A method for using neural activity to assess clinical response after deep brain stimulation for psychiatric disorders

Abstract

The present disclosure pertains to a method of assessing a psychiatric disorder in a subject by (1) measuring a neural activity of the subject's brain; and (2) classifying the psychiatric disorder based on the measured neural activity. The methods may also include a step of implementing a treatment based on the classification of the psychiatric disorder. The present disclosure also pertains to systems for assessing a psychiatric disorder in a subject in accordance with the methods of the present disclosure.

Description

TITLEA METHOD FOR USING NEURAL ACTIVITY TO ASSESS CLINICAL RESPONSE AFTER DEEP BRAIN STIMULATION FOR PSYCHIATRIC DISORDERSSTATEMENT REGARDING FEDERALLY SPONSORED RESEARCH

[0001] This invention was made with government support under UH3NS 100549 awarded by the National Institutes of Health. The government has certain rights in the invention.CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 670,049, filed on July 11, 2024. The entirety of the aforementioned application is incorporated herein by reference.BACKGROUND

[0003] Conventional approaches to diagnosing and monitoring outcomes for psychiatric disorders have numerous limitations. Numerous embodiments of the present disclosure aim to address the aforementioned limitations.SUMMARY

[0004] In some embodiments, the present disclosure pertains to a method of assessing a psychiatric disorder in a subject. In some embodiments, such methods include: (1) measuring a neural activity of the subject’s brain; and (2) classifying the psychiatric disorder based on the measured neural activity. In some embodiments, the methods of the present disclosure also include a step of implementing a treatment based on the classification of the psychiatric disorder. For instance, in some embodiments where the psychiatric disorder is classified as treatment-sensitive and / or moderately symptomatic, the treatment may include the administration of a therapeutic agent to the subject. In other embodiments where the psychiatric disorder is classified as treatment-resistant and / or severely symptomatic, the treatment may include neuromodulation.

[0005] Additional embodiments of the present disclosure pertain to systems for assessing a psychiatric disorder in a subject. In some embodiments, the system includes one or more computer readable storage mediums having a program code embodied therewith. In some embodiments, the program code includes: (1) programming instructions for measuring a neural activity of the subject’s brain; and (2) programming instructions for classifying the psychiatric disorder based on the measured neural activity. In some embodiments, the systems of the present disclosure also include programming instructions for recommending or implementing a treatment based on the classification of the psychiatric disorder.BRIEF DESCRIPTION OF THE DRAWINGS

[0006] FIG. 1A illustrates a method for assessing a psychiatric disorder in a subject.

[0007] FIG. IB illustrates a system for assessing a psychiatric disorder in a subject

[0008] FIGS. 2A-2B show ventral striatum (VS) recordings in patients with obsessive compulsive disorder (OCD), which show narrow band ~9 Hz power feature. FIG. 2A shows a reconstruction of deep brain stimulation (DBS) lead placement in ventral capsule (VC) / VS (n - 12) in standard atlas space. FIG. 2B shows power spectral density plots for each patient. In some cases, Applicant was only able to acquire recordings from one hemisphere due to stimulation configuration. Insets zoom in on 9 Hz, marked by the vertical dotted line. Gray shaded region shows the frequency band that we configured for chronic recordings (9 ± 2.5 Hz).

[0009] FIGS. 3A-3F show that temporal dynamics of VS neural activity reflect behavioral phenotype and clinical status. FIGS. 3A-3D provide illustrations depicting prototypical clinical / behavioral states associated with OCD and treatment with DBS. FIG. 3A provides avoidant rituals of the severely symptomatic state. FIG. 3B provides adaptive behaviors of clinical response. FIG. 3C shows excessively approachful behaviors of overstimulation, including disinhibition and decreased need for sleep. FIG. 3D shows balanced activities and sleep / wake patterns characteristic of clinical response. FIG. 3E shows left, heatmaps of left hemisphere VS 9-Hz power versus time of day (y axis) and days since VC / VS DBS activation (x axis) in the three clinical responders from Cohort 1. Bars over heatmaps indicate behavioral / clinical state. Right, polar plots with cosinor fit amplitude versus acrophase over time. The circadian pattern amplitude clearly distinguishes among the symptomatic state (strong circadian pattern), the overly disinhibited state (abolished circadian pattern) and clinically stable response (reduced circadian pattern). FIG. 3F is the same as FIG. 3E but for non-responders. There is less separation between the pre-DBS and long-term non-response states.

[0010] FIGS. 4A-4G show that VS neural activity is highly circadian and predictable in the severe OCD symptom state. All figure panels were constructed using data from one patient as an example non-rcspondcr, B006. FIG. 4A shows circularized plots of daily neural activity patterns showing z- scored 9-Hz power (radial axis) versus time of day (angular axis). The first two plots show representative days from the pre-DBS period, and the second two plots show representative days from the post-DBS period. The data within each day were rotated to align the peak daily amplitude to 3JC / 2 to allow averaging across days. FIG. 4B shows a circularized plot showing the average daily neural activity pattern over the entire pre-DBS period and the post-DBS period. FIG. 4C shows z-scored 9- Hz power over days throughout the entire monitoring period. FIG. 4D shows callouts of normalized 9-Hz power and model fits showing 3-d periods before DBS (left) and after DBS (right). Applicant fit a cosinor model and linear autoregressive model to the raw data. The linear autoregressive model was better able to capture intra-day fluctuations in 9-Hz power than was the cosinor model. The nonlinear model fit is not depicted as it was visually indistinguishable from the linear model fit. Daily cosinor R2(FIG. 4E), linear autoregressive R2(FIG. 4F) and sample entropy values (gray) (FIG. 4G) are plotted over time and overlaid by a 5-d exponential moving average (EMA) line (left). The color of the EMA line over time reflects clinical status as described in FIG. 4B. The half-violin plots (right) indicate the distribution of daily values within each clinical state. The daily values are not significantly different before versus after DBS activation in any of the metrics, via a two-tailed Welch’s / -test without multiple comparisons. Respective P values are 0.066, 0.510 and 0.772. AR, autoregressive; NS, not significant.

[0011] FIGS. 5A-5G show that clinical response is marked by a significant decrease in the predictability of VS neural activity. All figure panels were constructed using data from one patient as an example responder, BOOl. FIG. 5A provides the same scheme as in FIG. 5A, showing two representative days from the pre-DBS period and two from the post-DBS responder period. FIG. 5B shows a circularized plot showing the average daily neural activity pattern over the entire pre-DBS symptomatic period and the post-DBS responder period. FIGS. 5C-5G show the same scheme as in FIGS. 4C-4G. As opposed to the close fits in the non-responder data in FIGS. 4A-4G, the statistical metrics do not fit the post-DBS period well in this clinical responder. Cosinor amplitude (FIG. 5D) and R2(FIG. 5E) are reduced in the post-DBS versus pre-DBS data. The linear autoregressive model (FIG. 5D) does not match the raw data (FIG. 5D) as well as it did in FIGS. 4A-4G, and the model R2(FIG. 5F) is also reduced in the post-DBS versus pre-DBS data. Consistently, sample entropy increases over time. All three metric outputs (half- violin plots in FIGS. 5E-5G) are significantly different between the pre-DBS and post-DBS periods (asterisks), via a two-tailed Welch’s / -test without multiple comparisons. Respective P values are 1.02 x 10-16, 2.46 x IO-40and 4.10 x 10-14. AR, autoregressive.

[0012] FIGS. 6A-6D show that autoregressive model output metrics accurately distinguish responder status. FIG. 6A is a clothesline plot showing per-patient delta (that is, normalized to pre-DBS) values for each of the four output measures for the seven patients with pre-DBS and post-DBS data, such that a mean difference could be calculated and five-fold cross-validated (from left to right: cosinor A / ?2, linear autoregressive AT?2, nonlinear autoregressive A / ?2and A Sample entropy). Triangles represent responders, and circles represent non-responders. Separation between responders and non-responders was achieved for linear and nonlinear autoregressive A / ?2values and A Sample entropy using a maximum margin classifier (horizontal dashed black line). FIG. 6B shows ROC curves demonstrating classifier performance (symptom burdened versus symptom unburdened) of a logistic regression model trained and tested on delta values for each of the four output measures (cosinor AT?2; linear autoregressive AT?2; nonlinear autoregressive AT?2; and A Sample entropy). FIG. 6C shows analogous clothesline plots to FIG. 6A but including data from all 12 patients, including those with only segments of pre-DBS or post-DBS recordings. U002 does not appear in linear and nonlinear autoregressive T?2estimations due to missing data preventing calculation of the daily output measures. Pre-DBS and post-DBS non-responder points represent symptom burdened states, and post-DBS responder points represent symptom unburdened states. Only the linear autoregressive model narrowly achieved separation between symptom burdened and unburdened states. FIG. 6D shows analogous ROC curves to FIG. 6B but including mean estimates of daily output measures rather than delta values. AR, autoregressive.

