Systems and methods for monitoring neural activity

Abstract

The present application discloses a method and a system which detects a respective neural signal at each of one or more electrodes of a plurality of electrodes of an electrode array implanted in the brain. For each of the one or more electrodes, one or more waveform characteristics are extracted from the neural response signal and a weighting is determined for each of the one or more electrodes based on the one or more extracted waveform characteristics. In some examples the system or method determines an indicator of the distribution of the neural signals among the plurality of electrodes, based on the determined weighting for each of the one or more electrodes. In some examples the method or system determines whether the lead is located within a target neural region, based on the determined weighting of each of the one or more electrodes and the location of each of the one or more electrodes relative to the lead.

Description

"Systems and Methods for Monitoring Neural Activity"Technical Field

[0001] The present disclosure relates to deep brain stimulation (DBS) and, in particular, methods and systems of monitoring neural activity during DBS.Background

[0002] Deep brain stimulation (DBS) is an effective therapy for people with movement disorders, such as Parkinson’s disease. Accurate implantation of electrodes enables stimulation to be delivered to a target neural region while limiting the spread of stimulation beyond the boundaries of the target neural region.

[0003] There are a limited number of techniques utilised by neurosurgical teams in the intraoperative environment to confirm that a DBS lead is implanted acceptably in relation to a target neural region, such as the dorsolateral subthalamic nucleus (STN). Such methods are commonly employed to compensate for error intrinsic to the transferal of preoperative plans into the stereotactic environment along with error induced by brain shift due to loss of cerebrospinal fluid.

[0004] A microelectrode recording (MER) does not contain information to enable automated localisation of the dorsolateral STN. Additionally, MER assisted surgeries are primarily performed with patients awake, making some patients ineligible for the procedure due to symptom severity or other conditions. The interpretation of MER is also technically demanding of neurosurgical teams, requiring extensive experience for accurate recognition of signal characteristics indicative of the STN.

[0005] Local field potentials are similar to MER, being spontaneous signals that represent the activity of the local neuronal population. While local field potentials have been shown to better correlate with clinical outcomes than MER, the small amplitudes of such spontaneous signals are difficult to reliably record. Methods of intraoperativeguidance using local field potentials have also relied on the usage of multiple microelectrode implantations, increasing risk of intracranial haemorrhage.

[0006] As imaging technology has improved enabling the direct visualisation of the STN, intraoperative imaging has become an emerging intervention for lead placement confirmation in STN-DBS. The most common implementation involves the fusing of intraoperative CT with preoperative MRI. Such methods allow for implantation under general anaesthetic, but CT and MRI fusion relies on software to effectively align the CT and MRI such that the anatomy is aligned and clinical expertise to interpret the fused images. Inaccuracies in the software and subjective clinical assessments can introduce error. Additionally, the brain can shift within the cranial cavity and deform during the surgery thus contributing to error between the preoperative MRI and intraoperative CT fusion. Intraoperative MRI has also been investigated, however the artefact produced by the DBS lead is significant and irregular due to varying alignment between the magnetic field and the DBS lead. Such systems also increase surgery time and are not readily available in all centres due to cost or lack of expertise. Moreover, it is increasingly understood that the optimal stimulation location is not always indicated by imaging alone.

[0007] Any discussion of documents, acts, materials, devices, articles or the like which has been included in the present specification is not to be taken as an admission that any or all of these matters form part of the prior art base or were common general knowledge in the field relevant to the present disclosure as it existed before the priority date of each claim of this application.Summary

[0008] Some embodiments provide a method of monitoring neural activity in a brain, the method comprising: detecting a respective neural signal at each of one or more electrodes of a plurality of electrodes of an electrode array implanted in the brain; for each of the one or more electrodes, extracting one or more waveform characteristicsfrom the neural response signal; determining a weighting for each of the one or more electrodes based on the one or more extracted waveform characteristics; and determining an indicator of the distribution of the neural signals among the plurality of electrodes of the electrode array based on at least the determined weighting for each of the one or more electrodes.

[0009] The method may further comprise applying a stimulus to at least one electrode of the electrode array to evoke a neural signal. In some embodiments, determining an indicator may comprise determining a ratio of the neural signals detected at different electrodes of the plurality of electrodes. In some embodiments, determining an indicator may comprises determining a focal point of the neural signals based on spatial information of each electrode and the weighting corresponding to each electrode.

[0010] Determining the focal point may comprise: determining an origin; determining an electrode position with respect to the origin from the spatial information of each electrode of the plurality of electrodes; and calculating coordinates of the focal point using the electrode position and the weighting corresponding to each electrode. In some embodiments, the method may further comprise determining the spatial information of each electrode of the one or more electrodes.

[0011] In some embodiments, the focal point may be the centre of the neural signals. The focal point may comprise coordinates relative to an origin. The origin may be a point within the electrode array. The origin may be the centre of the electrode array. The focal point may comprise two-dimensional coordinates or three-dimensional coordinates. The focal point may be located within the electrode array.

[0012] In some embodiments, the coordinates of the focal point may be calculated using the equation:wherein:Cx, cy, Cz are the coordinates of the focal point;Xi, yi, zi are the coordinates of each electrode; wi are the weightings of the waveform characteristics applied to each electrode; and b is the number of electrodes.

[0013] In some embodiments, the method may further comprise converting the coordinates of the focal point into another coordinate space. The coordinate space may be a spherical coordinate space or a cylindrical coordinate space. The method may further comprise applying an offset to the coordinates of the focal point. The method may further comprise calculating a distance and / or a direction to the focal point.

[0014] Calculating a distance (r) of the focal point may comprise: treating the coordinates as a vector; and applying one of the following equations: r =r =

[0015] Calculating a direction (<9) of the focal point may comprise: treating the coordinates as a vector; and applying the equation:0 = tan-1(— ) cy where <9 is the direction of the focal point relative to the origin in an axial plane.

[0016] In some embodiments, the method may comprise obtaining a position of the lead relative to a target neural region in the brain based on the determined focal point of the neural signals. The method may further comprise calculating a distance, direction and / or depth of the focal point relative to the lead.

[0017] In some embodiments, the method may comprises: normalizing the one or more extracted waveform characteristics to produce one or more normalized values corresponding to the extracted waveform characteristics; and applying the one or more normalized values as the weighting for each electrode.

[0018] In some embodiments, one or more electrodes of the plurality of electrodes may be segmented electrodes. In some embodiments, the electrode array may be disposed on a lead. The method may further comprise determining a respective signal recording at each electrode of the one or more electrodes, each signal recording corresponding to the neural signal received by each electrode.

[0019] In some embodiments, the neural activity may be evoked neural activity. The neural activity may be spontaneous local field potentials. The neural activity may be spontaneous single or multi-unit activity.

[0020] In some embodiments, the one or more waveform characteristics may comprise one or more of: an amplitude, a frequency, a rate of change of frequency; a rate of change of amplitude; a rate of decay, a bandwidth, or a latency. The one or more waveform characteristics may comprise a relative difference or ratio between one or more of an amplitude, a frequency, a latency, a rate of decay, or a rate of change.

[0021] The one or more waveform characteristics may be extracted over a period of time of the neural signal. The one or more waveform characteristics may be combined and / or averaged over the period of time. The period of time may be a subset of the length of the signal recording of the neural signal.

[0022] In some embodiments, the method may further comprise: detecting a resting signal during a period of no stimulation; calculating a baseline from the resting signal; and removing noise from the neural signal based on the calculated baseline.

[0023] In some embodiments, the method may further comprise: extracting a resting waveform characteristic from the resting signal; removing the resting waveformcharacteristic from the extracted waveform characteristic to remove baseline noise. The noise and / or the baseline noise may comprise stimulation artefacts.

[0024] Some embodiments provide a method of analysing signals received by a deep brain stimulation lead, the method comprising: applying a stimulus to at least one electrode of the deep brain stimulation lead; detecting a respective neural signal at each of one or more electrodes of a plurality of electrodes of the deep brain stimulation lead; for each of the one or more electrodes extracting one or more waveform characteristics from the neural signal; determining a weighting for each of the one or more electrodes based on the one or more extracted waveform characteristics; and determining whether the lead is located within a target neural region based on the determined weighting of each of the one or more electrodes and the location of each of the one or more electrodes relative to the lead.

[0025] In some embodiments, the target neural region may comprise a target neural structure.

[0026] Determining whether the lead is located within a target neural region may comprise: processing the extracted waveform characteristics and location of each electrode to generate two or more features for input into a machine learning model, inputting the two or more features into the machine learning model; and outputting from the machine learning model an indication of the location of the lead relative to the target neural structure.

[0027] The method may further comprise determining a focal point of the neural signals based on the location of each electrode relative to the lead and the weighting of each electrode. The two or more features may comprise a maximum extracted waveform characteristic, a vector distance of the focal point of the neural signal and a vector direction of the focal point of the neural signal.

[0028] In some embodiments, the method may further comprise obtaining a position of the lead relative to the target neural region based on the focal point. Responsive to determining placement of the lead is not within the target neural region, the method may further comprise providing information on relative proximity and / or relative direction of the lead with respect to the target neural region based on the determined focal point. The method may further comprise determining whether the lead is located within the target neural region when there is an overlap between at least part of the lead and at least part of the target neural structure.

