CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The current application claims the benefit of and priority under 35 U.S.C. § 119(e) to U.S. Provisional Patent Application No. 63/090,680 entitled “Systems and Methods for Neuronavigation” filed October 12, 2020, the disclosure of which is hereby incorporated by reference in its entirety for all purposes.

FIELD OF THE INVENTION

[0002] The present invention generally relates to neuromodulation therapy, and (more specifically) to generating personalized stimulation targets.

BACKGROUND

[0003] Brain stimulation therapies can be delivered in a number of ways such as (but not limited to) transcranial magnetic stimulation (TMS) and deep brain stimulation (DBS). TMS. Brain stimulation therapies are often delivered at or towards a particular region of a patient’s brain in order to treat a condition of the patient.

[0004] Radiological imaging enables non-invasive scanning of internal organs. Common brain imaging techniques involve the use of magnetic resonance imaging (MRI) machines, and a variant of MRI referred to as functional MRI (fMRI) which is capable of measuring brain activity by measuring changes associated with blood flow. MRI, as opposed to fMRI, is often referred to as “structural” as it examines only the anatomy of the brain, and not brain activity.

SUMMARY OF THE INVENTION

[0005] Systems and methods for targeted neuromodulation in accordance with embodiments of the invention are illustrated. One embodiment includes a targeted neuronavigation system including a processor and a memory containing a targeting application, where the targeting application directs the processor to obtain patient brain data, where the patient brain data comprises a structural magnetic resonance imaging (sMRI) scan and at least one functional magnetic resonance imaging (fMRI) scan of a patient’s brain, map a reference region of interest (ROI) and at least one search ROI to the patient’s brain using the sMRI scan and at least one fMRI scan, where the reference ROI describes a region to be indirectly impacted via a brain stimulation therapy, and where the at least one search ROI describes at least one region to be directly targeted by the brain stimulation therapy, derive an individualized map of ROI parcellation, where the individualized map of ROI parcellation describes the reference ROI as a plurality of reference parcels, and describes the at least one search ROI as a plurality of candidate parcels, extract relationships between the plurality of candidate parcels and the plurality of reference parcels, calculate a target score for candidate parcels in the plurality of candidate parcels based on the extracted relationships, select a target parcel from the plurality of candidate parcels based on the target score, and provide the target parcel.

[0006] In another embodiment, the targeting application further directs the process to provide the brain stimulation therapy to the target parcel in order to treat a mental condition of the patient.

[0007] In a further embodiment, the mental condition is major depressive disorder.

[0008] In still another embodiment, the mental condition is suicidal ideation.

[0009] In a still further embodiment, the brain stimulation therapy is selected from the group consisting of: transcranial magnetic stimulation; transcranial direct current stimulation; and electrical stimulation delivered via an implantable electrostimulator.

[0010] In yet another embodiment, the targeting application further directs the processor to discard fMRI scans that deviate from expected whole brain network connectivity.

[0011] In a yet further embodiment, to derive the individualized map of ROI parcellation, the targeting application further directs the processor to randomly subsample voxels in the reference and at least one search ROIs, cluster the subsample of voxels, record a clustering assignment, and label clusters in the clustering assignment as candidate parcels or reference parcels based on their location.

[0012] In another additional embodiment, to derive the individualized map of ROI parcellation, the targeting application further directs the processor to randomly subsample voxels in the reference and at least one search ROIs as a first subsample of voxels, cluster the first subsample of voxels, record a first clustering assignment, randomly subsample voxels in the reference and at least one search ROIs as a second subsample of voxels, cluster the second subsample of voxels, record a second clustering assignment, merge the first clustering assignment and second clustering assignment using consensus clustering, and label clusters in the merged clustering assignment as candidate parcels or reference parcels based on their location.

[0013] In a further additional embodiment, to derive the individualized map of ROI parcellation, the targeting application further directs the processor to split spatially disjoint clusters.

[0014] In another embodiment again, the target score is calculated based on at least one factor from the group consisting of: parcel size, parcel depth, parcel shape, parcel homogeneity, functional connectivity strength to the reference ROI, and a network connectivity score.