[0013] FIGS. 7A-7G show the methods used in an alternative simplified autoregressive model to calculate the R2feature. All figure panels were constructed using data from one patient as an example non-rcspondcr, BOOl. FIG. 7A shows z-scorcd 9-Hz LFP power over days throughout the entire monitoring period. A sliding window with a width of three days and a stride of one day is applied to this data, with two example windows highlighted, one prior to DBS activation, and one after clinical response was achieved. FIG. 7B shows the three days of data included in the pre-DBS window, with the third day being the day of interest. The first two days of data in the window were used to fit a single-lag autoregressive (AR(1)) model. The learnable parameter of the model was locked and used to predict neural activity on the third day, with the predictions overlaid over the ground truth LFP power. FIG. 7C shows the calculation of the R2feature for the day of interest. From the third day of the window, the AR(1) model’ s predictions are scattered against ground truth normalized LFP power, a perfect fit line is displayed for reference, and the R2metric is calculated to represent correlation between the two sets. FIGS. 7D-7E use the same schemes as FIGS. 7B and 7C, respectively, but for a three-day window after DBS. FIG. 7F illustrates the predictive mechanics of the AR(1) model, where for each time point, a prediction is made based on the product of the previous term in the series and a learned coefficient. FIG. 7G shows the result of sliding the window over the entire time series and calculating the R2feature for each day. Arrows connect the R2values calculated on the example days in FIGS. 7C and 7E to their corresponding locations on the time series.DETAILED DESCRIPTION

[0014] It is to be understood that both the foregoing general description and the following detailed description arc illustrative and explanatory, and arc not restrictive of the subject matter, as claimed. In this application, the use of the singular includes the plural, the word “a” or “an” means “at least one”, and the use of “or” means “and / or”, unless specifically stated otherwise. Furthermore, the use of the term “including”, as well as other forms, such as “includes” and “included”, is not limiting. Also, terms such as “element” or “component” encompass both elements or components comprising one unit and elements or components that include more than one unit unless specifically stated otherwise.

[0015] The section headings used herein are for organizational purposes and are not to be construed as limiting the subject matter described. All documents, or portions of documents, cited in this application, including, but not limited to, patents, patent applications, articles, books, and treatises, arc hereby expressly incorporated herein by reference in their entirety for any purpose. In the event that one or more of the incorporated literature and similar materials defines a term in a manner that contradicts the definition of that term in this application, this application controls.

[0016] Conventional approaches to diagnosing and monitoring outcomes for psychiatric disorders have relied on categorizing symptom phenomenology, a scheme that has provided standardization to the field of psychiatry and facilitated communication among its practitioners over the past half century. High on the list of modern era criticisms, however, is its inherently phenomenological nature, focused on statistical patterns of symptom clusters rather than on underlying cause and mechanism.

[0017] The past decade has evidenced an accelerating interest in developing a mechanistically oriented framework within which to study psychiatric disorders and assess their treatments. However, most widely adopted of these is the National Institute of Mental Health’s Research Domain Criteria (RDoC), a trans-diagnostic framework defined by fundamental neurobehavioral constructs rather than symptom clusters.

[0018] In sum, current approaches to diagnosing and monitoring outcomes for psychiatric disorders have numerous limitations. Numerous embodiments of the present disclosure aim to address the aforementioned limitations.

[0019] In some embodiments, the present disclosure pertains to a method of assessing a psychiatric disorder in a subject. In some embodiments illustrated in FIG. 1A, such methods include: measuring neural activity of the subject’s brain (step 10); and classifying the psychiatric disorder based on the measured neural activity (step 12). In some embodiments, the methods of the present disclosure also include a step of implementing a treatment based on the classification of the psychiatric disorder (step 14). For instance, in some embodiments where the psychiatric disorder is classified as treatmentsensitive and / or moderately symptomatic, the treatment may include the administration of a therapeutic agent to the subject (step 16). In other embodiments where the psychiatric disorder is classified as treatment-resistant and / or severely symptomatic, the treatment may include neuromodulation (step 18).

[0020] Additional embodiments of the present disclosure pertain to systems for assessing psychiatric disorder in a subject. In some embodiments, the system includes one or more computer readable storage mediums having a program code embodied therewith. In some embodiments, the program code includes: (1) programming instructions for measuring a neural activity of the subject’s brain; and (2) programming instructions for classifying the psychiatric disorder based on the measured neural activity. In some embodiments, the systems of the present disclosure also include programming instructions for recommending or implementing a treatment based on the classification of the psychiatric disorder.

[0021] As set forth in more detail herein, the methods and systems of the present disclosure can have numerous embodiments.

[0022] Measuring neural activity

[0023] The methods and systems of the present disclosure may measure neural activity in various manners. For instance, in some embodiments, the measurement of neural activity includes measuring changes in the periodicity of neural signals relative to average periodicity. In some embodiments, the measurement of neural activity occurs continuously. For instance, in some embodiments, the measurement of neural activity occurs continuously at sequential intervals that are separated from one another from about 1 second to about 10 minutes. In some embodiments, the measurement of neural activity occurs continuously from the ventral striatum.

[0024] In some embodiments, the measured neural activity is derived from the subject’s prefrontal cortical area, deep gray area, ventral striatum (VS), ventral capsule (VC), or combinations thereof. In some embodiments, the measured neural activity is derived from the subject’s ventral striatum (VS). In some embodiments, the measured neural activity is derived from the subject’s ventral striatum (VS) and ventral capsule (VC).

[0025] In some embodiments, the measurement of neural activity includes measuring neural activity at or near the theta-alpha border. In some embodiments, the theta-alpha border is at or near 9 Hz. In some embodiments, the measurement of neural activity includes measuring neural activity of the ventral capsule (VC) or ventral striatum (VS) at or near the theta-alpha border. In some embodiments, the measurement of neural activity includes measuring neural activity of the ventral capsule (VC) at or near the theta-alpha border. In some embodiments, the measurement of neural activity includes measuring neural activity of the ventral striatum (VS) at or near the theta-alpha border.

[0026] Electrode implantation

[0027] In some embodiments, the neural activity is measured from one or more electrodes implanted into a subject’s brain. For instance, in some embodiments, the electrodes may be implanted into the ventral striatum (VS) of the subject’s brain. In some embodiments, the electrodes may be implanted into the ventral striatum (VS) and ventral capsule (VC) of the subject’s brain.

[0028] In some embodiments, the methods of the present disclosure also include a step of implanting one or more electrodes into a subject’s brain. In some embodiments, the systems of the present disclosure may be associated with one or more electrodes. In some embodiments, the electrodes include, without limitation, pulse generating and sensing leads, deep brain stimulation (DBS) leads, or combinations thereof.

[0029] Psychiatric disorders

[0030] The methods and systems of the present disclosure may be utilized to assess various psychiatric disorders. For instance, in some embodiments, the psychiatric disorders include, without limitation, obsessive-compulsive disorder, depression, bipolar disorder, post-traumatic stress disorder, substance use disorders, or combinations thereof. In some embodiments, the psychiatric disorder includes obsessive-compulsive disorder. In some embodiments, the psychiatric disorder includes bipolar disorder.

[0031] Classification of the psychiatric disorder

[0032] The methods and systems of the present disclosure may classify psychiatric disorders in subjects based on measured neural activities in various manners. For instance, in some embodiments, the classification of the psychiatric disorder is based on the classification of clinical response after a treatment, such as a deep brain stimulation. In some embodiments, classification of clinical response after treatment (e.g., deep brain stimulation) includes classification of obsessive-compulsive disorder, depression, bipolar disorder, post-traumatic stress disorder, or other psychiatric disorders. In some embodiments, the classification of the psychiatric disorder includes classification of the psychiatric disorder as a treatment-resistant psychiatric disorder, a severely symptomatic psychiatric disorder, a treatment- sensitive psychiatric disorder, a moderately symptomatic psychiatric disorder, or combinations thereof.

[0033] In some embodiments, the classification of the psychiatric disorder includes classification of the psychiatric disorder as a treatment-resistant psychiatric disorder. In some embodiments, a treatment-resistant psychiatric disorder is characterized by non-rcsponsivcncss of the subject’s psychiatric disorder symptoms to drug treatment.

[0034] In some embodiments, the classification of the psychiatric disorder includes classification of the psychiatric disorder as a severely symptomatic psychiatric disorder. In some embodiments, the severely symptomatic psychiatric disorder is characterized by disinhibited behavior. In some embodiments, the disinhibited behavior includes at least one of recklessness, relatively increased libido, or combinations thereof.

[0035] In some embodiments, the classification of the psychiatric disorder includes classification of the psychiatric disorder as a treatment- sensitive psychiatric disorder. In some embodiments, treatment- sensitive psychiatric disorder is characterized by the responsiveness of the subject’s psychiatric disorder symptoms to drug treatment.

[0036] In some embodiments, the classification of the psychiatric disorder includes classification of the psychiatric disorder as a moderately symptomatic psychiatric disorder. In some embodiments, the moderately symptomatic psychiatric disorder is characterized by a lack of disinhibited behavior, such as recklessness, relatively increased libido, or combinations thereof.

[0037] Treatment

[0038] In some embodiments, the methods of the present disclosure also include a step of implementing a treatment based on the classification of the psychiatric disorder. In some embodiments, the systems of the present disclosure also include programming instructions for recommending or implementing a treatment based on the classification of the psychiatric disorder.

[0039] For instance, in some embodiments where the classification of the psychiatric disorder includes a treatment- sensitive psychiatric disorder and / or moderately symptomatic psychiatric disorder, the treatment includes administering a therapeutic agent to the subject. In some embodiments, the therapeutic agent includes selective serotonin reuptake inhibitors (SSRIs), clomipramine, or combinations thereof.

[0040] In some embodiments where the classification of the psychiatric disorder includes a treatmentresistant psychiatric disorder and / or severely symptomatic psychiatric disorder, the treatment includes ncuromodulation. In some embodiments, the ncuromodulation includes deep brain stimulation (DBS) therapy.

[0041] Algorithms

[0042] In some embodiments, the methods of the present disclosure occur through the utilization of an algorithm. For instance, in some embodiments, the methods of the present disclosure include: feeding measured neural activity of a subject’s brain into an algorithm; and utilizing the algorithm to classify the psychiatric disorder based on the measured neural activity. In some embodiments, the algorithm is operable to implement programming instructions of the systems of the present disclosure. For instance, in some embodiments, the algorithm receives the measured neural activity of a subject’s brain and classifies the psychiatric disorder based on the measured neural activity. In some embodiments, the systems of the present disclosure include the algorithm.