[0029] Some embodiments provide a system for monitoring neural activity in a brain, comprising: a lead having an electrode array, the lead adapted for implantation in or proximate to a target neural structure in the brain; a measurement device selectively coupled to one or more electrodes of a plurality of electrodes of the electrode array, the measurement device configured to detect a respective neural signal at each electrode of the plurality of electrodes; and a processing unit coupled to the measurement device and configured to: for each electrode, extract one or more waveform characteristics from the neural signal; determine a weighting for each electrode based on the one or more extracted waveform characteristics; and determine an indicator of the distribution of the neural signals among the plurality of electrodes of the electrode array based on at least the determined weighting for each electrode.Brief Description of Drawings

[0030] The appended drawings merely illustrate example embodiments of the present disclosure and cannot be considered as limiting its scope.

[0031] Figure 1A is a schematic view of an example DBS electrode lead tip, according to some embodiments;

[0032] Figure IB is a schematic diagram showing different types of DBS leads which may be used with methods disclosed herein, according to some embodiments.

[0033] Figure 1C is a graphical illustration of an example therapeutic patterned DBS stimulus and the associated neural response signal in the form of an evoked resonant response, according to some embodiments;

[0034] Figure 2 is an exploded view of the electrode lead tip of Figure 1 A, according to some embodiments;

[0035] Figure 3 is an illustration of respective neural response signals received at each electrode of the plurality of electrodes disposed on the lead tip of Figure 1 A, according to some embodiments;

[0036] Figure 4 is a process flow diagram of a method of monitoring neural activity in a brain, according to some embodiments;

[0037] Figure 5 is a process flow diagram of a method of determining a focal point, according to some embodiments;

[0038] Figure 6 is a process flow diagram of a method for calculating a focal point, according to some embodiments;

[0039] Figure 7 is an axial view of the lead of Figure 1 A, according to some embodiments;

[0040] Figure 8A is a process flow diagram of a first method for removing noise from the neural response signal recording, according to some embodiments;

[0041] Figure 8B is a process flow diagram of a second method for removing noise from the neural response signal, according to some embodiments;

[0042] Figure 9 is a process flow diagram of a method of analysing signals received by a DBS lead, according to some embodiments;

[0043] Figure 10 is a process flow diagram of a method for implanting a DBS lead, according to some embodiments;

[0044] Figure 11 is a schematic illustration of example recording of a neural response signal in different anatomical positions, according to some embodiments;

[0045] Figure 12 is a schematic diagram of a nested cross-validation method, according to some embodiments; and

[0046] Figure 13 is a schematic view of a system for monitoring neural activity in the brain according to some embodiments.Description of Embodiments

[0047] Embodiments of the present disclosure relate to improvements in neurostimulation in the brain, improvements in monitoring neural activity and / or improvements in positioning DBS devices within a target neural region of the brain. The inventors have determined an improved method for identifying an indicator of the distribution neural signals detected at each electrode of a plurality of electrodes of an electrode array implanted in the brain. The inventors have also determined an improved method for determining the position of DBS devices within the brain based on the indicator, and in particular determining whether the DBS device is positioned within a target neural region.

[0048] In some embodiments the electrode array may be used for DBS. One or more DBS electrode leads comprising an electrode array may be used for stimulation of one or more neural structures within one or both hemispheres of the brain. A DBS electrode lead may be referred to herein as a “DBS device”, “DBS lead”, “electrode lead” or “lead”. Each DBS electrode lead may comprise one or more electrodes located near the tip of each lead. The DBS electrode lead may further comprise one or more directional electrodes. Each of the electrodes may be used for stimulation, monitoring, or both stimulation and monitoring. One or more of these electrodes may be implanted.Implanted electrodes may be used independently or in addition to one or more electrodes placed on the outside of the brain or skull.

[0049] Figure 1A is a schematic view of an example DBS electrode lead tip 102, according to some embodiments. The DBS electrode lead tip may be a typical DBS electrode lead tip 100, such as that incorporated into the Vercise™ Cartesia™ DB- 2022. The lead tip 102 comprises an electrode array 104 including a plurality of electrodes. The lead may be a directional DBS lead or may include one or more directional or segmented electrodes. Directional electrodes, or segmented electrodes, are electrodes capable of emitting omnidirectional current or emitting current in a specific direction. Electrodes may be radially segmented, creating a plurality of separate directional electrodes, or contacts, which may be controlled to direct or steer a current flow in a particular direction. For example, electrodes may be radially segmented into three or four directional electrodes. The three directional electrodes may be used simultaneously and equally or may be used in various configurations in which one or more are used. The plurality of electrodes includes a first electrode 106 disposed at the distal end of the lead tip 102, a second electrode 108, a third electrode 110 and a fourth electrode 112. The first electrode 106 and the fourth electrodes 112 may be ring electrodes. The first electrode 106 and the fourth electrode 112 may each include two contacts from which an omnidirectional current can be emitted. The second electrode 108 is may be radially segmented into three directional electrodes 108a, 108b, 108c. Each directional electrode 108a, 108b, 108c, may span approximately 120 degrees of the circumference of the lead. However, it will be appreciated that the directional electrodes may span different angular widths of the circumference. The third electrode 110 is radially segmented into three directional electrodes 110a, 110b, 110c, each spanning approximately 120 degrees of the circumference of the lead. The electrode array 104 of the lead 100 thereby includes a 1-3-3-1 configuration. The lead 100 also includes a lead marker 114 at the top of the lead, which allows control of the rotational orientation of the lead 100.

[0050] In other embodiments, leads with more or less electrodes, or electrodes with different sizes or topologies may be used. In some embodiments, the electrodes may span different angular widths. The electrodes may be evenly segmented or unevenly segmented to form a plurality of segmented electrodes or directional electrodes. In some embodiments, leads having an electrode array with different configurations may be used, for example, the electrodes may be configured in a variety of positions. In some embodiments, an electrode lead may be an 8-channel electrode lead or a 16- channel electrode lead. In addition, one or more reference electrodes may be located at a remote site and used to complete the electrical circuit when one or more electrodes on the DBS lead are activated for stimulation or used for signal monitoring. Different types and variations of DBS leads which may be used with embodiments of the present disclosure are illustrated in Figure IB. Types of leads include, but are not limited, Medtronic 3387, Medtronic 3389, Abbott Infinity, Boston Scientific, Direct STNAcute, Medtronics-Sapiens, and / or pDBS. Some or all of the features and / or configurations of the leads illustrated in Figure IB may be used in embodiments disclosed herein.

[0051] The lead tip 102 may be implanted within a target neural structure of a brain. Once implanted into the brain, each of the electrodes 106, 108, 110, 112 may be used to apply a stimulus to one or more target neural structures. By controlling the stimulation parameters of each of the electrodes 106, 108, 110, and 112, a therapeutic or non- therapeutic stimulus can be administered to the target neural structure or a target neural region. The therapeutic or non-therapeutic stimulus may evoke a neural response signal. Such therapeutic or non-therapeutic stimulus may evoke neural activity in a patient without having any therapeutic impact or causing undesirable side effects.

[0052] Figure 1C graphically illustrates an example therapeutic patterned DBS stimulus 150 and the associated neural response signal in the form of an evoked resonant response according to an embodiment of the present disclosure. The patterned stimulus 150 is shown above the graph to illustrate the correlation between stimulus and response. In the patterned stimulus, a single pulse has been omitted from an otherwise continuous 130Hz pulse train. The pulse train therefore includes a pluralityof bursts of pulses of continuous stimulation, each burst separated by a first time period ti, and each of the plurality of pulses separated by a second time period t2. Continuation of the stimulus before and after omission of a pulse (or more than one pulse) maintains the therapeutic nature of the DBS, whilst the omission of a pulse allows for resonance of the evoked resonant neural activity (ERNA) to be monitored over several (3 in this example) resonant cycles before the next stimulation pulse interrupts this resonance.

[0053] In summary, by patterning non-therapeutic and therapeutic stimuli, an evoked response can be monitored over a longer period of time than with conventional nonpatterned stimulation. Accordingly, stimuli may be applied in bursts of multiple pulses, each burst separated by a first time period ti, of no stimulation, each pulse separated by a second time period t2. For example, a stimulus signal may comprise a series of 10- pulse bursts at 130 Hz. To increase repeatability of results, the multi-pulse burst may be repeated after a predetermined period of no stimulation. For example, the multi -pulse burst may be repeated each second. The duration of the first time period ti, is greater than that of the second time period t2. The ratio between the duration of the burst and the duration between bursts may be chosen so as to ensure that relevant properties of the neural response signal, in this case ERNA, can be monitored easily and efficiently. In some embodiments, the duration of each burst is chosen to be between 1% and 20% of the duration of no stimulation between bursts.

[0054] In another example, non-therapeutic stimulation in an intraoperative environment may include delivering a monopolar constant-current burst stimulation sequentially to each electrode at 130Hz with a pulse amplitude of 3360pA. For example, ten bursts of stimulation may be delivered with each burst containing 10 symmetric biphasic pulses (negative first, 60ps phase width).

