity. This decoding approach can be easily extended to incorporate additional brain structures, such as the insular cortex, to further improve the specificity of the decoding. Future studies can also combine neural signals with real-time behavioral analysis to achieve even more sensitive and more specific pain detection.

[0072] The PFC has multiple functions. Thus, non-specific effects can be expected in neuromodulation treatment deployed by PFC stimulation. Non-specific side effects are a common problem with neuromodulation and have indeed been observed in existing clinical applications of DBS. There are two strategies to reduce non-specific effects: targeting neural structures or neuronal populations that are highly specific for pain or limiting the treatment to a certain period of time. Currently, no single target in the central nervous system is known that can reliably treat pain without side effects. In this study, the second strategy was utilized: an exemplary closed-loop demand-based approach reduces side effects by limiting neuromodulation directly to the duration of the detected pain episode. As a result, no significant behavioral deficits were observed. In the future, if neuronal populations with specific pain modulation functions are discovered, they may be adapted to the treatment interface to further improve the specificity of the treatment and minimize side effects. At the same time, an exemplary BMI can also be used to facilitate such discoveries.

[0073] In conclusion, the multi-domain neural interface of system 100 has been designed and tested to demonstrate reliable pain detection and treatment. The use of stable LFP signals allows chronic use, and the pain decoder is also compatible with ECoG and EEG recordings. Given the clinical feasibility of EEG or ECoG recordings and DBS, the adoption of this technology may open new avenues for the treatment of patients suffering from chronic debilitating pain.

[0074] Additional refinements to the exemplary closed-loop pain detection and treatment system include, for example, training the exemplary pain decoder 106 using neural signals from pain processing regions such as the ACC and S1. Signals recorded from regions not known to process pain signals can be used as negative controls. Using such negative controls further increases the specificity of pain decoding. In further exemplary embodiments, the pain decoding thresholds in the exemplary algorithm can be adjusted. Thus, different models can be created for each subject for high, medium, or low sensitivity pain decoding. In these cases, the exemplary decoder 106 and subsequent processing allows for different levels of sensitivity and specificity. Such model adjustments can be easily performed in pain patients as well as in preclinical models. In clinical settings, participating subjects can determine the sensitivity level that best suits their clinical needs.

[0075] In addition to the PFC, the exemplary stimulator 108 of the system 100 can be developed to target additional brain regions for treating pain. These regions include, but are not limited to, the primary motor cortex, the prefrontal cortex, the periaqueductal gray, or the thalamus. The exemplary decoding strategy can be applied to other pain-processing brain regions, such as the insular cortex, or a combination of the insular cortex and the ACC. In another exemplary embodiment, the decoding analysis can be extended from being based on LFP measurements to intracranial EEG signals (accessible to epilepsy patients with deep electrode implants).

[0076] As confirmed by the results of the above studies, a method 200 based on the system 100 (shown in FIG. 11 ) can be used to detect and treat pain. In step 210, neural signals from multiple brain regions of the patient are detected via the probe 102. As discussed above, these neural signals include LFPs from brain regions indicative of pain. In an exemplary embodiment, these brain regions include the ACC and S1. However, one skilled in the art will appreciate that neural signals from other brain regions may additionally or alternatively be detected and received.

[0077] At step 220, the processor 104 processes the detected LFP and transmits the processed LFP data to the pain decoder model 106 at step 230. In an exemplary embodiment, the processor calculates the frequency-dependent power characteristics of the LFP (step 220) as described above and inputs these power characteristics to the pain decoder model 106 (step 230). At step 240, the pain decoder model 106 can identify relative changes indicative of pain. If pain is detected, stimulation of a brain region is triggered at step 250 to treat the pain. In particular, the processor 104 can trigger activation of a stimulator 108, including, for example, optical fibers and / or electrodes, to provide optical and / or electrical stimulation to the brain region. As described above, the brain region can include, for example, the prefrontal cortex.

[0078] If no pain is detected, method 200 continues to monitor the brain as described above. Additionally, once pain and stimulation of the brain region is identified, method 200 continues to monitor and treat pain in real time.

[0079] It will be apparent to those skilled in the art that various modifications can be made to the disclosed exemplary embodiments, methods, and alternatives without departing from the spirit or scope of the disclosure. Thus, the disclosure is intended to cover modifications and variations provided they come within the scope of the appended claims and their equivalents.