[0043] The methods and systems of the present disclosure can utilize various algorithms. For instance, in some embodiments, the algorithm includes a machine-learning algorithm trained on measured neural activities of a subject. In some embodiments, the machine-learning algorithm includes a logistic regression model.

[0044] In some embodiments, the machine-learning algorithm includes an autoregressive model. In some embodiments, for each time point of neural activity, the autoregressive model makes a prediction on the value of the next time point based on the value of a previous time point and a learned coefficient. In some embodiments, the autoregressive model is represented by the following formula:

[0045] In some embodiments, Xt represents measured neural activity at a certain time point, strepresents an error term, andrepresents the learned coefficient of the model.

[0046] Subjects

[0047] The methods and systems of the present disclosure may be utilized to assess psychiatric disorders in various subjects. For instance, in some embodiments, the subject is a human being suffering from a psychiatric disorder. In some embodiments, the subject is a human being suspected of having a psychiatric disorder.

[0048] Computer storage media

[0049] The systems of the present disclosure can include various types of computer-readable storage mediums. For instance, in some embodiments, the computer-readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. In some embodiments, the computer-readable storage medium may include, without limitation, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or combinations thereof.

[0050] A non-exhaustive list of more specific examples of suitable computer-readable storage mediums includes, without limitation, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device, or combinations thereof.

[0051] A computer-readable storage medium, as used herein, is not to be construed as being transitory signals per se. Such transitory signals may be represented by radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0052] In some embodiments, computer-readable program instructions for computing devices can be downloaded to respective computing / processing devices from a computer- readable storage medium or to an external computer or external storage device via a network, such as the Internet, a local area network (LAN), a wide area network (WAN) and / or a wireless network. In some embodiments, the network may include copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. In some embodiments, a network adapter card or network interface in each computing / processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium within the respective computing / processing device.

[0053] In some embodiments, computer-readable program instructions for carrying out operations of the present disclosure may be assembler instructions, instruction- set-architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object-oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the "C" programming language or similar programming languages.

[0054] In some embodiments, the computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected in some embodiments to the user's computer through any type of network, including a LAN or a WAN, or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field- programmable gate arrays (FPGA), or programmable logic arrays (PL A) may execute the computer- readable program instructions by utilizing state information of the computer-readable program instructions to personalize the electronic circuitry in order to perform aspects of the present disclosure.

[0055] Embodiments of the present disclosure for assessing a psychiatric disorder as discussed herein may be implemented using a computing device illustrated in FIG. IB. Referring now to FIG. IB, FIG. IB illustrates an embodiment of the present disclosure of the hardware configuration of a computing device 30 which is representative of a hardware environment for practicing various embodiments of the present disclosure.

[0056] Computing device 30 has a processor 31 connected to various other components by computing device bus 32. An operating system 33 runs on processor 31 and provides control and coordinates the functions of the various components of FIG. IB. An application 34 in accordance with the principles of the present disclosure runs in conjunction with operating system 33 and provides calls to operating system 33, where the calls implement the various functions or services to be performed by application 34. Application 34 may include, for example, a program for assessing a psychiatric disorder as discussed in the present disclosure.

[0057] Referring again to FIG. IB, read-only memory ("ROM") 35 is connected to computing device bus 32 and includes a basic input / output computing device ("BIOS") that controls certain basic functions of computing device 30. Random access memory ("RAM") 36 and disk adapter 37 are also connected to computing device bus 32. It should be noted that software components including operating system 33 and application 34 may be loaded into RAM 36, which may be computing device’s 30 main memory for execution. Disk adapter 37 may be an integrated drive electronics ("IDE") adapter that communicates with a disk unit 38 (e.g., a disk drive).

[0058] Computing device 30 may further include a communications adapter 39 connected to computing device bus 32. Communications adapter 39 interconnects computing device bus 32 with an outside network (e.g., wide area network) to communicate with other devices.

[0059] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and systems according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams and combinations of blocks in the flowchart illustrations and / or block diagrams can be implemented by computer-readable program instructions.

[0060] These computer- readable program instructions may be provided to a processor of a computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer- readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer-readable storage medium having instructions stored therein includes an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0061] The computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer- implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0062] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of computing devices, methods, and computing devices according to various embodiments of the present disclosure. In this regard, each block in the flowchart orblock diagrams may represent a module, segment, or portion of instructions, which includes one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be accomplished as one step, executed concurrently, substantially concurrently, in a partially or wholly temporally overlapping manner, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based computing devices that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

[0063] Example 1. Disruption of neural periodicity predicts clinical response after deep brain stimulation for obsessive-compulsive disorder

[0064] Recent advances in surgical neuromodulation have enabled chronic and continuous intracranial monitoring during everyday life. Applicant used this opportunity to identify neural predictors of clinical state in 12 individuals with treatment-resistant obsessive-compulsive disorder (OCD) receiving deep brain stimulation (DBS) therapy.

[0065] Applicant developed neurobehavioral models based on continuous neural recordings in the region of the ventral striatum in an initial cohort of five patients and tested and validated them in a held-out cohort of seven additional patients. Before DBS activation, in the most symptomatic state, theta / alpha (9 Hz) power evidenced a prominent circadian pattern and a high degree of predictability.

[0066] In patients with persistent symptoms (non-responders), predictability of the neural data remained consistently high. On the other hand, in patients who improved symptomatically (responders), predictability of the neural data was significantly diminished. This neural feature accurately classified clinical status even in patients with limited duration recordings, indicating generalizability that could facilitate therapeutic decision-making.

[0067] In this Example, Applicant focused on a pair of related and well-studied constructs within this framework — response selection and inhibitory control — which lie within the broader category of cognitive control. In the context of decision-making, these terms refer to the ability to identify environmental or internal factors relevant to a pending decision, ignore irrelevant factors and arrive at an optimal decision. The terms also refer to the ability to suppress pre -potent, automatic responses in favor of reasoned ones that may be more adaptive in the longer term. These factors, therefore, reflect a latent axis, or continuum, of phenotypes.

[0068] As with most axes defined in such behavioral terms, an adaptive regime typically resides in the middle ground, a balance between the extremes, with the capacity to adjust the dial according to situational demands. Extrema along this axis, however, constitute well-characterized mental illnesses. On one end lie unbridled ‘approach’ behaviors and their attendant disorders of impulsiveness. On the other lie ‘avoidant’ phenotypes, marked by disproportional fear and anxiety.

[0069] Obsessive-compulsive disorder (OCD) is a classic example of the latter extreme. A dysfunctional response selection has been highlighted as a central feature in the pathophysiology of OCD.

[0070] OCD is common and debilitating, with a prevalence of 2-3% in the general population. In the most severe cases, patients spend an extraordinary amount time performing repetitive, seemingly senseless compulsions and pcrscvcrating on distressful, intrusive thoughts. Despite having insight that the thoughts are irrational and the compulsions unproductive, these individuals feel an overwhelming urge to continue these behaviors.

[0071] Unlike disorders of addiction in which behavior is reward seeking, OCD-related compulsions are difficult-to-control avoidant responses to internal thoughts or external triggers that develop to evade distress or harm. The avoidant diathesis is more obvious in certain subtypes of OCD, such as the contamination subtype (that is, avoiding germs by washing hands), but it also characterizes others. For example, individuals with taboo thoughts avoid saying or thinking certain things lest doing so causes harm to someone. Those with the ‘just right’ subtype are compelled to repeat actions until performed ‘correctly’ to avoid a negative consequence of performing them ‘incorrectly’.

[0072] Although pathological avoidance is not the only form of cognitive dysfunction in OCD, a substantial literature supports the idea that OCD symptomatology projects appreciably onto the axis of pathological avoidance and motivates Applicant’s focus on this behavioral dimension.

[0073] Greater degrees of avoidance are associated with greater OCD severity and greater treatment resistance. Exposure and response prevention (ERP) therapy is an effective, evidence-based treatment built on the premise that obsessive-compulsive symptoms are a manifestation of an avoidant phenotype and that overcoming that avoidance (facilitating approach) can improve symptoms. During ERP, a therapist guides the patient through interactions with OCD triggers (that is, exposures) that elicit distress while encouraging the patient to not engage in their typical attendant compulsions. Over time, engaging in the systematic approach toward and tolerance of distress-evoking triggers facilitates formation of new non- threatening associations, leading to symptom improvement.

[0074] For the 10-20% of individuals with OCD who are treatment resistant, deep brain stimulation (DBS) of the ventral striatum (VS) and adjacent capsular white matter (ventral capsule (VC)) is an effective therapy, with response rates in the 66% range and regulatory approval in the form of a humanitarian device exemption from the US Food & Drug Administration. Whereas improvement in OCD symptoms per se can take months to manifest, initiation of stimulation can have immediate and profound effects on mood and energy. These acute effects of VC / VS DBS activation have been labeled ‘mirth’ or ‘positive affect’ responses, variably consisting of smiling (often a contralateral hemi-smile), laughter or feelings of increased energy.

[0075] The early appearance of these behaviors is a possible predictor of eventual clinical response, although notable counter-examples exist. Based on Applicant’s experience with this cohort and others, Applicant adopted a more general terminology of ‘approach’ behavior to characterize this response, as one of the most common observable changes is an increase in talkativeness, a desire to engage in activities and extroversion. Indeed, a common side effect of overstimulation is the induction of overly approachful, disinhibited (often termed hypomanic) behavior, characterized by impulsivity, overabundance of energy, increased libido and decreased need for sleep.

[0076] Thus, Applicant focused on the approach-avoidance axis as one of the critical neurobehavioral axes underlying the pathophysiology of OCD. Although certainly not the only pathological axis in OCD, it does capture important behavioral manifestations of the disorder. To better understand its neurophysiological basis, Applicant sought to identify neural signatures of transitions along this behavioral axis produced by VC / VS DBS. Doing so would further anchor key diagnostic features within this cognitive framework, help advance its mechanistic understanding and, ideally, influence monitoring and treatment strategics.