[0055] The neural activity detected and / or monitored by the electrodes of the electrode array 104 may be spontaneous or evoked neural activity. In some embodiments, the neural activity may include spontaneous neural activity. For example, the neural activity may include spontaneous local field potentials. In anotherexample, the neural activity may include a spontaneous single or multi-unit activity. In further embodiments, the neural activity may include spontaneous brain signals such as beta, gamma and or high frequency oscillations (HFO). In some embodiments, HFO activity may be measured from local field potentials in the brain. In other embodiments, the neural activity may be evoked neural activity. In some embodiments, the evoked neural activity may be an evoked resonant neural activity (ERNA). The neural activity may have a corresponding neural signal. The neural signal may be an evoked neural signal, such as a neural signal evoked in response to an applied stimulus, or a neural signal resulting from neural activity that occurs spontaneously in the brain. Although embodiments of the present disclosure will be described with reference to a neural response signal, it will be appreciated that any neural signal may be used, for example, evoked or spontaneous, and is not limited to a neural signal corresponding to neural activity which occurs in response to a stimulus.

[0056] In some embodiments, the electrodes 106, 108a, 108b, 108c, 110a, 110b, 110c, 112 may be configured to each monitor and / or record a neural signal corresponding to the neural activity from neural circuits. In some embodiments, the neural signal may be in response to an applied stimulus. The neural response signal corresponding to the neural activity may be individually monitored at each electrode of the electrode array 104. Each electrode 106, 108a, 108b, 108c, 110a, 110b, 110c, 112 of the plurality of electrodes in the electrode array 104 therefore detects a respective neural response signal. Additionally, each electrode 106, 108a, 108b, 108c, 110a, 110b, 110c, 112 may be configured to record a respective neural response signal.

[0057] Figure 2 is an exploded view of the electrode lead tip 102, according to some embodiments. The plurality of electrodes shown on the lead tip 102 includes the first electrode 106 labelled as El disposed at the distal end of the lead tip 102, and the fourth electrode 112 labelled as E8. The three directional electrodes 108a, 108b, and 108c are directional electrodes labelled as E2, E3, and E4 respectively. The three directional electrodes 110a, 110b, 110c are directional electrodes labelled as E5, E6 and E7 respectively. Each of E2, E3 and E4, are disposed at the same depth on the lead tip102, and each cover a particular direction of approximately 120 degrees of the circumference of the lead. Each of E5, E6 and E7, are disposed at the same depth on the lead tip 102, and each cover a particular direction of approximately 120 degrees of the circumference of the lead. The directional electrodes E2, E3, E4, E5, E6 and E7 are configured to be selectively activated to direct stimulation or current flow towards a target neural region. Similarly, electrodes E2, E3, E4, E5, E6 and E7 are each configured to detect respective neural response signals from their position on the lead tip 102.

[0058] In some embodiments, the neural response signal may be an evoked neural response signal. For example, the evoked neural response signal may be a signal of evoked resonant neural activity (ERNA) detected at each electrode. The present disclosure may utilise the systems and methods of monitoring evoked resonant neural activity (ERNA) disclosed in WO 2018 / 213872, the contents of which are incorporated herein by reference. In other embodiments, the neural response signal may include a spontaneous neural response signal or an enhanced neural response signal.

[0059] In some embodiments, the method may further comprise determining a respective signal recording at each electrode, each signal recording corresponding to the neural response signal stimulated by one or more electrodes of the plurality of electrodes. Determining a respective signal recording at each electrode may include recording a respective neural response signal during an intraoperative procedure. For example, in some embodiments, neural response signal recordings may obtained using a bio-signal amplifier and a configurable custom-built neurostimulator. To allow for monopolar recording and stimulation, surface patch electrodes may be placed on each shoulder of a patient, serving as the amplifier ground and stimulation return respectively.

[0060] Figure 3 is an illustration of respective neural response signals 300 received at each electrode El, E2, E3, E4, E5, E6, E7 and E8 of the plurality of electrodes disposed on lead tip 102, according to some embodiments. The depictions of graphs showing therespective neural response signals 300 are illustrative only and are not intended to limit the type or characteristics of the neural response signal being detected and / or received by the electrodes. It will be appreciated that other graphical illustrations of the recordings may be used to depict the respective neural response signals detected by the electrodes. In Figure 3, each graph represents a recording of a respective neural response signal received at each electrode El to E8. The example neural response signals 300 shown are neural response signals corresponding to evoked resonant neural activity (ERNA). However, the neural response signals may correspond to spontaneous neural activity or other types of neural activity. Each graph represents the signal as a function of amplitude over time. The x-axis of each graph represents time and the y- axis of each graph represents amplitude. The shading of the graphs represents the root- mean-square (RMS) amplitude.

[0061] A recording of a neural response signal at an individual electrode produces a neural response signal recording. The temporal length of a recording may be referred to herein as an “epoch”. The epoch may refer to all or part of the length of the full recording of the neural response signal. For example, the epoch may refer to an entire length of recording including pre- stimulation and post-stimulation phases. In some embodiments, the epoch may refer to the length of the recording corresponding to the neural response signal of interest, such as an ERNA recording. The graphs shown in Figure 3 are recordings of the neural response signal which correspond to approximately 20ms epochs recorded after stimulation. Here, the epochs capture the ERNA signal evoked by applied stimulation. Epochs of the neural response signal may be in the range of 1ms to 100ms.

[0062] In some embodiments, a resting period may be captured as part of recording the neural response signal. The resting period may be in the range of 1 to 20 seconds. For example, a 12 second resting period before stimulation may be recorded. The resting period may be used to capture baseline neural activity. Neural activity may be recorded for a period of time post-stimulation, that is, after the last pulse of each burst of stimulation from all electrodes. For example, neural activity may be recorded for atleast 100ms after the last pulse of each burst of stimulation from all electrodes. The temporal length of the post-stimulation recording may be in the range of 1 to 200ms.

[0063] Using the neural response signal, the inventors have developed a method for determining an indicator of the distribution of the neural response signals among the plurality of electrodes of the electrode array based on at least a determined weighting corresponding to each electrode that has been determined from one or more extracted waveform characteristics of the neural response signal. In some embodiments, the inventors have developed a method for determining a focal point of the neural response signal using the respective neural response signals detected by each electrode of the electrode array. The determination of an indicator of the distribution utilises recordings of neural response signals obtained from one or more electrodes in an electrode array. The method for determining an indicator of distribution may be use as part of an intraoperative system. For example, it may advantageously be used to advise clinicians intraoperatively of a sufficient or insufficient position of the DBS lead and / o, in the case of an insufficient position, guide clinicians as to a direction in which a subsequent DBS lead should be implanted. The method of determining an indicator of the distribution of the neural response signal may additionally be used to inform clinicians as to a direction to shift a suboptimal lead.

[0064] The method of determining an indicator of a distribution of the neural response signals has a number of technical advantages. The method is agnostic to lead type and is able to used with high channel count lead designs. The indicator may be advantageously used to communicate information about the directionality or centre of the neural response signal, which can be particularly useful with high channel count leads where 8 or 16 or more signals may need to be interpreted. Additionally, the determination of an indicator is practically workable with a variety of waveform characteristics of the neural response signal, and may be adapted based on patient requirements or signal features.

[0065] Figure 4 is a process flow diagram of a method 400 of monitoring neural activity in a brain, according to some embodiments. In some embodiments, the neural activity may be responsive to a stimulus in the brain. The method 400 includes, at 410, detecting a respective neural response signal at each electrode of a plurality of electrodes on an electrode array implanted in the brain. The electrode array may be a DBS lead such as one of the type shown in Figures 1 or 2.

[0066] At 420, for each electrode, one or more waveform characteristics are extracted from the neural response signal. At 430, a weighting for each electrode is determined based on the one or more extracted waveform characteristics. The weighting for an electrode may indicate the relative strength of the neural response signal at the electrode.

[0067] The one or more waveform characteristics of the neural response signal refer to characteristics, attributes or features of the signal waveform. Waveform characteristics of the neural response signal may comprise one or more of amplitude, frequency, bandwidth, latency, natural frequency, damping factor, envelop, fine structure, onset delay, rate of change, and / or rate of decay. In some embodiments, waveform characteristics may comprise a relative difference or ratio between one or more of an amplitude, a frequency, a latency, a rate of decay or a rate of change. The waveform characteristics may be dependent on various physiological conditions of the patient. Accordingly, waveform characteristics may be manually selected, or may be automatically determined, for example, based on patient information or the neural response signal being received. In some embodiments, waveform characteristics may be extracted over a period of time of the neural response signal. The period of time of the recording may refer to all or part of the temporal length of a neural response signal recording or epoch. The period of time over which the waveform characteristic is extracted may be a subset of the length of the signal recording of the neural response signal.

[0068] Extracting the one or more waveform characteristics may comprise extracting values of the waveform characteristic. In some embodiments extracting the one or more waveform characteristics may include combining and / or averaging the waveform characteristics over a period of time. For example, combining and / or averaging the waveform characteristics may comprise calculating a root-mean- square (RMS) average of the waveform characteristic over a period of time. In some embodiments, the average RMS waveform characteristic is calculated for all epochs of each electrode in the electrode array. In some embodiments, the average RMS waveform characteristic is an average RMS amplitude of the neural response signal. The average RMS amplitude may be used to reduce signal noise from the recorded neural response signal. In some embodiments, extracting the one or more waveform characteristics may include measuring the RMS of the waveform characteristic over a period of time for all electrodes, or for each individual electrode.

[0069] The respective neural response signals for each electrode, such as those shown in Figure 3, are shown as a function of amplitude over time. Here, amplitude is at least one waveform characteristic which has been extracted from the neural response signal. It can be seen that the amplitude varies between the respective neural response signals detected from El to E8. In particular, each electrode’s detected neural response signal shown in Figure 3 can be weighted based on the amplitude waveform characteristic.