Claims

1. 1. A computer-implemented method for detecting and treating chronic pain, comprising: receiving neural signals from a plurality of brain regions via probes implanted in a plurality of brain regions of the patient's brain, the neural signals including local field potentials (LFPs) of the plurality of brain regions; Processing the neural signals and inputting the processed neural signals into a machine learning pain decoder model; determining whether pain is indicated based on the processed nerve signals; A computer-implemented method comprising: causing a computer to produce stimulation to a target region of the patient's brain based on the pain indication when pain is indicated via the pain decoder model.

2. 2. The computer-implemented method of claim 1, further comprising: A computer-implemented method, wherein processing neural signals includes calculating frequency dependent power characteristics of local field potentials of multiple brain regions.

3. 3. The computer-implemented method of claim 2, further comprising: The computer-implemented method, wherein determining whether pain is indicated comprises identifying relative changes in neural activity in a plurality of brain regions.

4. 2. The computer-implemented method of claim 1, further comprising: The computer-implemented method, wherein the plurality of brain regions includes anterior cingulate cortex and primary somatosensory cortex.

5. 2. The computer-implemented method of claim 1, further comprising:

11. A computer-implemented method, wherein stimulating a target region of the patient's brain includes optical stimulation and electrical stimulation.

6. 2. The computer-implemented method of claim 1, further comprising: The computer-implemented method, wherein the target region of the patient's brain includes the prefrontal cortex.

7. 2. The computer-implemented method of claim 1, further comprising:

11. The computer-implemented method, wherein the target region of the patient's brain includes one of the primary motor cortex, the anterior cingulate cortex, and / or the periaqueductal gray, and the thalamus.

8. 2. The computer-implemented method of claim 1, further comprising: The computer-implemented method further includes training a pain decoder model using a state space model based on spectral features from low gamma (30-50 Hz), high gamma (50-100 Hz), and ultra-high frequency (300-500 Hz) bands.

9. 1. A system for treating pain, comprising: a plurality of probes implantable in a plurality of brain regions of the patient's brain for detecting neural signals including local field potentials in the plurality of brain regions; a processing device that receives neural signals from a plurality of brain regions of the patient's brain, processes the neural signals, and inputs the processed neural signals to a pain decoder model, the pain decoder model being a machine learning based model configured to be indicative of pain; A stimulation device implantable in a target region of the patient's brain, the stimulation device providing stimulation to the target region based on an indication of pain.

10. 10. The system of claim 9, The system, wherein the processing unit is configured to process neural signals by calculating frequency dependent power characteristics of local field potentials of multiple brain regions.

11. 11. The system of claim 10, The pain decoder model is trained to identify relative changes in neural activity in multiple brain regions.

12. 11. The system of claim 10, The system is characterized in that the pain decoder model is trained using a state space model based on spectral features from low gamma (30-50 Hz), high gamma (50-100 Hz), and ultra-high frequency (300-500 Hz) bands.

13. 10. The system of claim 9, The system, wherein the processing unit is configured to initiate activation of the stimulator in response to an indication of pain.

14. 10. The system of claim 9, The system, wherein the stimulation device is configured to provide either light stimulation or electrical stimulation to a target area.

15. 10. The system of claim 9, The system, wherein the stimulation device is configured to be implanted in one of the prefrontal cortex, the primary motor cortex, the anterior cingulate cortex, the periaqueductal gray, and the thalamus.

16. 10. The system of claim 9, The system, wherein the plurality of probes are configured to be implanted in a plurality of brain regions including the anterior cingulate cortex and the primary somatosensory cortex.

17. 10. The system of claim 9, The system, wherein each of the plurality of probes comprises a silicon probe array.

18. 10. The system of claim 9, The system further comprises a graphical user interface that displays the LFP signal in real time and provides an option to change the threshold criteria.

19. A non-transitory computer-readable storage medium, storing a sequence of instructions executable by a processor, the sequence of instructions being executed by the processor to receiving neural signals from a plurality of brain regions of the patient's brain, the neural signals including local field potentials (LFPs) of the plurality of brain regions; Calculate the frequency-dependent power characteristics of local field potentials in multiple brain regions; Power characteristics were fed into a machine learning pain decoder model, trained to identify relative changes in neural activity across multiple brain regions to indicate pain. A storage medium causing the processor to cause stimulation of a target region of the brain based on indications of pain.

Citation Information

Patent Citations

  • Systems and methods for modeling brain dynamics in normal and diseased conditions

    JP2016512441A

  • Methods and systems for identifying pain sources and treating chronic pain

    JP2017530745A

  • Brain signal feedback for pain management

    US20030100931A1

  • Wireless neural recording and stimulating system

    US20090157141A1

  • Cortical Implant System for Brain Stimulation and Recording

    US20150157862A1