[0077] To accomplish this goal, Applicant took advantage of a novel opportunity for continuous longterm intracranial recordings in patients with OCD. Recently available DBS systems have the ability to not only deliver stimulation but also record local field potentials (LFPs). Low-frequency oscillations in the theta (4-8 Hz) to alpha (8-12 Hz) range play a prominent role in the large body of neuroscientific literature on cognitive control, from animal studies to human non-invasive and intracranial neurophysiology.

[0078] As opposed to using traditional episodic, task-based cognitive neurophysiology approaches, Applicant adopted the more ethologically relevant approach made available with on-device recordings to determine whether variations in continuously acquired neural data could provide an objective readout of clinical state transitions (that is, from the avoidant state that characterizes severe OCD symptoms to the more approachful, tolerant state that characterizes clinical response).

[0077] Applicant hypothesized that changes in the periodicity of neural signals may provide insight into pathological network activity and, therefore, clinical status. Abnormalities in daily (circadian) periodicity are a cardinal feature of mental health disorders in general and of OCD in particular. Several studies identified a relationship between OCD symptom severity and circadian pattern alteration, including tendencies for delayed sleep onset and / or insomnia.

[0078] Given the frequent sampling required to identify circadian neural variations, as well as the extended time period required to study the clinical trajectory of response to DBS (months), Applicant leveraged the long-term recording capability of these DBS systems to track months’ worth of low- frequency power changes in 12 individuals with severe, treatment-resistant OCD. Applicant’s results demonstrate that 9-Hz (theta / alpha border) VS neural activity is highly periodic in the symptomatic state. Decreased periodicity and predictability of this signal after DBS initiation characterized clinical response, whereas persistence of these features marked non-response.

[0079] Twelve patients with severe OCD who met surgical criteria for DBS participated in this Example. Applicant bilaterally implanted DBS leads in the region of the VC / VS (FIG. 2A). Applicant also implanted a Medtronic Percept PC pulse generator, which not only delivers stimulation but can also record neural activity. Of the 12 patients, eight had clinical follow-up for more than 6 months (mean follow-up duration: 22.75 ± 8.4 months). Of those eight, five (62.5%) were responders to DBS (Yale-Brown Obsessive-Compulsive Scale (Y-BOCS) reduction^ 35%).

[0080] The 12 patients span two temporally contiguous cohorts. Cohort 1 consists of the first five patients (B001-B006), from whom Applicant has the longest recordings, starting before DBS initiation and spanning months afterwards until they achieved stable chronic clinical status. Data from these patients were used to identify promising models linking neural activity to clinical state. Cohort 2 consists of the next seven patients (B007-U003), whose data were more limited for several reasons. Applicant performed predictive testing on this expanded cohort to identify generalizable neurobehavioral relationships across patients.

[0081] Example 1.1. VS low-frequency power shows circadian periodicity before DBS

[0082] In the first few patients in Cohort 1, Applicant noticed a narrow-band peak in VS spectral power at the thcta / alpha border (at or near 9 Hz) (FIG. 2B). Given the prominence of this neural feature and its role in cognitive neurophysiology, Applicant chose 9 Hz as the center frequency to track upon implantation of the DBS system. Thus, Applicant obtained chronic, passive recordings in this band at continuous 10-min intervals in all patients. The duration of recordings amounted to several months’ worth in each of the five Cohort 1 patients (316 ± 128 d total duration per patient; 25 ± 20 d before VS DBS activation and 292 ± 136 d after) and several weeks’ worth in each of the seven Cohort 2 patients (67 ± 46 d total duration per patient).

[0083] By collecting these continuous recordings, Applicant sought to identify neural biomarkers of clinical states relevant to this disorder. One important state is that of severe OCD symptom burden, which describes the initial state for all patients before DBS activation. In the context of Applicant’s approach-avoidance framework, this is a state marked by highly avoidant behaviors performed in response to exaggerated or irrational representations of threat (FIG. 3A).

[0084] At the other extreme of the behavioral spectrum is a less common clinical state that can occur as a side effect of VC / VS DBS. This state consists of an overly approachful phenotype marked by notably disinhibited behavior (for example, risk taking, increased libido, impulsivity and overabundance of energy) (FIG. 3C). Between these two extremes lies the state of clinical response, which represents an adaptive balance between these behavioral poles (FIGS. 3B and 3D). More than 48,000 h of recordings in these patients demonstrated a remarkable correspondence between neural activity and clinical status.

[0085] Because LFP recordings arc enabled immediately upon implant, but stimulation typically starts weeks later, Applicant was able to obtain neural recordings for the baseline symptomatic clinical state. As evident from the heatmaps showing 9-Hz spectral power in the five Cohort 1 patients in FIGS. 3E- 3F (left hemisphere) and Extended Data in FIGS. 2A-2B (right hemisphere), this symptomatic state before DBS activation demonstrated a characteristic cyclic pattern of 9-Hz power with an approximately daily period.

[0086] Applicant applied four metrics (three model-based and one model-free) to assess this periodicity and its change after DBS activation. Applicant did so in the Cohort 1 patients first, because of their more extensive data. Given the circadian-appearing neural activity pattern, Applicant began by applying a cosinor regression model, a common model used for measuring circadian periodicity in biological systems. Modeling demonstrated a significant circadian (24-h period) fit in both the left (P < 10-63; FIGS. 3E-3F) and the right (P < 10-44; FIGS. 2A-2B) hemispheres. This circadian periodicity of 9-Hz neural activity was consistent across patients before DBS activation (FIGS. 3E- 3F and 2A-2B).

[0087] DBS initiation elicited notable behavioral and neurophysiological changes. In four of the five Cohort 1 patients (B001, B002, B004 and B005), Applicant observed acute pro-approach effects upon DBS initiation. The pro-approach response was potent enough in two patients (B004 and B005) that they exhibited clinically meaningful and maladaptive disinhibited behavior in the days immediately after DBS activation (dark bars in FIGS. 3E-3F).

[0088] In these two patients, the neural recordings demonstrated changes mirroring these behavioral observations. There was a reduction in amplitude of the circadian neural pattern that is evident in the polar plots in FIGS. 3E-3F as proximity of the red points to the origin. DBS programming adjustments (that is, decreased amplitude) at the next clinic visit alleviated those behaviors. Because of the short duration of these disinhibited periods and dearth of reliable behavioral data, Applicant did not attempt to model these acute effects.

[0089] Example 1.2. Loss of VS neural predictability indicates clinical response

[0090] Applicant focused efforts on identifying neural predictors of long-term clinical state (that is, responder versus non-responder status). Applicant classified ‘clinical response’ as time periods when a patient demonstrated clinically meaningful improvement in OCD symptoms and periods of ‘persistent OCD symptoms’ based on analogous opposite criteria. The heatmaps of 9-Hz power demonstrated a marked reduction in circadian periodicity during periods of clinical response relative to the pre-DBS symptomatic state (FIGS. 3E-3F and FIGS. 2A-2B). Consistently, the amplitude of the cosinor fit was reduced during these periods (FIGS. 3E-3F).

[0091] To study this effect, Applicant created circularized measures of daily neural activity patterns (FIGS. 4A and 5A). These circularized plots demonstrate several notable features. The circadian periodicity is evident as a lobed appearance or eccentricity in the daily circular plots of the prc-DBS (FIG. 4A, left panels, and FIG. 5A, left panels; light yellow) and non-responder (FIG. 4A, right panels) states and even more so in the state- averaged circularized plots (light and dark yellow in FIGS. 4B and 5B).

[0092] In contrast, the circularized plots of responder status (FIG. 5A, right panels) show less eccentricity and circadian periodicity. This effect is also most evident in the state-averaged plot (FIG. 5B). To illustrate these effects, Applicant created time-lapse animations of daily circularized plots, which, in responders, demonstrate the transition from patterns primarily characterized by eccentricity (strong circadian component) to ones with a prominent stellate shape (intra-day dispersion of neural power). In contrast, in non-responders, the pre-DBS eccentric pattern does not appreciably change.

[0093] Applicant quantified the change in strength of circadian periodicity before versus after DBS by comparing the cosinor model goodness of fit (7?2) between the two states. Based on the observations above, Applicant expected a significant decrease in R2(indicating decrement in periodicity) in clinical responders compared to no significant change in R2(indicating persistent periodicity) in non- responders. After accounting for autocorrelation in the data arising from frequent neural sampling (every 10 min) but lack of frequent variation in clinical status (responder versus non-responder) labels, the cosinor model performed relatively poorly under these conditions, with results consistent with expectations in only two of five hemispheres in clinical responders and in three of four hemispheres in non-responders. Without sample size correction, the cosinor model performed better, with consistent results in five of five responders and in three of four non-rcspondcrs.

[0094] Moderate performance and the overly simple nature of the cosinor model motivated Applicant to test other models. Besides decreased circadian periodicity, another feature evident in the responder circularized plots is increased daily variability or dispersion of neural activity, visible as spikiness. The cosinor model assumes a sinusoidal shape and is not well suited to capture these additional elements of periodicity. Thus, Applicant used a family of autoregressive models, which do not assume a particular shape of periodic data but simply test whether previous datapoints are predictive of future datapoints.

[0095] The lineai- and nonlinear autoregressive models fit the pre-DBS and non-responder neural data much more closely than did the cosinor model (FIG. 4D). On the other hand, the latter model simply captured the general daily / circadian trend, the former models matched within-day variations in neural activity with greater fidelity.