[0070] A weighting for each electrode may be determined based on the one or more waveform characteristics extracted from the neural response signal of each electrode. As shown in Figure 3, E8 has a smaller amplitude over time when compared with the other electrodes. The amplitudes of El, E3, E5 and E6 appear larger than E8 but smaller than the amplitude of E2. The amplitude of E2 appears smaller than the amplitudes of E7 and E4, with E4 having a slightly larger amplitude than E7. The shading of the graphs in Figure 3 indicates the RMS amplitude value of each electrode. Accordingly, based on the relative average RMS values of the amplitude over time for each neural response signal at each electrode, a weighting can be determined. Those with a comparatively stronger RMS amplitude will be weighted more than those with acomparatively weaker RMS amplitude. This results in a weighting being applied to each electrode based on the average values of the waveform characteristic over time.

[0071] Applying the weighting to each electrode may include normalising the one or more extracted waveform characteristics to produce one or more normalised values corresponding to the extracted waveform characteristic. For example, a normalised value between 0 and 1 may be determined for each electrode. The normalised values may then be applied as the weighting for each electrode.

[0072] Referring back to Figure 4, at 440, the method 400 includes determining an indicator of the distribution of the neural response signals detected at each electrode of the plurality of electrodes of the electrode array based on at least the determined weighting for each electrode.

[0073] The indicator of the distribution of the detected neural response signals provides an indication of how the neural response signal is distributed. For example, this may include an indication of where a characteristic of the neural response signal is the strongest, where a characteristic of the neural response signal is the weakest or where it cannot be detected, or how characteristics of the neural response signal changes over time or location. In some embodiments, the indicator provides an indication of the spatial characteristics of the neural response signal and / or neural activity. The indicator may summarise neural response signals received at each electrode into a central point. In some embodiments, determining an indicator of the distribution of neural response signals comprises determining a ratio of the neural response signals detected at different electrodes of the plurality of electrodes. In some embodiments, determining an indicator of the distribution of the neural response signals for each electrode of the electrode array may comprise determining a focal point of the neural response signals based on spatial information of each electrode, and the weighting corresponding to each electrode that was determined in 430.

[0074] Figure 5 is a process flow diagram of a method for determining a focal point, according to some embodiments. The method 500 includes determining an origin at 510. The origin may be manually input, selected or automatically determined. For example, the origin may be the centre of the electrode array. The origin may be determined with respect to spatial information regarding the electrode array, such as a diameter and / or length of the electrode array, or spatial positions of electrodes within the electrode array. The method 500 further includes, at 520, determining an electrode position with respect to the origin from the spatial information of each electrode of the plurality of electrodes. The spatial information of each electrode may be determined by receiving the spatial information, for example, by an input. In other embodiments, the spatial information may be accessed and / or extracted, for example from a file or data store containing the spatial information about each electrode. The spatial information for each electrode on the electrode array may be in accordance with manufacturer’s specifications. The method 500 further includes, at 530, calculating the focal point using the electrode position and the determined weighting corresponding to each electrode. The weighting may have been determined by step 430 of method 400 disclosed herein.

[0075] In some embodiments, the focal point may be the centre of the neural response signals detected at each electrode. That is, it may indicate a central point of the neural response signal or a point from which the neural response signal originates. The focal point may be determined by summarising neural recordings into a central point which provides insights into the distribution and directionality of the neural response signal around a DBS lead. In some embodiment, the focal point captures the spatial characteristics of the neural response signal. The focal point may be determined by selecting a waveform characteristic of the neural response signal, and determining weightings for each respective neural response signal received at each electrode of an electrode array. The focal point is calculated by multiplying the weightings with the spatial information of the electrodes (that is, the electrode position). In some embodiments, calculating the focal point includes summing over the multiplied weightings for each electrode. Calculating the focal point may further includemultiplying the weightings with the spatial information of the electrode for one or more individual coordinate values of the electrode with respect to the origin. The focal point may comprise coordinates relative to an origin. This may be the origin determined in 510 of method 500 disclosed herein, or it may be a point outside the electrode array. For example, the origin point may be defined relative to a target neural structure, or a reference electrode. The origin may be a point within the electrode array. In some embodiments, the origin may be the centre of the electrode array.

[0076] The focal point may comprise two-dimensional coordinates, for example, on an x-y plane. In some embodiments, the focal point may comprise three-dimensional coordinates, for example on an x-y-z plane. The focal point may itself be located within the electrode array. In some embodiments, calculating the focal point may further include summing over the determined weighting for each electrode of the plurality of electrodes. In some embodiments, a focal point representing the weighting of the neural response signal across an electrode array, such as a direction DBS lead, can be calculating using equation (1). The coordinates of the focal point may be converted into another coordinate space. For example, the coordinates of the focal point may be converted into a spherical coordinate space or a cylindrical coordinate space. In some embodiments, an offset may be applied to the coordinates of the focal point. The offset may be a spatial offset. The offset may be applied when, for example, a surgeon wants to implant and apply stimulation to a neural region that is offset from the neural region that generates a neural signal.

[0077] Figure 6 is a process flow diagram of a method 600 for calculating the focal point, according to some embodiments. At 610, the RMS waveform characteristics for each electrode are calculated. At 620, the RMS waveform characteristic values may be normalised between 0 and 1 across each electrode. The process for normalising the RMS waveform characteristic values may produce normalised values. At 630, the resulting normalised values may then be assigned as weights to each of the directional electrodes. It will be appreciated that normalising the RMS values in 620 is an optional step, and embodiments in which the RMS values are not normalised may be readilyused. In embodiments where the waveform characteristic values are not normalised, the waveform characteristic values may be used as weights and assigned to each of the directional electrodes in 630.

[0078] At 640, the coordinates of the directional electrodes may be assigned as the midpoint of the electrode surface. The midpoint of electrode surface may be defined with reference to the determined origin, and / or may use the spatial information about the electrode positions. The origin point (0,0) of the coordinate system is then set to the origin determined in 510 of method 500. In some embodiments, the origin point may be set to the centre of the electrode array or the DBS lead.

[0079] At 650 the coordinates of the focal point are then determined. The coordinates of the focal point may be determined using equation (1), wherein cx, cy, czare the coordinates of the focal point; xi, yi, zi are the coordinates of each electrode; pi are the weightings of the waveform characteristics applied to each electrode; and b is the number of electrodes. The electrode positions (xi, yi, zi) and corresponding weightings of the waveform characteristics (wi) may be used to calculate the coordinate components of the focal point from the origin, for example, from the centre of the electrode array. Equation (1) outputs the coordinate components (cx, cy, cz) of the focal point from an origin. In some embodiments, coordinate component czmay be used to calculate a depth of the neural response signal relative to a lead.

[0080]

[0081] The coordinates of the focal point output from equation (1) may then be used to calculate one or more indicator features. Indicator features are features derived from the neural response signal that provide additional information about the distribution of the neural response signal. Indicator features may provide features about the indicatorof the distribution of the neural response signal such as distance or direction. Indicator features may be used to determine information on an anatomical position of a lead relative to a target neural structure. Indicator features may be used by a machine learning model to determine the trajectory of a lead with respect to a target neural structure.

[0082] Using the coordinate components of the focal point, and treating them as a vector, a first indicator feature comprising a distance (r) can be calculated using equation (2) or equation (3). The distance may also be referred to as “range” or “magnitude". Equation (2) can be used to calculate the distance in 2 dimensions, where the distance is restricted to the axial plane of the lead, and equation (3) can be used to calculate the distance in 3 dimensions. Equations (2) and (3) represent the distance of the distribution of the waveform characteristics around the lead where, for example, a greater difference in the measured waveform characteristics of the neural response signal across the lead will result in a greater distance.

[0083] Using the coordinate components (ex, cy, cz) of the focal point determined from equation (1), and treating them as a vector, a second indicator feature comprising a direction (theta) can also be calculated relative to the origin in the axial plane of the lead by taking the inverse tangent of the ex and cy coordinate components using equation (4). The direction (theta) is the direction of the focal point relative to the origin in an axial plane.6 = tan-

[0084] A third indicator feature comprising a maximum RMS may be determined by measuring the RMS waveform characteristic across all electrodes of the electrode array and determining the maximum RMS waveform characteristic for the neural response signal. For example, where the waveform characteristic is amplitude, this may involve determined the maximum neural response signal RMS amplitude across each electrode of an electrode array.

[0085] Figure 7 is an axial view of the lead 102, showing a determined indicator as a vector 700 with respect to the electrodes E2, E3 and E4, according to some embodiments. The indicator has been determined on the basis of the neural response signals shown in Figure 3. The indicator is represented as a vector 700, with a direction of 287 degrees and a magnitude of 0.3 towards the focal point. Such a representation of the indicator overlaid on an axial view of the lead may be used during intraoperative procedures, for example, as a visual representation of the indicator of the distribution of the neural response signal with respect to an implanted lead.

[0086] In some embodiments, there are provided methods of removing noise form the neural response signal prior to determining the indicator of the neural response signal. Figure 8A is a process flow diagram of a first method 800 for removing noise from the neural response signal recording, according to some embodiments. The method 800, includes, at 810, detecting a resting signal during a period of no stimulation. The resting signal may be a period of time in the range of 1 to 20 seconds which is recorded before a non-therapeutic stimulus is applied. In some embodiments, the resting signal may be 12 second. The resting signal is recorded at each electrode. At 820, a baseline is calculated from the resting signal for each electrode. At 830, the method includes removing noise from the neural response signal based on the calculated baseline. In some embodiments, this includes calculating an average RMS waveform characteristic for the resting period and subtracting from the neural response signal the average RMS value to remove any baseline noise contributing to the calculation of the indicator.