[0015] In a further embodiment again, the network connectivity score reflects anticorrelation between a default mode network and a dorsal attention network of the patient’s brain.

[0016] In still yet another embodiment, a method of targeted neuronavigation includes obtaining patient brain data, where the patient brain data includes a structural magnetic resonance imaging (sMRI) scan and at least one functional magnetic resonance imaging (fMRI) scan of a patient’s brain, mapping a reference region of interest (ROI) and at least one search ROI to the patient’s brain using the sMRI scan and at least one fMRI scan, where the reference ROI describes a region to be indirectly impacted via a brain stimulation therapy, and where the at least one search ROI describes at least one region to be directly targeted by the brain stimulation therapy, deriving an individualized map of ROI parcellation, where the individualized map of ROI parcellation describes the reference ROI as a plurality of reference parcels, and describes the at least one search ROI as a plurality of candidate parcels, extracting relationships between the plurality of candidate parcels and the plurality of reference parcels, calculating a target score for candidate parcels in the plurality of candidate parcels based on the extracted relationships, selecting a target parcel from the plurality of candidate parcels based on the target score, and providing the target parcel. In many embodiments, obtaining patient brain data may be accomplished by accessing patient brain data that has previously been uploaded to or transmitted to the target identification system; requesting patient brain data from a remote institution, computer system, or database; or by accessing hardware such as MRI or other imaging hardware to cause acquisition of patient brain data.

[0017] In a still yet further embodiment, the method further includes providing the brain stimulation therapy to the target parcel in order to treat a mental condition of the patient. [0018] In still another additional embodiment, the mental condition is major depressive disorder.

[0019] In a still further additional embodiment, the mental condition is suicidal ideation. [0020] In still another embodiment again, the brain stimulation therapy is selected from the group consisting of: transcranial magnetic stimulation; transcranial direct current stimulation; and electrical stimulation delivered via an implantable electrostimulator.

[0021] In a still further embodiment again, the method further includes discarding fMRI scans that deviate from expected whole brain network connectivity.

[0022] In yet another additional embodiment, wherein deriving the individualized map of ROI parcellation includes randomly subsampling voxels in the reference and at least one search ROIs, clustering the subsample of voxels, and recording a clustering assignment, labeling clusters in the clustering assignment as candidate parcels or reference parcels based on their location.

[0023] In a yet further additional embodiment, wherein deriving the individualized map of ROI parcellation includes randomly subsampling voxels in the reference and at least one search ROIs as a first subsample of voxels, clustering the first subsample of voxels; recording a first clustering assignment, randomly subsampling voxels in the reference and at least one search ROIs as a second subsample of voxels, clustering the second subsample of voxels, and recording a second clustering assignment, merging the first clustering assignment and second clustering assignment using consensus clustering, and labeling clusters in the merged clustering assignment as candidate parcels or reference parcels based on their location.

[0024] In yet another embodiment again, deriving the individualized map of ROI parcellation further includes splitting spatially disjoint clusters.

[0025] In a yet further embodiment again, the target score is calculated based on at least one factor from the group consisting of: parcel size, parcel depth, parcel shape, parcel homogeneity, functional connectivity strength to the reference ROI, and a network connectivity score.

[0026] In another additional embodiment again, the network connectivity score reflects anticorrelation between a default mode network and a dorsal attention network of the patient’s brain.

[0027] In a further additional embodiment again, a system for treating major depressive disorder includes a transcranial magnetic stimulation device, a neuronavigation device, a processor, and a memory containing a targeting application, where the targeting application directs the processor to obtain patient brain data, where the patient brain data comprises a structural magnetic resonance imaging (sMRI) scan and at least one functional magnetic resonance imaging (fMRI) scan of a patient’s brain, map a reference region of interest (ROI) and at least one search ROI to the patient’s brain using the sMRI scan and at least one fMRI scan, where the reference ROI describes a region to be indirectly impacted via the transcranial magnetic stimulation device, and where the at least one search ROI describes at least one region to be directly targeted by the brain stimulation therapy, derive an individualized map of ROI parcellation, where the individualized map of ROI parcellation describes the reference ROI as a plurality of reference parcels, and describes the at least one search ROI as a plurality of candidate parcels, extract relationships between the plurality of candidate parcels and the plurality of reference parcels, calculate a target score for candidate parcels in the plurality of candidate parcels based on the extracted relationships, select a target parcel from the plurality of candidate parcels based on the target score, and apply transcranial magnetic stimulation to the target parcel using the transcranial magnetic stimulation device and/or neuronavigation device in order to treat major depressive disorder.