[0096] The close fits of the autoregressive models (FIGS. 3A-3F, 4A-4G, and 5A-5G) indicate a high degree of predictability of 9-Hz VS neural activity in the pre-DBS symptomatic state. Both linear and nonlinear autoregressive models discriminated clinical response from non-response more consistently than did the cosinor model. Results for these models matched the expected pattern in five of five hemispheres in responders and in three of four hemispheres in non-responders, not only without correction for non-independence but even with the conservative sample size correction. In other words, clinical response was associated with decreased predictability of neural activity, whereas nonresponse was associated with persistent neural predictability. The sole inconsistency was the right hemisphere of patient B006, in whom the neural pattern looked like that of a responder despite the fact that, clinically, the patient was a non-responder.

[0097] Examples of hemispheric asymmetries between brain activity and clinical outcomes are common in OCD. Functional imaging studies identified lateralized changes in blood flow, metabolic activity and functional connectivity in response to pharmacological, cognitive-behavioral and even DBS therapy. Applicant’s data also demonstrate more reliable predictability of clinical status from left VS neural data than from the right, as further demonstrated below for individual patient analyses in Cohort 1 and in the across-patient cross-validated predictive analyses in the full cohort.

[0098] In addition to these model-based metrics, Applicant also used sample entropy, a model-free metric, to capture this change in periodicity and variability. Sample entropy quantifies the regularity and (un)predictability of time- series data, making it a good candidate for data with these features. Because it is a computed value rather than a model, Applicant compared its value rather than goodness of fit pre-DBS versus post-DBS.

[0099] A finding consistent with Applicant’ s expectation would be a significant increase in sample entropy in responders compared to no change in non-responders. Sample entropy performed better than the cosinor model but not as well as the autoregressive models. It behaved as expected in five of five responder hemispheres and in three of four non-responder hemispheres without sample size correction but in only four of five responder and in three of four non-responder hemispheres with the conservative sample size correction.

[0100] Example 1.3. VS predictability indicates response status in full cohort

[0101] To determine whether these metrics generalize to a larger sample, Applicant applied them to the seven additional Cohort 2 patients (FIGS. 3A-3F and 4A-4G) who had less data for a variety of reasons. The data show distributions of the four output measures (cosinor and autoregressive R2values and sample entropy) for the 10 remaining patients not shown in FIGS. 4A- 4G and 5A-5G. Three Cohort 2 patients — one responder (B007) and two non-responders (B008 and U001) — had at least one hemisphere’s worth of both pre-DBS and post-DBS data, but the remaining four Cohort 2 patients only had intervals of pre-DBS or post-DBS data.

[0102] Applicant used the combined sample of all 12 patients’ data across both cohorts to test the four metrics for generalizability using cross-validated predictive modeling. To do so, Applicant dichotomized the data into symptom burdened and unburdened states. The symptom burdened state included periods of time before DBS activation (when all patients are symptomatic; light lines in FIGS. 3A-3F, 4A-4G, and 5A-5G) and after DBS activation in clinical non-responders (persistent symptoms; dark lines in FIGS. 3A-3F, 4A-4G, and 5A-5G). The symptom unburdened state simply included post-DBS periods of clinical response (dark lines in FIGS. 3A-3F, 4A-4G, and 5A-5G).

[0103] Applicant began by comparing the distributions of each of the four output measures between the symptom burdened and unburdened states. As Applicant did for Cohort 1 above, Applicant did so both with and without sample size correction to account for non-independence arising from autocorrelation of the clinical state labels. Without correction, the autoregressive models and_ Art _ -I rr sample entropy could reliably distinguish clinical state in both left (P < 10 ) and right (P < 10 ) hemispheres, but the cosinor model could do so only in the left hemisphere (P < 10 ). With conservative correction, the autoregressive models and sample entropy were still successful in both left (P < 10-15) and right (P < 10-5) hemispheres, but the cosinor model was no longer successful in either.

[0104] Applicant first performed the analyses using data from the seven patients (all five from Cohort 1 and two from Cohort 2) who had both pre-DBS and post-DBS neural data, such that a delta value could be calculated (FIG. 6A; A / ?2for the autoregressive models and A Sample Entropy for the model-free measure). For both autoregressive models and sample entropy, a maximum margin classifier could identify a delta value that separated responders from non-responders. The cosinor model was unable to make this distinction.

[0105] Applicant then used daily delta values to train a logistic regression model with leave- one-patient-out cross-validation to predict responder status on any given day. Applicant found that linear and nonlinear autoregressive R2measures outperformed cosinor R~ and sample entropy measures (FIG. 6B) and that left hemisphere data yielded more accurate predictions than did right hemisphere data. Left hemisphere linear and nonlinear autoregressive features yielded a balanced accuracy of 82% and 84%, respectively, corresponding to an area under the receiver operating characteristic curve (AUROC) of 85% and 89%, respectively. Classifier performance was significantly above chance levels.

[0106] An even more demanding requirement would be for a parameter to reliably indicate clinical status based on partial data — that is, recordings lacking a pre-DBS versus post-DBS comparison. FIG. 6C includes data from all 12 patients, including those with only a segment of preDBS or post-DBS data. Even with these limited data, one of the metrics (the linear autoregressive model) has a maximum margin R2classifier value that separates symptom burdened from unburdened states. The margin between these distributions is very narrow and would likely not hold across a much larger dataset, but these results suggest that the autoregressive model R2value is a strong candidate predictor of clinical status.

[0107] As above, Applicant trained a classifier using these daily predictability measures without normalizing to the pre-DBS value. Again, the linear and nonlinear autoregressive R2measures outperformed cosinor R and sample entropy (FIG. 6D) and enabled accurate classification of clinical status with a balanced accuracy of 82% and 80%, respectively, corresponding to an AUROC of 88% and 87%. This performance was, again, well above chance. Thus, despite the lack of pre-DBS and post-DBS recordings in the expanded cohort of 12 patients, predictability measures from the autoregressive model fits were still able to reliably classify responder status.

[0108] Example 1.4. Discussion

[0109] Applicant’s results identify a neurophysiological biomarker in the VS that enables prediction of clinical state from neural data. The best-performing feature — daily predictability of low- frequency (~9 Hz) power — is most prominent in the severely symptomatic state. Chronically, retention versus loss of this predictability distinguishes clinical non-responders from responders, respectively. Applicant initially identified this neurobehavioral relationship in a cohort of five patients with plentiful data, and its robustness was maintained in an expanded cohort of seven additional patients with sparser data.

[0110] A consistent relationship between a readily trackable neural feature and clinical status would have important practical implications. Such a biomarker could help determine whether an individual is heading in the direction of clinical response. Programming DBS for OCD is challenging because of the inconsistent relationship between DBS parameter adjustments and symptomatic changes. Unlike the field of movement disorder DBS, in which adjustments are often associated with immediate changes in symptoms (for example, reduction in tremor or stiffness), reduction in OCD symptom severity per se often requires months to manifest. An early neural indicator of emerging treatment response would be extremely informative and could help guide therapy delivery, thus demystifying the process of DBS programming for OCD and making the therapy more accessible to a greater number of clinicians and patients.

[0111] Further contributing to the generalizability of Applicant’s findings is the straightforward nature of obtaining these neural recordings. The presence of the spectral feature that Applicant observed in the first few participants near the theta / alpha border was consistent with the relevance of this frequency range to the cognitive processes underlying OCD. The prominence and exact center frequency of this approximately 9-Hz feature varied across the 12 patients, but Applicant’s results demonstrated robustness even with all patients’ recordings fixed at 9 Hz.

[0112] Avoiding the need to individualize recording settings reduces complexity and increases the likelihood that novice centers adopt and make use of these findings. Data availability and utility are also facilitated by the passive nature of the recordings. Applicant’s previous work identified low- frequency neural correlates of OCD symptom severity, but the active data acquisition requirement demanded considerable effort from the research team as well as exceptional commitment from the patients. Reliable results obtained from passive data streams will further democratize their accessibility and usability.

[0113] The availability of chronic, continuous neural data provides new perspectives for neurobehavioral analyses. Traditionally, neural biomarkers of psychiatric symptom states have relied on snapshots of neural data collected during behavioral tasks, time-locked exposures to triggers or variation in symptom severity. Such measures may be appropriate in the context of adaptive or responsive neuromodulation development, which prioritizes relatively rapid adjustments of stimulation delivery based on episodic changes in symptom state.

[0114] Applicant’s findings, on the other hand, indicate that neural biomarkers of slowly evolving clinical states may not be episodic variations from baseline but, rather, features of variation in the baseline itself. These high- amplitude periodic variations must be taken into account when studying the paroxysmal changes, as the simple effect of time of day will otherwise obscure interpretability. Failure to do so may still produce statistically significant and meaningful relationships between clinical state and neural activity but would fail to fully capture the dynamics.

[0115] These results have a direct implication on efforts related to closed-loop neuromodulation therapy. DBS adjustments (and even medication administration) in Parkinson’s disease, for example, have rapid effects on motor symptoms. OCD, on the other hand, is an example of a disorder in which symptom changes usually take weeks to months after DBS initiation or adjustment to become evident. Consistently, the principal tool used to measure OCD symptom severity, the Y-BOCS, asks the patient to consider an approximately week-long period in their responses. Applying this concept to the design of closed-loop algorithms, the temporally tight neural symptom relationship in Parkinson’s disease has led to the use of very short time constants relating detection of a neural signal (for example, increased beta power in the subthalamic nucleus) and adjustment of DBS settings.

[0116] The slower neurobehavioral time constant in Applicant’s results, akin to the one for depression severity recently described by others, suggests that building closed-loop classifiers for psychiatric disorders may benefit from using slowly evolving neural signals (for example, daily measures of neural predictability) rather than episodic ones. An important question raised by these results is that of a mechanistic explanation: what neurophysiological processes underlie the circadian periodicity and neural predictability that Applicant observed that seem so closely related to clinical state? The literature suggests physiological and anatomical relationships that may be explanatory and could generate testable hypotheses for future animal or human research.