[0087] Figure 8B is a process flow diagram of a second method 850 for removing noise from the neural response signal, according to some embodiments. The method 850 includes, at 860, detecting a resting signal during a period of no stimulation. Then, at 870 a resting waveform characteristic is extracted from the resting signal. The resting waveform characteristic may correspond to an extracted waveform characteristic. At 880, the resting waveform characteristic is removed from an extracted waveform characteristic to remove baseline noise. In some embodiments, methods 800 and 850 may be performed as part of step 420 of method 400 disclosed herein.

[0088] The inventors have determined that the calculation of an indicator of distribution of the neural response signal has practical applications in using the neural response signal for intraoperative confirmation. For example, to confirm that a DBS lead is accurately implanted relative to a target neural structure. One embodiment of the present disclosure provides a system and method for localising a lead tip within a target neural structure of the brain using an indicator of a distribution of neural response signals, such as ERNA or HFO activity. During an operation for implantation of the lead tip into the brain, instead of relying on low accuracy positioning techniques as described above to estimate the location of electrodes relative to neural structures within the brain, the system may be used to provide real-time feedback to the surgeon based on a determined indicator of the distribution which is calculated from characteristics of the neural response signal detected by each electrode of the lead tip. The feedback may be used to more accurately position the lead tip within the target structure in three dimensions and to guide the direction of repositioning or reimplanting the electrodes along a different trajectory.

[0089] In some embodiments, there is provided a method of analysing signals received by a DBS lead. The method comprises detecting a respective neural response signal at each electrode of a plurality of electrodes of the DBS lead, and for each electrode, extracting one or more waveform characteristics from the neural response signal. The method further comprises determining a weighting for each electrode based on the one or more extracted waveform characteristics, and determining whether thelead is located within a target neural region based on the weighting of each electrode and the location of each electrode relative to the lead. In some embodiments, the method may further comprise applying stimulus to at least one electrode of the DBS lead to stimulate a neural response signal.

[0090] Figure 9 is a process flow diagram of a method 900 of analysing signals received by a DBS lead, according to some embodiments. The method 900 of analysing signals received by a DBS lead may be used to determine whether the DBS lead is located within a target neural region. The target neural region may comprise a target neural structure. In some embodiments, the target neural region may further comprise a buffer around the target neural structure. The target neural region may define a region in which the DBS lead is considered to be satisfactorily or accurately positioned. The target neural structure may be a structure within one or both hemispheres of the brain. The target neural structure may include, but is not limited to, the subthalamic nucleus (STN) the substantia nigra pars reticulata (SNr) and globus pallidus interna (GPi). At 910, a stimulus is applied to at least one electrode of the DBS lead. 910 may be an optional step within method 900, and in some embodiments the stimulus may be applied from another source, such as one or more electrodes on a secondary DBS lead. At 920, a respective neural response signal is detected at each electrode of a plurality of electrodes of the DBS lead. For each electrode, at 930, one or more waveform characteristics are extracted from the neural response signal. In one example, the extracted waveform characteristic may be an amplitude of the neural response signal.

[0091] At 940, a weighting for each electrode is determined based on the one or more extracted waveform characteristics. At 950, it is determined whether the DBS lead is located within a target neural region based on the weighting of each electrode and the location of each electrode relative to the lead. In some embodiments, the method may further comprise determining a focal point of the neural response signals based on the location of each electrode relative to the lead and the weighting of each electrode. In some embodiments, the method further comprises obtaining a position of the lead relative to the target neural region based on the focal point. Responsive to determiningplacement of the lead is not within the target neural region, the method may provide information on relative proximity and / or relative direction of the lead with respect to the target neural region based on the determined focal point. In some embodiments, this may be in the form of a graphical representation which indicates the direction and proximity of the lead relative to a determined focal point of the neural response signal. In some embodiments, the method may further comprise determining whether the lead is located within the target neural region when there is an overlap between at least part of the lead and at least part of the target neural structure.

[0092] Figure 10 is a process flow diagram of a method 1000 for implanting a DBS lead, according to some embodiments. The method 1000 utilises the determination of an indicator of distribution of a neural response signal in the form of a focal point, in order to determine whether the lead should be repositioned in the x-y plane with reference to a target neural structure. The method 1000 describes a general exemplary surgical workflow that may be used for implanting a DBS lead. At 1010, a DBS lead having a plurality of electrodes is implanted according to preoperative planning. The plurality of electrodes may be advanced towards a target neural structure along a predefined trajectory in accordance with preoperative planning. The step size (or spatial resolution) by which the electrode lead is advanced may be chosen by the surgeons and / or clinicians. In some embodiments, the step size is 1 mm.

[0093] At 1020, respective neural response signals are detected and recorded at each electrode on the DBS lead. The neural response signal may be an evoked neural response signal, for example, such as ERNA. The neural response signals may be detected and recorded by applying a non-therapeutic stimulus at 1015. The stimulus may be applied for the all or part of the time that the lead tip is being implanted. The evoked response may be measured by the same electrode(s) as that used to apply the stimulus, and / or it may be measured at one or more different electrodes on the lead tip. That is, a neural response signal may be recorded from electrodes neighbouring (or distant to) the stimulated electrode as well as the stimulated electrode itself.

[0094] At 1030, an indicator in the form of a focal point is determined for the neural response signal. For example, the indicator may be determined using methods 400, 500 and / or 600 as disclosed herein. At 1040, the focal point is used to determine whether adjustment should be made to the depth of the electrode. In some embodiments, the depth adjustment may be dorsal or ventral. Responsive to determining that a depth adjustment should be made, at 1050 a depth adjustment is performed. The method then reverts back to 1020 to detect and record new neural response signals at the new depth, and confirm that no further depth adjustments are required. Steps 1020 to 1040 are repeated until the electrode lead tip has been inserted to the maximum allowable depth, which may be in the target neural structure or slightly beyond, or until the focal point indicates that no further adjustments should be made to the depth of the electrode.

[0095] Responsive to determining that no depth adjustments are required, at 1060, the focal point is used to determine whether the lead is positioned within a target neural region. In some embodiments, the target neural region indicates a satisfactory or acceptable position for the lead. In some embodiments, the focal point may guide or indicate that the DBS should be shifted in the x-y plane with respect to the lead. For example, the focal point may suggest that the DBS lead should be shifted in the mediolateral and / or anteroposterior plane with respect to the lead. In some embodiments, the focal point may be used to generate a graphical representation for display to a surgeon and / or clinician which indicates the direction in which the lead should be shifted and / or the amount to which the lead should be shifted in that direction.

[0096] Responsive to determining that an adjustment in the x-y plane should be made to the DBS lead, at 1070 a new DBS lead is inserted through a new trajectory. In some embodiments, the rather than inserting a new lead, the original DBS lead may be removed and reimplanted. The new trajectory may be guided by the determined focal point of the neural response signal which provides an indicator of the distribution of the neural response signal. After the insertion of a new DBS lead in the new trajectory, the method reverts back to 1020 to detect and record new neural response signals for thenew DBS lead, and confirm no further shifts are required for the DBS lead. In some embodiments, where two or more DBS leads are implanted in the brain, stimulation may be provided through one or more electrodes on one lead, and the neural response may be measured from a different lead. For example, the stimulation of 1015 may be provided by one or more electrodes of an originally planted lead, but the neural response signals detected at 1020 may be measured by one or more electrodes of the newly planted lead. In some embodiments, two or more electrode arrays may be implanted simultaneously. For example, if the first implanted DBS lead is determined to be suboptimal, a second lead that is 2mm lateral to the first may be implanted while keeping the first lead in place. An assessment of the second lead may be performed simultaneous to the first lead in order to determine which of the two leads is better, and which should be kept permanently implanted. In some embodiments, a neural response signal may be detected and recorded at both DBS leads, irrespective of which lead contains the stimulated electrode.

[0097] Steps 1020 to 1060 are repeated until the determined focal point indicates that no further adjustments in the x-y plane should be made to the DBS lead and that the DBS lead has been positioned within a target neural region. In some embodiments, the lead may be determined to be within the target neural region, and therefore in a satisfactory or acceptable position when the coordinates of the focal point are determined to be within the lead, or the coordinates of the focal point are determined to be within a predetermined range of the lead. In some embodiments, the lead may be determined to be acceptably positioned when a machine learning model determines that the focal point suggests the lead has an “acceptable” trajectory. Responsive to determining that the lead has been positioned in the target neural region, at 1080 the DBS lead is fixed in position.

[0098] Figure 11 is a schematic illustration of example recordings (A) and (B) of a neural response signal in different anatomical positions, according to some embodiments. Recordings (A) and (B) may be example recordings measured during method 1000. By determining a focal point of the neural response signal recordings of(A), and analysis of the neural response signal predicted an anterolateral direction for the centre of the neural response signal. This is represented by arrow 1100 of axial representation 1102 located above the graph representations in (A). The focal point from the recordings in (A) are determined to indicate that the lead is not positioned within a target neural region. That is, the focal point indicates that the lead could be moved towards the centre of the neural response signal. Accordingly, the neural response signal recordings of (B) are produced as a result of shifting the lead 2mm in the anterolateral direction. In (B), the graphs depict the neural response signal recordings after the 2mm anterolateral move. As can be seen, the amplitude of the neural response signals are much greater and / or indicate a greater relative strength, and can be used to indicate an anatomically acceptable location of the lead, which may be confirmed with post-operative neuroimaging.