[0028] In yet another additional embodiment again, the target parcel is transmitted from a cloud computing platform to a neuronavigation system.

[0029] In still yet another additional embodiment, a method of treating major depressive disorder includes obtaining patient brain data, where the patient brain data comprises a structural magnetic resonance imaging (sMRI) scan and at least one functional magnetic resonance imaging (fMRI) scan of a patient’s brain, mapping a reference region of interest (ROI) and at least one search ROI to the patient’s brain using the sMRI scan and at least one fMRI scan, where the reference ROI describes a region to be indirectly impacted via a brain stimulation therapy, and where the at least one search ROI describes at least one region to be directly targeted by the brain stimulation therapy, deriving an individualized map of ROI parcellation, where the individualized map of ROI parcellation describes the reference ROI as a plurality of reference parcels, and describes the at least one search ROI as a plurality of candidate parcels, extracting relationships between the plurality of candidate parcels and the plurality of reference parcels, calculating a target score for candidate parcels in the plurality of candidate parcels based on the extracted relationships, selecting a target parcel from the plurality of candidate parcels based on the target score, and treating major depressive disorder by applying transcranial magnetic stimulation to the target parcel using a transcranial magnetic stimulation device and/or a neuronavigation device.

[0030] In still yet again another additional embodiment, the transcranial magnetic stimulation is accelerated theta burst stimulation.

[0031] Additional embodiments and features are set forth in part in the description that follows, and in part will become apparent to those skilled in the art upon examination of the specification or may be learned by the practice of the invention. A further understanding of the nature and advantages of the present invention may be realized by reference to the remaining portions of the specification and the drawings, which forms a part of this disclosure.

BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The description and claims will be more fully understood with reference to the following figures and data graphs, which are presented as exemplary embodiments of the invention and should not be construed as a complete recitation of the scope of the invention.

[0033] FIG. 1 illustrates a targeted neuromodulation system in accordance with an embodiment of the invention.

[0034] FIG. 2 illustrates a target generator in accordance with an embodiment of the invention. [0035] FIG. 3 is a flow chart illustrating a targeting process for generating an individualized target in accordance with an embodiment of the invention.

[0036] FIG. 4 is a flow chart illustrating a targeting process for evaluating expected network connectivity in accordance with an embodiment of the invention.

[0037] FIG. 5 is a flow chart illustrating a targeting process for deriving individualized ROI parcellations in accordance with an embodiment of the invention.

[0038] FIG. 6 is a flow chart illustrating a targeting process for splitting spatially disjoint clusters in accordance with an embodiment of the invention.

DETAILED DESCRIPTION

[0039] Mental health conditions and other neurological problems are a significant field of medicine with profound importance for both patients and society as a whole. For example, depression and suicidal ideation represent chronic public health issues. However, treatment for these conditions have conventionally been addressed with pharmaceuticals, and in some treatment resistant cases, using surgery and/or electroconvulsive therapy (ECT). These methods can have significant side effects that are both mental and physical. In contrast, a form of therapy called transcranial magnetic stimulation (TMS) has arisen as a viable non-invasive treatment option with minimal side effects reported.