[0117] One possibility is that the 9-Hz circadian rhythm that Applicant observed is produced by pattern generators in the hypothalamus, as are many other such rhythms. Among the most common cell types related to circadian pattern generation arc the orexin neurons of the lateral hypothalamus, which regulate not only wakefulness and arousal but also limbic processes, including fear, stress response and anxiety. Their pathological activity may induce hypervigilant states that promote fearful / avoidant behavior characteristic of the OCD symptomatic state. From the anatomical point of view, the DBS target borders the VS but may, in addition, impinge on the gray matter of the bed nucleus stria terminalis (BNST), a small paired septal nucleus that, in animals, is involved in fear conditioning. Due to the direct connectivity between VS / BNST and the hypothalamus, DBS in this region may have direct physiological effects on the hypothalamus.

[0118] Combining these lines of evidence, it is possible that pathological activity of pattern generators, such as orexin neurons, produces the avoidant behaviors observed in OCD concomitantly with the strongly periodic neural activity. DBS of the VS / BNST region may interfere with this lateral hypothalamic dysfunctional activity, disrupting the neural pattern and reversing behavior toward a more approachful diathesis characterized by tolerance of previous triggers and, thus, decreased OCD symptom severity.

[0119] These hypothalamic effects may help explain the daily periodicity that Applicant observed in neural activity in the symptomatic state (for example, FIG. 4C). Atop this 24-h cycle was a higher-frequency neural activity pattern with a high degree of predictability before DBS that persisted after DBS in non-responders. This neural predictability is consistent with the predictability in behavior that characterizes the OCD symptomatic state. Individuals with OCD are often seized by rituals, whether overt (for example, washing, checking, touching, counting and repeating) or covert (for example, silently reciting prayers to neutralize unwanted thoughts). The need to perform these rituals defines these individuals’ response to a situation or trigger and, thereby, severely limits their behavioral repertoire, making it highly predictable when exposed to triggers. This rigid repertoire and perseverative behavior may stem from the cognitive inflexibility that has long been postulated to underlie the OC phenotype.

[0120] In contrast, healthy, adaptive behavior is characterized by the flexibility to respond to external or internal cues in the context of longer-term goals and plans. Indeed, cognitive flexibility is one of the hallmark features of intelligent behavior. The transition from repetitive, ritualistic actions in the symptomatic state to more adaptive and flexible behaviors in the responder state is accompanied by increased dispersion and decreased predictability of VS neural activity.

[0121] Applicant’s working hypothesis is that greater variability in neural activity in the VS, a structure known to influence motivated, goal-directed behavior, allows the agent to shed maladaptive ritualistic behavior in favor of reasoned behavior aligned with long-term goals.

[0122] Example 1.5. Study design

[0123] Twelve adults with a principal diagnosis of severe, treatment-resistant OCD underwent DBS implantation after informed consent. All patients had OCD for more than 5 years and had failed adequate trials of selective serotonin reuptake inhibitors (SSRIs), clomipramine and augmentation as well as expert ERP therapy. Nine patients (B001-B010) underwent DBS surgery and were followed at Baylor College of Medicine (BCM), and three patients underwent DBS surgery and were followed at the University of Utah (U001-U003).

[0124] The full group consisted of two patient cohorts. Cohort 1 consisted of the first five patients (B001, B003, B004, B005 and B006) from whom Applicant has the longest recordings, starting before DBS initiation and spanning months afterwards until they achieved stable chronic clinical status. Data from these patients were used to construct the neurobehavioral models. Cohort 2 consisted of the next seven patients (B007-U003) whose data are more limited for several reasons. One of these patients (B009) received the recording-enabled DBS generator at the time of a generator replacement procedure, such that recordings were only available well into their therapy. Three patients (B007, U001 and U002) underwent a programming change soon after DBS initiation to a recordingincompatible stimulation configuration, precluding recordings for a duration of their therapy. Three patients (B008, B010 and U003) were implanted relatively recently or have shorter clinical follow-up (less than 6 months) and, therefore, do not have recordings spanning the full trajectory from pre-DBS to stable chronic status.

[0125] Gender was determined based on self-report. Neither sex nor gender was considered in the study design, and sex / gender analyses were not carried out due to the small sample size and lack of pre-existing hypotheses on sex / gender differences.

[0126] Example 1.6. DBS surgery

[0127] DBS leads (model 3387 or SenSight) with 1.5-mm spacing were placed bilaterally in the VC / VS region. Lead locations were determined using direct targeting on the prc-opcrativc magnetic resonance imaging (MRI). Applicant refined the location of the leads using awake intraoperative testing as Applicant previously described. The pair of leads was connected to extensions that were tunneled to a Medtronic Percept PC pulse generator.

[0128] Example 1.7. Imaging protocol and DBS electrode localization

[0129] Applicant acquired pre-operative high-resolution Tl-weighted MRI sequences and post-operative computed tomography (CT) as previously described. Electrode reconstruction was performed using the advanced processing pipeline in Lead-DBS. Lead Group software was used to visualize all electrodes from single-implant and dual-implant patients using the DISTAL Minimal Atlas in 7T MRI ex vivo atlas (FIG. 2A).

[0130] Example 1.8. Determination of clinical states

[0131] The Y-BOCS and the Y-BOCS II were administered before surgery and subsequently periodically throughout the course of DBS treatment. Applicant supplemented these clinical indicators with reports obtained from reliable informants (for example, caregivers and family members) whose communications we considered trustworthy. This combination of data takes advantage of impressions from both clinicians and caregivers and reflects the real-world nature of this study. Applicant used these data to classify four clinical states. (1) Severe OCD symptoms. By definition, this was the state before DBS initiation. (2) Clinical response. Applicant classified clinical response as periods when a patient demonstrated clinically meaningful improvement in OCD symptoms, defined quantitatively as >35% improvement in Y-BOCS (conventional ‘responder status’) or qualitatively as significant improvement allowing successful return to professional / social life. (3) Persistent OCD symptoms. These were periods after DBS initiation during which OCD symptom severity remained near pre-DBS levels. (4) Disinhibited behavior. Applicant defined these as periods when the patient exhibited disinhibition (recklessness, increased libido and poor decision-making) to an extent that warranted DBS adjustment. Unlabeled periods are those during which clinical status is undetermined (that is, insufficient data) or intermediate (that is, improved but not reaching clinical response criteria).

[0132] Example 1.9. Time-domain neural recordings

[0133] Time-domain neural recordings were acquired directly from the Percept PC device at a sample rate of 250 Hz. During recordings with DBS on, Applicant used BrainSense Streaming mode to record LFPs in a bipolar configuration around the active stimulation contact. During recordings with DBS off, Applicant leveraged Indefinite Streaming mode to enable recordings from each of the three possible contact pairs (0-2, 1-3 and 0-3; assuming monopolar stimulation at contact 1 or 2) simultaneously.

[0134] Example 1.10. Power spectral density estimations

[0135] To analyze the oscillatory content of the time-domain LFP data, Applicant estimated power spectral density from data collected while the patient was at rest. First, Applicant computed power spectral density for the time-domain signal using Welch’ s method (MATLAB pwelch function) with a Hamming window of 1 s, a window overlap of 600 ms, and a 256-point fast Fourier transform (FFT). Applicant then converted to decibels by taking the base 10 logarithm of the result and multiplying by 10. The resulting power spectral density plots (power in decibels versus frequency) are shown in FIG. 2B.

[0136] Example 1.11, Data collection

[0137] The Percept PC can be configured to record a continuous estimate of average LFP amplitude from a user-defined frequency band of interest (center frequency ±2.5 Hz). These data are saved onboard the device every 10 min (144 samples per day) and can be offloaded upon connection to the Medtronic DBS Clinician Programmer Application. Applicant configured the device to record LFP signal amplitude (units of microvolts peak) in the 8.79 ± 2.5 Hz band (referred to in the main text and hereafter as 9 Hz).

[0138] Example 1,12, Data downloads

[0139] Chronic data were exported in JSON format during clinic visits. Gaps in data occurred for one of two reasons: (1) errors in recording configuration or (2) if more than 60 d have passed since the data were last downloaded. After 60 d, the device begins to overwrite already saved data due to onboard storage limitations.

[0140] Example 1,13. Data conditioning

[0141] Timestamps were converted from UTC to local time zone. Outliers were defined as values above 30 standard deviations over the global median of a continuous chunk of data with no losses and were then replaced with a piecewise cubic hcrmitc interpolating polynomial (pchip) method. Less than 0.1% of data met the criteria for outliers. Subsequently, segments of missing values with a length of, at most, 1 h (six samples) were interpolated. Otherwise, missing timepoints were stored as not a number (NaN). Interpolation was used for less than 0.17% of data, and the average interpolation length was 1.78 ± 1.28 samples.

[0142] Example 1,14, Normalization

[0143] To account for increases over time in baseline LFP amplitude and stimulation amplitude, daily power arrays were z-scored using daily means and standard deviations.

[0144] Example 1.15. Visualization

[0145] The raw amplitude values were discretized into one of 144 10-min bins based on timestamp and reshaped from a one-dimensional vector of chronological data to a two-dimensional matrix of date versus time of day for visualization. These data were then plotted as a heatmap (FIGS. 3A-3F) where greater power values are represented by lighter blue colors.

[0146] Example 1,16. Construction of circularized plots of daily 9-Hz power

[0147] To better visualize the 9-Hz power fluctuations over the course of the 24-h period, Applicant generated circularized polar plots of 9-Hz power on each day.