[0099] In some embodiments, determining whether the lead is located within a target neural region comprises processing the extracted waveform characteristics of the respective neural response signals at each electrode and / or the location of each electrode to generate two or more indicator features for input into a machine learning model. Once the indicator features are generated, the two or more indicator features are input into the machine learning model and the machine learning model is configured to output an indication of the location of the lead relative to the target neural region. The two more indicator features may comprise a maximum extracted waveform characteristic, a vector distance of the focal point of the neural response signal and a vector direction of the focal point of the neural response signal.

[0100] In some embodiments, determining whether the lead is located within a target neural region comprises applying a machine learning model to determine whether the lead is located within target neural region. For example, a machine learning model may be applied to determine whether the lead is located in an anatomically acceptable or unacceptable position relative to a target neural structure. In some embodiments, the machine learning model may be utilised to provide recommendations or predictions asto a direction where a subsequent lead should be implanted based on predicting that the lead is in an unacceptable or insufficient position.

[0101] The inventors have developed a machine learning model, capable of distinguishing between anatomically acceptable and unacceptable lead positions. The machine learning model may be in the form of a binary classifier. The binary classifier may utilise a supervised K-Nearest Neighbour (KNN) method. The KNN method has distinct advantages due to its simplicity and performance in small datasets. In other embodiments, the binary classifier may utilise other supervised or unsupervised methods including, but not limited to, support vector machines (SVMs), Naive Bayes, Nearest Neighbour, Decision Trees, Logistic Regression and / or neural networks.

[0102] In some embodiments, the binary classifier is trained to predict the acceptability of lead positions with reference to a target neural structure. The binary classifier may be configured to take inputs in the form of one or more indicator features, labels, and hyperparameter ranges as input. The one or more indicator features may include the maximum extracted waveform characteristic, a vector distance of the focal point of the neural response signal and / or a vector direction of the focal point of the neural response signal. In one example, where the extracted waveform characteristic is amplitude of the neural response signal, the one or more indicator feature to be input into the machine learning model may include a maximum RMS amplitude, a distance of the focal point calculated using amplitude, and a direction of the focal point calculated using amplitude. However, it will be appreciated that additional features relating to the indicator of the distribution of the neural response signal and / or the focal point of the neural response signal may be used and / or input as an indicator feature.

[0103] The training data for the binary classifier may comprises intraoperative neural response signal recordings corresponding to one or more electrodes on an electrode array of a DBS lead having an individual trajectory. Each trajectory is labelled, and the labels are used as an input to the binary classifier. Labels comprise an identifyingelement that describes whether the training data is an acceptable trajectory with respect to the target neural structure. With respect to the training data used for the binary classifier, each trajectories which formed the training data was assigned one of three labels: (a) “acceptable”, (b) “unacceptable”, or (c) “uncertain”. Trajectories that were subsequently shifted to a new trajectory following standard surgical procedure were assigned the label “unacceptable”.

[0104] The “uncertain” label may be assigned where image quality is inadequate to clearly define the borders of the STN. Trajectories assigned the “uncertain” label may be excluded from the training data. The “unacceptable” label may be assigned where the lead electrode tiers at and below the target neural structure, for example, if most of the lead diameter lies outside the boundary of the target neural structure, or if the lead is greater than a predetermined threshold from the centroid (or central point) of the target neural structure. The “acceptable” label may be assigned if the above “unacceptable” conditions are not met, for example, if the lead is less than a predetermined threshold from the centroid of the target neural structure, and / or if the lead is located at least partially within a target neural region. For example, the lead may be assigned an “acceptable” label when there is a predetermined level of overlap between at least part of the lead and at least part of the target neural structure. The “Acceptable” label may be considered the positive class by the binary classifier. In some embodiments, the labels may be assigned manually, for example, by a neurologist (optionally blinded to participant information) reviewing the anatomical positioning of the leads. In some embodiments, the labels may be assigned automatically, for example, by using an automated recognition algorithm to identify the anatomical positioning of the leads.

[0105] Hyperparameters are different parameters of the machine learning model that specify how the machine learning model learns. Hyperparameters for the KNN method may include, but are not limited to, Number of Neighbours, Weight function, Algorithm, Leaf Size, and Metric. Examples of potential inputs for each of thehyperparameters are provided in Table 1. However, it will be appreciated that a number of other appropriate potential inputs may be used.

[0106] Table 1:

[0107] Much like feature selection, techniques employed in the development and assessment of the binary classifier may be chosen to provide realistic performance given a relatively small dataset. Accordingly, a number of features may be employed to reduce bias and overfitting in the machine learning model.

[0108] A nested cross-validation (CV) method may be utilised to optimise and evaluate the ability of the binary classifier to predict the acceptability of a lead’s position in unseen data. In a nested CV, optimisation of the classifier hyperparameters is nested inside the outside classifier assessment layer. Each training fold of the outer CV is passed into the inner layer for hyperparameter optimisation through an exhaustive grid search method, which outputs a classifier with optimal hyperparameters for the training data provided. The resulting classifier is then applied to the testing fold in the outer layer. In a nested CV approach, hyperparameter optimisation is not given the opportunity to overfit to the dataset as only a subset of the dataset is passed into the inner CV provided by the outer CV. This results in a less biased estimate of classifier performance. An added benefit of this method is that training and testing utilise the entire dataset, maximizing the potential performance in the case of a limited dataset length. Further, a stratified shuffled 5-fold CV may be utilised for each layer of thenested CV. In a stratified shuffled CV, the dataset is first randomly shuffled before being divided into 5 folds where each fold maintains the same proportions of the labels observed in the whole dataset. This approach ensures each fold approximates the population, maintaining that neither label is over-represented or under-represented in the training and testing splits of the data. This benefits classifier training in the case of an imbalanced dataset by maintaining realistic proportions and reducing the likelihood of classifier bias towards one label over another.Additionally, optimisation of the classifier within the inner CV may be set to maximize the balanced accuracy score. Being the arithmetic mean of true positive and true negative rates, balanced accuracy avoids potential performance score inflation in imbalanced datasets where the classifier may predict one label substantially more reliably than another. Balanced accuracy applies equal weighting to both labels in the dataset, combatting potential overfitting. Lastly, hyperparameter ranges fed into the grid search optimisation layer of the nested CV may be restricted to prevent the output classifier from being overly complex. This may be implemented to ensure that resulting classifier performance estimates were generalisable and not able to overfit to the training dataset with complex model parameters.

[0109] Figure 12 is a schematic diagram of a nested cross-validation method 1200, according to some embodiments, utilised to evaluate the binary classifier. The nested CV 1200 consists of an outer assessment 1202 and an inner optimisation level 1204. The inner CV 1204 is configured to perform model optimisation. The outer CV 1202 splits the dataset into training sets and testing sets. The outer assessment CV 1202 uses a stratified 5-fold shuffled CV, where four of the folds 1206 are used as training data and passed into the inner optimisation CV 1204 and the remaining fold 1208 is used as testing data. The inner CV xxx also used a stratified 5-fold shuffled CV and an exhaustive grid-search method to select the optimum hyperparameters from a list (such as the lists provided in Table 1) to maximize the balanced accuracy of the classifier.

[0110] The optimised classifier trained using the 4 folds 1206 from the outer CV 1202 may then be tested on the remaining fold 1208 of the outer CV 1202. The outer CV 1202 may evaluate the classifier on balanced accuracy, positive predictive value (PPV), and negative predictive value (NPV). Scores may then be averaged for each of the 5 divisions of the dataset of the outer CV 1202. In some embodiments, the nested CV 1200 may be repeated at least 10 times with the final scores being an average of the at least 10 repetitions. Utilising a nested cross-validation method provides technical advantages by providing a generalisable estimate of model performance be preventing overfitting and information leakage. Further, using the classifier with the optimised hyperparameters, enables feature importance of the indicator features to be measured using a feature permutation test and scoring for balanced accuracy. In some embodiments, the feature permutation test may be repeated at least 20 times.

[0111] The indicator features derived from the neural response signal and used in the machine learning model may include maximum RMS amplitude, centre-of-RMS distance, and centre-of-RMS direction. These features may be selected as they advantageously contain information on both the amplitude and the distribution of the neural response signal across the directional DBS lead. The distribution of the neural response signal may be used to determine a lead’s anatomical position relative to a target neural structure, such as the dorsolateral STN.

[0112] The maximum RMS waveform characteristics, such as amplitude, may be used as it advantageously increases or decreases with proximity to the target neural structure. Further, the centre-of-RMS based features of distance and direction may be used as they advantageously provide a representation of the neural response signal’s distribution recorded from the directional electrodes of the DBS lead. Centre-of-RMS distance represents how heavily ‘weighted’ the neural response signal’s waveform characteristic, such as amplitude, is across the directional electrodes of the DBS lead. An increased weighting of the neural responses signal’s waveform characteristic on one side of the lead may translate to a greater distance and inform the machine learning model on the lead’s proximity to the target neural structure. Centre-of-RMS directiondescribes the direction of this weighting outwards from the lead and may be used to provide additional information on lead position.