[0040] TMS involves applying a magnetic field to a particular region of the brain in order to depolarize or hyperpolarize neurons at the target region. Generally, the target region is selected by a medical professional based on its relationship with the patient’s condition. For example, the dorsolateral prefrontal cortex (DLPFC) is known to be involved with major depressive disorder. However, the exact location of the DLPFC in an individual can be difficult to manually identify. Even when it can be identified, there may in fact be a particular subregion of the DLPFC which would be the most effective target for the individual patient based on their idiosyncratic brain. Further, there may even be other regions in the brain that would provide better stimulation targets for the patient. As every brain is at least slightly different, a personalized way of generating stimulation targets for an individual can provide better treatment outcomes. [0041] An additional limitation of many TMS devices is the depth at which they can induce a current in a patient’s brain. Often, TMS devices cannot target deep brain structures. However, there are numerous large-scale networks throughout the brain that have been identified. For example, the default mode network (DMN) is a network which appears to be involved with numerous tasks such as wakeful rest. By way of further example, the dorsal attention network (DAN) is thought to be key in voluntary orienting of visuospatial attention, and similarly the ventral attention network (VAN) reorients attention towards salient stimuli. Connectivity between different regions of the brain can provide an opportunity in TMS and other brain stimulation therapies whereby a more surface brain structure which is strongly connected to a deeper brain structure can be stimulated to effect change in the deeper brain region. Further, stimulation of connected networks can have significant impacts on structures within or otherwise connected to the network. Some networks in particular such as (but not limited to) the DMN, the DAN, and the VAN have particular experimentally determined relationships to major depressive disorder and suicidal ideation. Networks with relationships to a particular mental condition to be treated can be given additional priority.

[0042] Given the complex nature of the brain, when applying a neuromodulation therapy (like TMS), the location at which the stimulation is delivered can have a significant impact on the outcome of the treatment. Targeting as discussed herein refers to the process of identifying target structures within a patient’s brain for stimulation in order to treat mental health conditions. While current targeting methods can yield workable targets, many conventional methods have significant failings. For example, targeting often takes place using one scan from a patient and cannot incorporate multiple scans over time. Due to scanning noise and limited test-retest reliability of fMRI, deriving a target based on a single scan is more likely to be affected by noise and lead to a compromised levels of target reliability. Reliability limitation may be even more prominent for methods that employ voxel clustering for target detection, especially if clustering procedure is highly sensitive to noise and signal loss. Further, clustering procedures used for this purpose do not always consider the spatial relations between the voxels, which may lead to impractical results. Turning now to the drawings, systems and methods described herein seek to address these limitations, and provide a more robust targeting framework that produces more effective individualized stimulation targets for more effective treatment. In many embodiments, the targets produced using systems and methods described herein are subsequently used as the target in a neuromodulation therapy such as (but not limited to), TMS, transcranial direct current stimulation (tDCS), as the implantation location for one or more stimulation electrodes, and/or as the target for any number of different neuromodulation modalities as appropriate to the requirements of specific applications of embodiments of the invention. Targeting systems in accordance with embodiments of the invention are discussed below.

Targeted Neuromodulation Systems

[0043] Targeted neuromodulation systems are capable of obtaining and/or accessing scans of a patient’s brain, and identifying one or more individualized targets for brain stimulation therapy. In many embodiments, targeting systems may be integrated into other medical devices, such as (but not limited to) TMS devices or neuronavigation devices. In various embodiments, targeting systems not only can generate individualized targets, but also include or be integrated with neuronavigation devices to identify where a TMS coil should be placed to correctly stimulate the target. In many embodiments, targeted neuromodulation systems can further apply neuromodulation to the generated target via a neuromodulation device such as (but not limited to) a TMS device, a tDCS device, an implantable neurostimulator, and/or any other neurostimulation device as appropriate to the requirements of specific applications of embodiments of the invention. [0044] Turning now to FIG. 1 , a targeted neuromodulation system in accordance with an embodiment of the invention is illustrated. Targeted neuromodulation system 100 includes a target generator 110. Targeting generators can be implemented using any number of different computing platforms such as (but not limited to) desktop computers, laptops, server computers and/or clusters, smartphones, tablet PCs, and/or any other computing platform capable of executing logic instructions as appropriate to the requirements of specific applications of embodiments of the invention. In many embodiments, target generators determine personalized and/or partially-personalized targets within an individual’s brain. [0045] Targeted neuromodulation system 100 further includes an fMRI machine 120 and a TMS device 130. In many embodiments, the fMRI machine is capable of obtaining both structural and functional MRI images of a patient. The TMS device 130 can deliver brain stimulation therapy to the target selected by the target generator 110. However, as can readily be appreciated, alternative imaging modalities (e.g. computed tomography, positron emission tomography, electroencephalography, etc.), and alternative brain stimulation devices can be used (e.g. implantable stimulators) as appropriate to the requirements of specific applications of embodiments of the invention; alternatively, the targeting system 100 may not include its own imaging equipment, and may receive imaging or other brain data from one or more imaging systems that are distinct from the neuromodulation system 100.