[0148] Example 1,17, Rotation and smoothing

[0149] To reduce the impact of slow phase shifts on data visualizations and averaging (see below), the polar plots were constructed such that data from each 24-h period were rotated to align peak activity to midday (12:00 local time). For each day in which an acrophasc was fit with a significant P value, the power values for that day were rotated circularly so that the acrophase pointed to 3?r / 2 radians (downward on the polar plot). A Gaussian smoothing blur kernel was applied to each day after rotation.

[0150] Example 1.18. Visualizations

[0151] Representative examples of the pre-DBS severe OCD symptom, clinical response, and persistent OCD symptom are shown in FIGS. 4A-4G and 5A-5G. Single-day circularized plots were visualized over time for patients B004 and B006.

[0152] Example 1.19. Overview of model-based and model-free measures

[0153] To analyze the periodicity and predictability of Applicant’ s data, Applicant applied a suite of model-based measures (cosinor R2and linear and nonlinear autoregressive R2) and one model- free measure (sample entropy). For each of these four output measures using data collected in the VS, Applicant computed an average representing the pre-DBS severe OCD symptom state, the post-DBS persistent OCD symptom state and the post-DBS clinical response state. In addition to the average measures computed per state, Applicant computed these measures on each day so that Applicant could visualize the trajectory over time and build distributions that are representative of each clinical state. The following sections describe the metrics and the respective analyses related to each in more detail.

[0154] Example 1.20, Cosinor model

[0155] Normalized LFP data were fitted to a sum of cosine functions using a linear regression model as set forth in Equation 1, where the fitted sinusoid with N components can be expressed as:

[0156] In Equation 1, y(t) represents the raw signal; eif) is an error term; M is the midline statistic of rhythm (MESOR); P is the period of the cosinor used to fit the data; A is amplitude; j is component; and t is time. A period of exactly P = 24 h was selected for all analyses. The number of sinusoidal components was selected per patient to reflect the maximum number of daily peaks. Cosine fits were generated using 5-d windows centered on the day of interest (for example, day of interest ±2 d). To determine strength of periodicity before DBS, Applicant also fit a single cosinor using the pre-DBS severe symptom state for each patient.

[0157] Example 1,21, Cosinor metrics

[0158] Cosine fits were expressed as acrophase (for example, time shift of daily cosine peak from 0:00 local time), amplitude and the coefficient of determination (R2). The resulting phase and amplitude values are plotted per patient in the polar plots shown in FIGS. 3A-3F and FIG. 2B.

[0159] Amplitudes and acrophases reported here display only the ‘primary’ (that is, largest) peak. The cosinor model was fitted to individual patient states (that is, pre-DBS, persistent OCD symptoms and clinical response) using a five-fold cross-validation strategy implemented across all 10- min timepoints. For each fold, Applicant used the model to generate a prediction of the -scored data at each 10-min timepoint and calculated an R2value representing the strength of the fit. These R2values were then averaged across all test folds within each clinical state to generate an average 7?2value representative of each clinical state, per patient. Applicant then aggregated the predicted data points from all five models to calculate an R2value representing the strength of the fit on each day. The cross validated R2values and daily2values were calculated using Equation 2, which can be expressed as:

[0160] In Equationis the true value; ytis the predicted value at sample i; and y is the mean of true values.

[0161] Example 1 ,22, Linear autoregressive model analysis and statistics

[0162] A lineai- autoregressive model is a statistical model used for predicting time-series data, where the current value in a series is modeled as a linear combination of previous values. A linear autoregressive model can be expressed in Equation 3 as:

[0163] In Equation 3, Xtis the value of the series at time t c is a constant; <f>k are p parameters of the model; and stis an error term.

[0164] Example 1.23. Determining significant lag terms

[0165] Lag terms, including up to 144 previous timepoints (from Xf-i to Xr-144), were used in the model to predict the value of the series at each time, t. To prevent overfitting, Applicant selected significant lag terms from the time-series data for each patient. The model initially considered one day’s worth of previous values, amounting to 144 timepoints, as potential regressors. To determine the robustness and relevance of these lag terms, a five-fold cross-validation technique was conducted with an ordinary least squares regression model. During each fold of the cross-validation, lag terms that showed statistical significance (P < 0.05) were identified and recorded.

[0166] A lag term was retained for further analysis if it was deemed significant in more than three of the cross-validation folds. This process ensured that only consistently relevant lag terms were considered and prevented overfitting of the model. Subsequently, all selected significant lag terms were tested against the entire dataset. In cases where any lag terms proved to be statistically nonsignificant when applied to the full dataset, the above selection process was iteratively repeated. The process continued until only statistically significant lag terms remained.

[0167] Example 1.24, Training and evaluation

[0168] After identifying a set of statistically significant lag terms for each patient’s entire dataset, Applicant applied a linear autoregressive model to individual patient states (that is, pre-DBS, persistent OCD symptoms and clinical response). To assess the model’s predictive accuracy within each state, Applicant employed a five-fold cross-validation strategy implemented across all 10-min timepoints. The key metric for evaluating the model’s performance was the coefficient of determination (7?2). Within each fold, Applicant used the model to generate predictions of the data at each 10-min timepoint. Similar to the cosinor model evaluation, Applicant aggregated datapoints across all five models to calculate an R2value representing the strength of the fit on each day. The results of Applicant’s analysis were robust to hetero skedastic conditions, as confirmed by implementing a hetero skedastic autocorrelation robust covariance in Applicant’s regression, which yielded predictions consistent with those from a standard regression.

[0169] Example 1.25. Nonlinear autoregressive model analysis and statistics

[0170] A nonlinear autoregressive model extends the scope of a linear autoregressive model by capturing complex, nonlinear relationships within the data. Applicant modeled the relationship between past data points and current values using a neural network due to the expressive nature of these models for nonlinear relationships.

[0171] Example 1.26. Neural network architecture

[0172] Applicant designed a simple neural network architecture using PyTorch 2.1 ,0+cul 18. The network is composed of two layers. The initial layer of the network is a fully connected linear layer that transforms the input, consisting of 144 lagged features of the previous day, into a hidden layer comprising 32 neurons. A nonlinear activation function, rectified linear unit (ReLU), is applied after the first layer to introduce nonlinearity into the model. The second layer then maps the 32-neuron hidden layer to the single-neuron output layer that predicts the time series.

[0173] Example 1,27, Training and evaluation

[0174] Similar to the linear autoregressive model, Applicant applied the nonlinear autoregressive model to individual patient states (that is, pre-DBS, persistent OCD symptoms and clinical response). Within each state, a five-fold cross-validation strategy was implemented across all 10-min timepoints. The training for each cross-validation split was conducted over 50 epochs, using the Adam optimizer with a learning rate of 0.001 to optimize performance. To prevent overfitting and to enhance the model’s generalizability, LI regularization was applied with a lambdaThe mean absolute error served as the loss function. The model architecture, the loss function, the regularization scheme and the learning rate were all optimized through testing. Throughout the training phase, Applicant monitored the R2for both training and testing sets as a function of epochs to prevent overfitting. Applicant selected the epoch with the best test R2as the best model for each fold in the cross-validation strategy. Similar to the cosinor and linear autoregressive model evaluation strategies, Applicant aggregated data points from all five models to calculate an R2value measuring the goodness of the fit on each day.

[0175] Example 1.28. Sample entropy analysis and statistics

[0176] Sample entropy is a model-free measure used to quantify the regularity and (un)predictability of time-series data. Intuitively, this measure provides an estimate of the complexity in a time-series by quantifying the similarity of small windows within the time-series. Mathematically, sample entropy is defined and calculated in Equation 4 as follows:Sample entropy

[0177] In Equation 4, A is the number of unique pairs of distinct subsequences / windows ( ,, xf) of X of length m + 1, and B is the number of unique pairs of distinct subsequences / windows (xt, Xj) of X of length m. The Manhattan distance d(xi, Xj) between any two distinct subsequences / windows must be smaller than a user- specified tolerance, r, to be considered similar. The implementation of sample entropy was based on the EntropyHub version 0.2 package. To generate sample entropy values for each day, Applicant used single-day - scored 9-Hz power values as input and set the input parameters as m = 2, r = 3.6 and r= 1.

[0178] Example 1,29. Statistical calculations

[0179] To compare the mean difference between the pre-DBS state (light yellow) and postDBS clinical response state (blue) or the persistent OCD symptom state (dark yellow), Applicant computed a delta value representing the difference between the pre-DBS and post-DBS mean values for each of the four output measures described above (cosinor R2. linear- autoregressive R2, nonlinear autoregressive R2and sample entropy) (FIG. 6A).

[0180] Applicant quantified the separation between symptom burdened (for example, pre-DBS and persistent OCD symptoms) and symptom unburdened (for example, clinical response) states by using a maximum margin classifier to demonstrate the linear separability of average output measures across these two states. The maximum margin classifier is visualized as a horizontal dotted line in FIGS. 6A and 6C for the average deltas and average daily output measures, respectively.

[0181] Within each patient, the four daily output measures described above were compared between the pre-DBS and the post-DBS distributions using two-sample two-tailed Welch’s / -tests. Applicant reports the results of the / -test and modified / -test both before and after sample size correction, respectively.

[0182] To account for the autocorrelation present within each distribution, Applicant computed a measure of true information present in the data known as effective sample size (ESS). ESS quantifies the number of independent and identically distributed samples that would be equivalent to the correlated samples in terms of the amount of information that they contain about the population parameter. The ESS of highly autocorrelated data is much smaller than the actual sample size because correlated samples contain less unique information. Mathematically effective sample size is expressed in Equation 5 as:E f fective sample size1

[0183] In Equation 5, N is the original sample size, and pk is the autocorrelation coefficients at lag k up to a maximum lag in. In Applicant’s analysis, Applicant chose a maximum lag of 10 d. For distributions with fewer than 10 d of data, the maximum lag was the number of days present in the distribution. The effective sample size was then used to calculate a two-sample two-tailed Welch’s t- test and a Hedge’s g in place of a traditional sample size.