[0113] In some embodiments, converting the coordinate components of the focal point into a distance and direction provide technical advantages. First, a distance and direction of the distribution of the neural response signal are explainable and easily understood by a user reviewing the input data. Such features can be intuitively visualised for review. For example, the graphical representation 1102 provided in Figure 11 provides an intuitive and easy to understand visual guide. Further, the derivation of the distance and direction as indicator feature benefits performance of the machine learning model disclosed herein. The machine learning model does not have the capability to infer higher dimensional relationships between indicator features, and this would have been necessary if only the focal point coordinate components were included as indicator features. However, by deriving the focal point distance and direction for use in the machine learning model, this improves the explainability of input features and provides an improved model performance.

[0114] Feature importance was determined for each of the three metrics through the random shuffling of values of each feature and observing the degradation of the model performance. The maximum RMS had the highest importance with a mean decrease in the balanced accuracy of 0.42 ±0.03. Both the centre-of-RMS distance (0.21 ±0.03) and direction (0.20 ±0.02) had similar importance in contributing to the machine learning models performance.

[0115] The determined feature importance values reflect the level of contribution each of the neural response signal features had to the final machine learning model performance. In some embodiments, the maximum RMS waveform characteristic may hold the greatest importance to performance. In some embodiments, this confirms that the maximum waveform characteristic value of the neural response signal is related to an anatomically acceptable implantation location. The focal point based indicator features also provided useful information contributing to machine learning modelperformance, confirming that the distribution of the neural response signal’s waveform characteristics across the directional electrodes varies with the lead’s position relative to the target neural structure. That is, there is a relationship between the one or more waveform characteristics of the neural response signal and the proximity of the lead to a target neural structure. In some embodiments, the indicator features are derived to harness this relationship of the neural response signal across directional electrodes.

[0116] Distilling the neural responses signal recordings from each electrode of the electrode array into three indicator features provides several advantages. The indicator features generated provide the machine learning model with information that it would otherwise no be capable of inferring id only raw data were included, such as the spatial relationship between each of the recording electrodes along the lead. The features are also generalisable to a variety of lead designs, as focal point features can be adjusted to any lead geometry, including high channel count designs, for example, the Boston Scientific Cartesia X™ / HX™ 16-channel leads.

[0117] A common challenge in machine learning applications and the injection of many raw signal features during training is that the resulting model becomes a ‘black box’ in which it is not fully understood what aspects of the input contribute to the output. The inexplainable nature of some implementations minimises their appeal for adoption in the clinical environment. The machine learning model disclosed herein utilises readily explainable indicator features, in which the output of the machine learning model is a result of indicator features that can be easily interpreted by visualizing the raw neural response signal waveforms recorded across the lead.

[0118] Furthermore, limiting the number of features maintains the generalisability of the machine learning model and reduces the likelihood of overfitting. Given the relatively small dataset on which the machine learning model is trained, to avoid the curse of dimensionality, it was important to minimize the number of features used in classifier training. The initial neural response signal recordings were condensed across all 8 electrodes into 3 information dense indicator features for use in the machinelearning model training. In some embodiments where larger training datasets are used, additional features could be added from the neural response signal. Although examples described herein refer to neural response signal indicator features that are based on amplitude, it will be appreciated that indicator features may be based on other waveform characteristics such as frequency and decay rate. In some embodiments, larger datasets may allow for greater complexity in machine learning approaches that could infer higher dimensional relationships between neural response signal waveform characteristics.

[0119] Using the three features calculated from the neural response signal comprising ERNA recorded on 8-channel directional DBS leads, the inventors determined that the KNN classifier predicted an acceptable or unacceptable trajectory with a mean balanced accuracy of 82.0% (±2.6 percentage points standard deviation). Mean positive and negative predictive values achieved were 84.6% (±2.7 percentage points) and 81.8% (±2.7 percentage points) respectively.

[0120] Some embodiments of the present disclosure provide a system for monitoring neural activity in a brain. In some embodiments, the neural activity may be responsive to a stimulus in the brain. The system may comprise a lead having an electrode array, the lead adapted for implantation in or proximate to a target neural structure in the brain. The system may further comprise a selectively coupled to one or more electrodes of a plurality of electrodes of the electrode array, the configured to detect a respective neural response signal at each electrode of the plurality of electrodes. The system may further comprise a processing unit coupled to the and configured to: for each electrode, extract one or more waveform characteristics from the neural response signal; determine a weighting for each electrode based on the one or more extracted waveform characteristics; and determine an indicator of the distribution of the neural response signals among the plurality of electrodes of the electrode array based on at least the determined weighting for each electrode. In other embodiments, the system may comprise a processing unit coupled to the and configured to: for each electrode, extract one or more waveform characteristics from the neural response signal; determine aweighting for each electrode based on the one or more extracted waveform characteristics; and determine whether the lead is located within a target neural region based on the weighting of each electrode and the location of each electrode relative to the lead. In some embodiments, the system is configured to perform one or more of the methods 400, 500,600, 800, 850, 900, and / or 1000 disclosed herein.

[0121] An example DBS system 1300 for monitoring neural activity in the brain according to an embodiment of the present disclosure is illustrated in Figure 13. The system comprises the lead 102 of Figure 1 including the plurality of electrodes 106, 108a, 108b, 108c, 110a, 110b, 110c, 112, together with a processing unit 1302, a signal generator 1304, a measurement circuit 1306 and an optional switch matrix 1308. The processing unit comprises a central processing unit (CPU) 1310, memory 1312, and an input / output (VO) bus 1314 communicatively coupled with one or more of the CPU 1310 and memory 1312.

[0122] In some embodiments, the switch matrix 1308 is provided to control whether the electrodes 106, 108a, 108b, 108c, 110a, 110b, 110c, 112 are connected to the signal generator 1304 and / or to the measurement circuit 1306. The switch matrix 1308 may be configured to connect multiple signal paths at the same time. In some embodiments, the switch matrix 1308 may comprise a multiplexer. In other embodiments the switch matrix may not be required. For example, the electrodes 106, 108a, 108b, 108c, 110a, 110b, 110c, 112 may instead be connected directly to both the signal generator 1304 and the measurement circuit 1306. Although in Figure 13 all of the electrodes 106, 108a, 108b, 108c, 110a, 110b, 110c, 112 are connected to the switch matrix 1308, in other embodiments, only one or some of the electrodes 106, 108a, 108b, 108c, 110a, 110b, 110c, 112 may be connected.

[0123] The measurement circuit 1306 may include one or more amplifiers and digital signal processing circuitry including but not limited to sampling circuits for measuring neural responses to stimulation such as neural response signals, including ERNA. In some embodiments the measurement circuit 1306 may also be configured to extractother information from received signals, including local field potentials for measurement of HFOs and the like. The measurement circuit 1306 may also be used in conjunction with the signal generator 1304 to measure electrode impedances. The measurement circuit 1306 may be external to or integrated within the processing unit 1302. Communication between the measurement circuit 1306 and / or the signal generator 1304 on the one hand and the I / O port on the other may be wired or may be via a wireless link, such as over inductive coupling, WiFi (RTM), Bluetooth (RTM) or the like. Power may be supplied to the system 1300 via at least one power source 1316. The power source 1316 may comprise a battery such that elements of the system 1300 can maintain power when implanted into a patient.

[0124] The signal generator 1304 is coupled via the switch matrix 1308 to one or more of the electrodes 106, 108a, 108b, 108c, 110a, 110b, 110c, 112 and is operable to deliver electrical stimuli to respective electrodes based on signals received from the processing unit 1302. To this end, the signal generator 1304, the switch matrix 1308 and the processing unit 1302 are also communicatively coupled such that information can be transferred therebetween.

[0125] Whilst the signal generator 1304, switch matrix 1308, and the processing unit 1302 in Figure 13 are shown as separate units, in other embodiments the signal generator 1304 and switch matrix 1308 may be integrated into the processing unit 1302. Furthermore, either unit may be implanted or located outside the patient’s body.

[0126] The system 1300 may further comprise one or more input devices 1318 and one or more output devices 1320. Input devices 1318 may include but are not limited to one or more of a keyboard, mouse, touchpad and touchscreen. Examples of output devices include displays, touchscreens, light indicators (LEDs), sound generators and haptic generators. Input and / or output devices 1318, 1320 may be configured to provide feedback (e.g. visual, auditory or haptic feedback) to a user related, for example, to characteristics of ERNA or subsequently derived indicators (such as proximity of the electrode 102 relative to neural structures in the brain. To this end, one or more of theinput devices 1318 may also be an output device 1320, e.g. a touchscreen or haptic joystick. Input and output devices 1318, 1320 may also be wired or wirelessly connected to the processing unit 1302. Input and output devices 1318, 1320 may be configured to provide the patient with control of the device (i.e. a patient controller) or to allow clinicians to program stimulation settings, and receive feedback of the effects of stimulation parameters on ERNA and / or HFO characteristics.

[0127] One or more elements of the system 1300 may be portable. One or more elements may be implantable into the patient. In some embodiments, for example, the signal generator 1304 and lead 102 may be implantable into the patient and the processing unit 1302 may be external to the patient’s skin and may be configured for wireless communication with the signal generator via RF transmission (e.g. induction, Bluetooth (RTM), etc.). In other embodiments, the processing unit 1302, signal generator 1304 and lead 102 may all be implanted within the patient’s body. In any case, the signal generator 1304 and / or the processing unit 1302 may be configured to wirelessly communicate with a controller (not shown) located external to the patient’s body.