[0046] In many embodiments, the targeted neuromodulation system 100 includes a neuronavigation device which guides delivery of brain stimulation therapy by TMS device 130 to a target selected by the target generator 110. This neuronavigation device may be integrated into the targeting generator 110 or separate (not shown) from the targeting system 110. In numerous embodiments, neuronavigation devices assist in delivering brain stimulation therapy to one or more targets generated by a targeting system; for instance, by determining the rotational and translational position of a stimulating coil and head and displaying an image to guide a user to position the stimulating coil correctly, or by additionally using a mechanical actuator such as a robotic arm to position the stimulating coil correctly. As can be readily appreciated the specific function of a neuronavigation device can be varied depending on the type of neuromodulation being applied.

[0047] In many embodiments, the fMRI, TMS device, targeting system, and/or neuronavigation device are connected via a network 140. The network can be a wired network, a wireless network, or any combination thereof. Indeed, any number of different networks can be combined to connect the components. However, it is not a requirement that all components of the system be in communication via a network. Target generators are capable of performing without operative connections between other components. Indeed, as can be readily appreciated, while a specific targeted neuromodulation system is illustrated in FIG. 1 , any number of different system architectures can be used without departing from the scope or spirit of the invention. For example, in many embodiments, targeted neuromodulation systems can include different neuromodulation devices that provide different stimulation modalities.

[0048] When targeting systems are provided with patient brain data, they are capable of generating individualized targets. Turning now to FIG. 2, a target generator architecture in accordance with an embodiment of the invention is illustrated. Target generator 200 includes a processor 210. However, in many embodiments, more than one processor can be used. In various embodiments, the processor can be made of any logic processing circuitry such as (but not limited to) central processing units (CPUs), graphics processing units (GPUs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), and/or any other circuit as appropriate to the requirements of specific applications of embodiments of the invention.

[0049] The target generator 200 further includes an input/output (I/O) interface 220. I/O interfaces are capable of transferring data between connected components such as (but not limited to) displays, TMS devices, fMRI machines, other treatment devices and/or imaging devices, and/or any other computer component as appropriate to the requirements of specific applications of embodiments of the invention. The target generator further includes a memory 230. The memory can be implemented using volatile memory, non-volatile memory, or any combination thereof. As can be readily appreciated, any machine-readable storage media can be used as appropriate to the requirements of specific applications of embodiments of the invention.

[0050] The memory 230 contains a targeting application 232. The targeting application is capable of directing the processor to execute various target generation processes. The memory 230 is also capable of storing patient brain data 234. Patient brain data describes brain scans of the patient such as, but not limited to, structural MRI and functional MRI scans. In numerous embodiments, the memory 230 can further contain normative connectivity data 236 describing expected generalized connectivity networks for a standard brain model.

[0051] While particular target generator architectures and target generators are discussed in accordance with embodiments of the invention above, any number of different architectures and hardware designs can be used without departing from the scope or spirit of the invention. For example, in many embodiments, different stimulation modalities can be used. In various embodiments, transcranial direct current stimulation is used. In numerous embodiments, implantable electrical neurostimulators are used to directly stimulate brain tissue. Target generation processes for generating individualized stimulation targets are discussed in further detail below.

Generating Individualized Stimulation Targets

[0052] Some brain stimulation methods will work with some degree of efficacy without individualized, precision targeting. However, providing stimulation to a particular region of the brain to attempt to maximize the impact of treatment for an individual is highly beneficial. Various existing methodologies that attempt to generate personalized targets fail to fully consider the existing network connectivity in the brain and/or naively cluster regions within the brain. Target identification processes described herein can provide higher accuracy stimulation targets for an individual based on their personal brain network connectivity.

[0053] Turning now to FIG. 3, a flow chart of a target identification process for generating an individualized stimulation target for a patient in accordance with an embodiment of the invention is illustrated. Process 300 includes obtaining (310) patient brain data. As noted above, patient brain data can include structural and/or functional brain scans. In many embodiments, patient brain data includes both a structural MRI and a functional MRI scan. In various embodiments, multiple structural and/or functional MRI scans are included in the patient brain data which may have been captured at different times. MRI scans can be checked for quality. In various embodiments, scan quality is examined using commonly used fMRI quality control (QC) tools, and/or by matching whole brain connectivity structure against expected normative connectivity structure. Target identification processes for performing quality control using expected normative connectivity structure are discussed in further detail in a below section with reference to FIG. 4.

[0054] Process 300 further includes mapping (320) search and reference regions of interest (ROIs) onto the patient’s brain. ROIs can be any brain structure, substructure, or group of structures of interest in the brain as decided by a user. Reference ROIs are ROIs that describe a region that the brain stimulation therapy should indirectly affect. In contrast, search ROIs describe regions in which individualized brain stimulation targets may reside. In this way, applying stimulation to an individualized brain stimulation target in a search ROI has an effect on the reference ROI. ROIs can be made up of one or more voxels depending on the size of the particular ROI. In some embodiments, ROIs may overlap. In numerous embodiments, a brain atlas is used to map ROIs onto a structural scan of the patient’s brain. In various embodiments, target ROIs are indicated by applying a mask to the brain structure, where the mask flags desired target ROIs. In various embodiments, the mask can have different weight metrics for different desired target ROIs. ROIs can also be mapped onto functional scans. In various embodiments, a structural scan can be used as a template to align other functional scans. In various embodiments, multiple fMRI scans can be combined by integrating functional connectivity data to yield a “combined fMRI”. In this way, multiple fMRIs taken of a patient with similar or identical protocols can be merged to yield a more complete picture of an individual’s network connectivity.

[0055] fMRI signals (i.e. activity levels for a particular voxel or set of voxels over time) are extracted (330) from the ROIs. Voxels with poor signal quality can be excluded (335) and/or discarded. In numerous embodiments, poor quality signal can be caused due to various scanner limitations, scanning parameters and/or movement during the scanning process. In various embodiments, poor quality signals are detected by calculating voxellevel signal-to-noise ratio (SNR). By removing low quality signals from consideration, targeting accuracy can greatly increase. An individualized map of ROI parcellation is derived (340) from the extracted fMRI signals. The individualized map of ROI parcellations describes multiple parcels (or groups of adjacent voxels). Candidate parcels are derived from search ROIs, and constitute candidate targets for brain stimulation therapy. Reference parcels are derived from the reference ROI, and constitute areas of the reference ROI which will be impacted by the stimulation. Methods for deriving ROI parcellations in accordance with embodiments of the invention are discussed in further detail below with respect to FIG. 5.

[0056] Relationships between potential candidate and reference parcels are extracted (350) and a target score for potential candidate parcels are generated (360). In many embodiments, the functional connectivity between two parcels (a candidate and a reference) is measured and the target score is based on the strength of the functional connection. A target which has a stronger functional connectivity to a reference ROI (e.g. any parcel within the reference ROI), and therefore impacts functioning of the reference more strongly, can be given a higher target score. In many embodiments, other factors contribute to the score including (but not limited to) parcel depth, other functions of the parcel and/or surrounding brain structures, size, shape, and homogeneity of the parcel, fit to known/expected system/network-level connectivity profile, as well as numerous other factors can be considered as appropriate to the requirements of specific applications of embodiments of the invention. For example, a larger target may not have as strong functional connectivity to the reference, but is much larger and therefore easier to target with a specific brain stimulation device.

[0057] By way of additional example, a network connectivity score can be included which incorporates network-level expectations regarding which brain region to target can be included. If, in the literature, the field believes that a particular brain structure or network (i.e. set of structures) is involved with a particular condition, parcels that interact strongly with that brain structure/network may be weighted more heavily as potential targets. As noted above, the DLPFC is believed to be strongly linked to clinical depression and suicidal ideation, and therefore targets that strongly interact with that region may be more desirable based on current expectations.