[0128] In some embodiments, the processing unit 1302 of system 1300 may be configured to execute computer-readable or machine readable media storing instructions to perform methods of monitoring neural activity. In some embodiments, there is provided a machine -readable medium storing instructions which, when executed by one or more processors of a system, cause the system to apply a stimulus to at least one electrode of the deep brain stimulation lead; detect a respective neural response signal at each electrode of a plurality of electrodes of the deep brain stimulation lead; for each electrode, extract one or more waveform characteristics from the neural response signal; determine a weighting for each electrode based on the one or more extracted waveform characteristics; and determine whether the lead is located within a target neural region based on the weighting of each electrode and the location of each electrode relative to the lead.

[0129] In some embodiments, there is provided a machine-readable medium storing instructions which, when executed by one or more processors of a system, cause the system to detect a respective neural response signal at each electrode of a plurality of electrodes of an electrode array implanted in the brain; for each electrode, extract one or more waveform characteristics from the neural response signal; determine a weighting for each electrode based on the one or more extracted waveform characteristics; and determine an indicator of the distribution of the neural response signals among the plurality of electrodes of the electrode array based on at least the determined weighting for each electrode. The machine-readable medium storing instruction may be non-transitory.

[0130] It will be appreciated by persons skilled in the art that numerous variations and / or modifications may be made to the above-described embodiments, without departing from the broad general scope of the present disclosure. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive.

Claims

CLAIMS:

1. A method of monitoring neural activity in a brain, the method comprising: detecting a respective neural signal at each of one or more electrodes of a plurality of electrodes of an electrode array implanted in the brain; for each of the one or more electrodes, extracting one or more waveform characteristics from the neural response signal; determining a weighting for each of the one or more electrodes based on the one or more extracted waveform characteristics; and determining an indicator of the distribution of the neural signals among the plurality of electrodes of the electrode array based on at least the determined weighting for each of the one or more electrodes.

2. The method according to claim 1, further comprising applying a stimulus to at least one electrode of the electrode array to evoke a neural signal.

3. The method according to claim 1 or claim 2, wherein determining an indicator comprises determining a ratio of the neural signals detected at different electrodes of the plurality of electrodes.

4. The method according to claim 1 or claim 2, wherein determining an indicator comprises determining a focal point of the neural signals based on spatial information of each electrode and the weighting corresponding to each electrode.

5. The method according to claim 4, wherein determining the focal point comprises: determining an origin; determining an electrode position with respect to the origin from the spatial information of each electrode of the plurality of electrodes; calculating coordinates of the focal point using the electrode position and the weighting corresponding to each electrode.

6. The method according to claim 4 or claim 5, further comprising determining the spatial information of each electrode of the one or more electrodes.

7. The method according to any one of claims 4 to 6, wherein the focal point is the centre of the neural signals.

8. The method according to any one of claims 4 to 7, wherein the focal point comprises coordinates relative to an origin.

9. The method according to claim 8, wherein the origin is a point within the electrode array.

10. The method according to claim 8 or claim 9, wherein the origin is the centre of the electrode array.

11. The method according to any one of claims 4 to 10, wherein the focal point comprises two-dimensional coordinates or three-dimensional coordinates.

12. The method according to any one of claims 4 to 10, wherein the focal point is located within the electrode array.

13. The method according to any one of claims 4 to 12, wherein the coordinates of the focal point are calculated using the equation:wherein: cx, cy, Cz are the coordinates of the focal point;Xi, yi, zi are the coordinates of each electrode; wi are the weightings of the waveform characteristics applied to each electrode; and b is the number of electrodes.

14. The method according to any one of claims 4 to 13, further comprising converting the coordinates of the focal point into another coordinate space.

15. The method according to claim 14, wherein the coordinate space is a spherical coordinate space or a cylindrical coordinate space.

16. The method according to claim any one of claim 4 to 15, further comprising applying an offset to the coordinates of the focal point.

17. The method according to any one of claims 4 to 16, further comprising calculating a distance and / or a direction to the focal point.

18. The method according to any one of claims 4 to 17, wherein calculating a distance (r) of the focal point comprises: treating the coordinates as a vector; and applying one of the following equations:r =19. The method according to any one of claims 4 to 18, wherein calculating a direction (<9) of the focal point comprises: treating the coordinates as a vector; and applying the equation:0 = tan1(— )Cy where <9 is the direction of the focal point relative to the origin in an axial plane.

20. The method according to any one of claims 4 to 19, further comprising obtaining a position of the lead relative to a target neural region in the brain based on the determined focal point of the neural signals.

21. The method according to any one of claims 4 to 20, further comprising calculating a distance, direction and / or depth of the focal point relative to the lead.

22. The method according to any one of claims 1 to 21, wherein the method further comprises: normalizing the one or more extracted waveform characteristics to produce one or more normalized values corresponding to the extracted waveform characteristics; and applying the one or more normalized values as the weighting for each electrode.

23. The method according to any one or claims 1 to 22, wherein one or more electrodes of the plurality of electrodes are segmented electrodes.

24. The method according to any one of claims 1 to 23, wherein the method further comprises determining a respective signal recording at each electrode of the one or more electrodes, each signal recording corresponding to the neural signal received by each electrode.

25. The method according to any one of claims 1 to 24, wherein the neural activity is evoked neural activity.

26. The method according to any one of claims 1 to 24, wherein the neural activity is spontaneous local field potentials.

27. The method according to any one of claims 1 to 24, wherein the neural activity is spontaneous single or multi-unit activity.

28. The method according to any one of claims 1 to 27, wherein the one or more waveform characteristics comprises one or more of: an amplitude, a frequency, a rate of change of frequency; a rate of change of amplitude; a rate of decay, a bandwidth, or a latency.

29. The method according to any one of claims 1 to 27, wherein the one or more waveform characteristics comprises a relative difference or ratio between one or more of an amplitude, a frequency, a latency, a rate of decay, or a rate of change.

30. The method according to any one of claims 1 to 29, wherein the one or more waveform characteristics are extracted over a period of time of the neural signal.

31. The method according to claim 30, wherein the one or more waveform characteristics are combined and / or averaged over the period of time.

32. The method according to claim 30 or claim 31, wherein the period of time is a subset of the length of the signal recording of the neural signal.

33. The method according to any one of claims 1 to 32, wherein the method further comprises: detecting a resting signal during a period of no stimulation; calculating a baseline from the resting signal; and removing noise from the neural signal based on the calculated baseline.

34. The method according to any one of claims 1 to 33, wherein the method further comprises: extracting a resting waveform characteristic from the resting signal; removing the resting waveform characteristic from the extracted waveform characteristic to remove baseline noise.

35. The method according to any one of claims 33 or 34, wherein the noise and / or the baseline noise comprise stimulation artefacts.

36. The method according to any one of claims 1 to 35, wherein the electrode array is disposed on a lead.

37. A method of analysing signals received by a deep brain stimulation lead, the method comprising: applying a stimulus to at least one electrode of the deep brain stimulation lead; detecting a respective neural signal at each of one or more electrodes of a plurality of electrodes of the deep brain stimulation lead; for each of the one or more electrodes extracting one or more waveform characteristics from the neural signal; determining a weighting for each of the one or more electrodes based on the one or more extracted waveform characteristics; and determining whether the lead is located within a target neural region based on the determined weighting of each of the one or more electrodes and the location of each of the one or more electrodes relative to the lead.

38. The method according to claim 37, wherein the target neural region comprises a target neural structure.

39. The method according to claim 37 or claim 38, wherein determining whether the lead is located within a target neural region comprises: processing the extracted waveform characteristics and location of each electrode to generate two or more features for input into a machine learning model, inputting the two or more features into the machine learning model; and outputting from the machine learning model an indication of the location of the lead relative to the target neural structure.

40. The method according to claim 39, the method further comprising determining a focal point of the neural signals based on the location of each electrode relative to the lead and the weighting of each electrode.

41. The method according to claim 40, wherein the two or more features comprise a maximum extracted waveform characteristic, a vector distance of the focal point of the neural signal and a vector direction of the focal point of the neural signal.

42. The method according to claim 40 or claim 41, the method further comprising obtaining a position of the lead relative to the target neural region based on the focal point.

43. The method of any one of claims 40 to 42, wherein responsive to determining placement of the lead is not within the target neural region, the method further comprises providing information on relative proximity and / or relative direction of the lead with respect to the target neural region based on the determined focal point.

44. The method of any one of claims 37 to 44, further comprising determining whether the lead is located within the target neural region when there is an overlap between at least part of the lead and at least part of the target neural structure.

45. A system for monitoring neural activity in a brain, comprising: a lead having an electrode array, the lead adapted for implantation in or proximate to a target neural structure in the brain; a measurement device selectively coupled to one or more electrodes of a plurality of electrodes of the electrode array, the measurement device configured to detect a respective neural signal at each electrode of the plurality of electrodes; and a processing unit coupled to the measurement device and configured to: for each electrode, extract one or more waveform characteristics from the neural signal; determine a weighting for each electrode based on the one or more extracted waveform characteristics; and determine an indicator of the distribution of the neural signals among the plurality of electrodes of the electrode array based on at least the determined weighting for each electrode.

46. The system of claim 45 wherein the system is configured to perform the method of any one of claims 1 to 44.

Images