[0058] As an example, in numerous embodiments, for each parcel, the difference between the functional connectivity to the DAN and the DMN can be calculated. Anticorrelation between the DAN and the DMN can be used as the network connectivity score, where a higher degree of anticorrelation suggests a stronger candidate parcel. In various embodiments, the difference between functional connectivity to the VAN and the DMN is calculated and used as a network connectivity score. In some embodiments, a weighted average of the network connectivity scores for different networks can be used as an overall network connectivity score, where the weights are based on the relevance of particular networks to a condition at issue. In various embodiments, the functional connectivities are calculated on a per-voxel basis and averaged to get an overall parcel score. [0059] An individualized target parcel is then selected (370) from the group of candidate parcels based on the target scores. In many embodiments, the highest scored candidate parcel is selected. In many embodiments, the center for the target parcel is extracted (380) in order to more precisely determine TMS coil alignment. In many embodiments, the center is calculated by averaging the position of each voxel making up the target candidate.

[0060] While a particular method for generating an individualized target is illustrated in FIG. 3, as can be readily appreciated, any number of different modifications can be made without departing from the scope or spirit of the invention. For example, not every quality control step needs to be taken or every parameter considered for generating a network score as appropriate to the requirements of specific applications of embodiments of the invention. Further, different weights may be given to different parameters as to their relative importance in calculating a target score. Additional description of various steps of the above processes are found below.

Network Connectivity Quality Control

[0061] Patient brain data can include one or more fMRI scans, however there is rarely an immediate guarantee that the data is high quality (e.g. having a high SNR). Measurement noise and head movement are known causes of fMRI reliability limitation and are thus estimated and partially addressed as common practice during data preprocessing. However, in some cases, poor scan quality and or preprocessing errors are missed which can lead to deriving a target based of faulty brain functional connectivity structure. To prevent making clinical decisions based on faulty data, additional means are desirable.

[0062] Under the probable and acceptable assumption of overall preservation in system-level organization of the human brain, matching measured whole-brain connectivity against expected normative connectivity can reduce errors from bad scans and, in some cases, provide a flag to medical professionals the presence of atypical brains for further manual scrutiny. In many embodiments, identified bad scans are discarded. Turning now to FIG. 4, a target identification process for measuring expected network connectivity in accordance with an embodiment of the invention is illustrated. [0063] Process 400 includes assigning (410) each voxel to a predefined network. Many large-scale brain networks are known and have been mapped based on large samples of the population such as (but not limited to) the visual network (VIS), the sensorimotor network (SMN), the dorsal attention network (DAN), the ventral attention network (VAN), the limbic network, the frontoparietal control network (FPCN), and the default mode network (DMN). These networks can be overlaid onto an MRI of a patient such that each voxel is assigned to at least one network. For each voxel pair, a functional connectivity score (FC) can be calculated (420), where the FC represents the strength of the connectivity between the voxels in an fMRI (including a combined fMRI). All of the FC values that link voxels that are assigned to the same network are averaged (430) to yield a “within FC” value.

[0064] A “between FC” value is obtained by averaging (440) all FC values that link voxels from different networks. The between FC value is subtracted (450) from the within FC value to obtain a network fit for the voxel. While individual voxels may vary in their network association due to expected individual differences in brain function and structure, the average network fit across voxels (termed network quality control (QC) metric) is expected to remain positive (within FC > between FC) If the network QC metric is not significantly positive (mean between FC >= within FC), it is an indicator that there may be something either wrong with the scan, the preprocessing procedure or a significantly atypical structural issue occurring within the patient’s brain. A statistical significance of network QC metric can be obtained by randomly permuting the data while considering voxel spatial positions and repeating the network QC estimation process. In this way, intake fMRIs can be cleared for quality. If an fMRI scan is flagged as having a poor overall network fit it can allow detailed inspection by a medical professional of the data and prevent deriving a target from faulty information.

[0065] While a particular method for QC control based on brain network connectivity is illustrated in accordance with an embodiment of the invention in FIG. 4, network connectivity can be used as a control using any of a number of different algorithms as appropriate to the requirements of specific applications of embodiments of the invention. Ensuring quality data can increase the accuracy of generated targets. A discussion of how to parcellate the brain into individualized ROIs is discussed further below. ROI Parcellation

[0066] It is well known that while the overall structure of the human brain is relatively conserved across individuals, each person has idiosyncratic brain functionality and circuitry based on any number of factors both environmental and genetic. As such, merely dividing the brain based on a standardized model can yield inaccurate or insufficient results. While previous attempts have been made at parcellating the brain into ROIs, the particular methodologies used have often failed to robustly cluster voxels in an effective manner. Turning now to FIG. 5, a target identification process for deriving an individualized map of ROI parcellation in accordance with an embodiment of the invention is discussed.

[0067] Process 500 includes randomly subsampling (510) a percentage of all voxels. In many embodiments, the percentage is any number greater than 80%, however depending on the amount of data and compute available, this number can be less than 80%. The fMRI signals within the subsampled voxels are then clustered based on signal similarity (520). Any number of different clustering processes can be used including (but not limited to) agglomerative (hierarchical) clustering, Cluster Identification via Connectivity Kernels (CLICK) clustering, k-means clustering, and/or Spectral clustering. In some embodiments, clustering methods that incorporate spatial information (e.g. spatially constrained spectral clustering) can be used.

[0068] The clustering assignment is recorded (530) and a new random subsampling (510) is obtained. The process can be repeated many times to increase accuracy. In many embodiments, this process is repeated 100 or more times to ensure enough data, although fewer can suffice. The subsample clustering solutions are then merged (540). In many embodiments, they are merged using a consensus clustering approach. Any resulting spatially disjoint clusters can then be split (550) into sub-clusters. The clusters (and any sub-clusters) are then labeled (560) as parcels, either reference or search based on their locations within reference and search ROIs.

[0069] By repeatedly subsampling and clustering, noise in the neural signals can be accounted for and a more accurate picture of the individual’s true brain connectivity can emerge. Furthermore, multiple fMRI scans can be run through this process and the resulting clusters can be integrated using consensus clustering. In this way, multiple fMRIs, including those taken on different days, can contribute to the overall dataset used for targeting. In various embodiments, spatially disjoint clusters can be avoided by using a spatially constrained clustering process. However, depending the requirements of specific applications of embodiments of the invention, it may be desirable to select a spatially unconstrained clustering process, which may yield spatially disjoint clusters. A target identification process for splitting spatially disjoint clusters in accordance with an embodiment of the invention is illustrated in FIG. 6.

[0070] Process 600 includes recording (610) the spatial position of each voxel in the spatially disjoint cluster. A distance matric indicating the physical distance between every two voxels is generated (620) which is then converted (630) into a graph representation. Long edges in the graph (edges that exceed a predefined threshold) are pruned (640) to yield a partially connected graph that is then split into connected sub-graphs (components) if such emerge. The set of voxels in each connected component can then be defined as a separate cluster (650).

[0071] In this way, a disjoint cluster can be split and separately used as potential candidate parcels for stimulation. In many embodiments, these disjoint clusters are problematic the “center” of a disjoint cluster may be outside any part of the disjoint cluster and nowhere near a viable target location.

[0072] Processes 300, 400, 500, and/or 600, and their variations, may be performed by a target identification system in order to provide a target parcel which then may be archived, stored for later use, transmitted to a neuronavigation device, used in further analysis, or combined with one or more other target parcels (for example, by union or intersection) to yield a composite target parcel. The target identification system may be distinct from, separate from, and/or integrated or partially integrated with a neuronavigation device. The target identification system may be implemented on a cloud computing platform, on a computing platform local to the site of treatment, on a computing platform incorporated into or part of a neuronavigation device, or any combination of such platforms.

[0073] Although specific methods of ROI parcellation are discussed above, many different methods can be implemented in accordance with many different embodiments of the invention, such as (but not limited to) those that use different specific clustering processes, and/or utilize different thresholds and parameters. It is therefore to be understood that the present invention may be practiced in ways other than specifically described, without departing from the scope and spirit of the present invention. Thus, embodiments of the present invention should be considered in all respects as illustrative and not restrictive. Accordingly, the scope of the invention should be determined not by the embodiments illustrated, but by the appended claims and their equivalents.