[0184] Lastly, Applicant then pooled the data across patients and computed two-sample two- tailed Welch’s Z-tests comparing the daily output measures associated with symptom burdened and symptom unburdened states. The effective sample size for each output measure was calculated as the sum of the effective sample size of each individual patient’s data included in the analysis.

[0185] Example 1 ,30. Classifier development

[0186] Applicant used a logistic regression model to predict clinical status (symptom burdened versus symptom unburdened) on any given day. Applicant trained the model a total of eight times on different sets of input features. The first four classifiers were trained using the ‘delta values’ corresponding to each of the four output measures, where the delta value corresponds to the difference between the average output measure for the pre-DBS state and the output measure on any given day. These features can be calculated only if both pre-DBS and post-DBS data are available within the same patient. The second set of four classifiers was trained without the pre-DBS and post-DBS comparison and, instead, used the raw output measures calculated on each day as feature inputs. Therefore, these features can be calculated for any patient regardless of missing data.

[0187] Applicant balanced the class weights of the logistic regression model proportionally to reflect the class imbalance in the data for each patient and performed leave-one -patient-out cross- validation to test the model. This process was repeated until data from all patients were tested. Applicant computed the ROC curve, the AUROC and the balanced accuracy for all the classifiers in accordance with Equation 6.Sensitivity 4- SpecificityBalanced accuracy ™ (6)

[0188] In Equation 6, TN, TP, FN and FP denote true negative, true positive, false negative and false positive rates, respectively.

[0189] Example 1,31. Chance performance estimation

[0190] Due to the imbalance of data between the two states (that is, non-responder versus responder), Applicant performed a permutation test to determine chance level performance of the classifier and whether classifier performance was significantly different than chance. Applicant assigned labels in two different ways: (1) randomly and (2) via a circular shift.

[0191] For the first permutation test, the labels between the two states were randomly shuffled. For the second permutation test, the labels for each patient were circularly shifted by a random amount. Applicant implemented this shuffling strategy to preserve the temporally contiguous nature of the data labels (that is, response / non-response labels do not occur randomly but, rather, in large contiguous blocks). A random shuffling procedure would reduce the amount of autocorrelation between the train / test splits and, thereby, inflate the difference between the true value and shuffle distribution.

[0192] For both methods, Applicant then used a leave-one -patient-out cross-validation strategy (as previously described) to evaluate the model’ s performance on the shuffled data. As before, Applicant computed the AUROC score and balanced accuracy to quantify performance. This procedure was repeated 10,000 times to create a chance classification distribution for each patient.

[0193] Example 1.32, Statistical significance testing

[0194] To determine if the classifier performance using true labels significantly outperformed chance performance, Applicant compared the true test metrics (AUROC score and balanced accuracy measures) with the chance distributions estimated for each classifier. To estimate P values, Applicant implemented a randomization test where Applicant counted the number of values in the distribution that were greater than the test metric and divided the result by the total number of permutations (10,000).

[0195] Example 1,33. Model comparison

[0196] Within each model (that is, delta or daily model), the classifier performance was compared across the four different sets of feature inputs (that is, output measures) in a pairwise fashion using a non-paramctric statistical significance test (DcLong’s test). This test allows for comparison of two AUROCs calculated on the same dataset. The analysis was performed using the auroc-matlab package in MATLAB.

[0197] Later iterations of the linear autoregressive model used just a single lag term; in Equation 3, only <P1. Moreover, because the data was z-scored, its mean was 0, so the constant term c was unnecessary and removed. This resulted in the model having just a single learnable parameter. The predictive mechanics of the AR(1) model are shown in FIG. 7F, where for each time point, a prediction is made based on the product of the previous term in the series and a learned coefficient. This model is expressed by Equation 7.

[0198] In Equation 7, Xtis the value of the series at time t, etis an error term, andCP1is the learned coefficient of the model.

[0199] After predictions are made, the R2feature is calculated in the same way as the larger model, comparing the model’s predictions to ground truth z-scored LFP power, as shown in FIGS. 7C and 7E. Although results from this simplified model arc not reported in this example, its performance was similar to that of the larger model.

[0200] Alternative training strategies were also employed. Notably, FIGS. 7A-7G show a sliding window being applied to an example patient. At each iteration, the first two days of data in the window were used to fit the autoregressive model, the model’s parameters were locked, and predictions were made on the third day, which was the day of interest. Unlike the state-specific strategy described previously, where the autoregressive model was fit across a wide zone of the patient’s data using a cross validation method, this strategy is entirely causal. Both strategies resulted in similar performance.

[0201] Without further elaboration, it is believed that one skilled in the ail can, using the description herein, utilize the present disclosure to its fullest extent. The embodiments described herein are to be construed as illustrative and not as constraining the remainder of the disclosure in any way whatsoever. While the embodiments have been shown and described, many variations and modifications thereof can be made by one skilled in the art without departing from the spirit and teachings of the invention. Accordingly, the scope of protection is not limited by the description set out above, but is only limited by the claims, including all equivalents of the subject matter of the claims. The disclosures of all patents, patent applications and publications cited herein are hereby incorporated herein by reference, to the extent that they provide procedural or other details consistent with and supplementary to those set forth herein.

Claims

CLAIMS1. A method of assessing a psychiatric disorder in a subject, said method comprising: measuring a neural activity of the subject’s brain; and classifying the psychiatric disorder based on the measured neural activity.

2. The method of claim 1, wherein the measurement of neural activity comprises measuring changes in the periodicity of neural signals relative to average periodicity.

3. The method of claim 1, wherein the neural activity is derived from the subject’s prefrontal cortical area, deep gray area, ventral striatum (VS), ventral capsule (VC), or combinations thereof.

4. The method of claim 1 , wherein the neural activity is derived from the subject’s ventral striatum.

5. The method of claim 1, wherein the measurement of the neural activity occurs continuously.

6. The method of claim 5, wherein the measurement of the neural activity occurs continuously from the ventral striatum.

7. The method of claim 1, wherein the measurement of neural activity comprises measuring neural activity at or near the theta-alpha border, wherein the theta-alpha border is at or near 9 Hz.

8. The method of claim 7, wherein the measurement of neural activity comprises measuring neural activity of the ventral capsule (VC) or ventral striatum (VS) at or near the theta-alpha border.

9. The method of claim 1, wherein the neural activity is measured from one or more electrodes implanted into the subject’s brain.

10. The method of claim 9, wherein the one or more electrodes are implanted into the ventral striatum (VS) of the subject’s brain.

11. The method of claim 9, wherein the one or more electrodes are implanted into the ventral striatum (VS) and ventral capsule (VC) of the subject’s brain.

12. The method of claim 9, further comprising a step of implanting one or more electrodes into the subject’s brain.

13. The method of claim 1, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as treatment-resistant psychiatric disorder, severely symptomatic psychiatric disorder, treatment-sensitive psychiatric disorder, moderately symptomatic psychiatric disorder, or combinations thereof.

14. The method of claim 1, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as treatment-resistant psychiatric disorder, wherein treatment-resistant psychiatric disorder is characterized by non-responsiveness of the subject’s psychiatric disorder symptoms to drug treatment.

15. The method of claim 1, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as severely symptomatic psychiatric disorder.

16. The method of claim 15, wherein the severely symptomatic psychiatric disorder is characterized by disinhibited behavior, wherein the disinhibited behavior comprises at least one of recklessness, relatively increased libido, or combinations thereof.

17. The method of claim 1, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as treatment-sensitive psychiatric disorder, wherein treatment- sensitive psychiatric disorder is characterized by the responsiveness of the subject’s psychiatric disorder symptoms to drug treatment.

18. The method of claim 1, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as moderately symptomatic psychiatric disorder.

19. The method of claim 18, wherein the moderately symptomatic psychiatric disorder is characterized by a lack of disinhibited behavior, wherein the disinhibited behavior comprises at least one of recklessness, relatively increased libido, or combinations thereof.

20. The method of claim 1, wherein the psychiatric disorder is selected from the group consisting of obsessive-compulsive disorder, depression, bipolar disorder, post-traumatic stress disorder, substance use disorders, or combinations thereof.

21. The method of claim 1, wherein the psychiatric disorder comprises obsessive-compulsive disorder.

22. The method of claim 1, wherein the method occurs through the utilization of an algorithm.

23. The method of claim 22, wherein the method comprises: feeding the measured neural activity of the subject’ s brain into the algorithm; and utilizing the algorithm to classify the psychiatric disorder based on the measured neural activity.

24. The method of claim 23, wherein the algorithm comprises a machine-learning algorithm trained on the measured neural activity.

25. The method of claim 24, wherein the machine-learning algorithm comprises an autoregressive model, wherein for each time point of neural activity, the autoregressive model makes a prediction on the value of the next time point based on the value of a previous time point and a learned coefficient.

26. The method of claim 25, wherein the autoregressive model is represented by the following formula:•> wherein Xtrepresents measured neural activity at a certain time point, strepresents an error term, and <7^ represents the learned coefficient of the model.

27. The method of claim 1, further comprising a step of implementing a treatment based on the classification of the psychiatric disorder.

28. The method of claim 27, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as treatment-sensitive psychiatric disorder, moderately symptomatic psychiatric disorder, or combinations thereof, and wherein the treatment comprises administering a therapeutic agent to the subject.

29. The method of claim 27, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as treatment-resistant psychiatric disorder, severely symptomatic psychiatric disorder, or combinations thereof, and wherein the treatment comprises neuromodulation.

30. The method of claim 29, wherein the neuromodulation comprises deep brain stimulation (DBS) therapy.

31. The method of claim 1, wherein the subject is a human being suffering from psychiatric disorder.

32. A system for assessing psychiatric disorder in a subject, wherein the system comprises one or more computer readable storage mediums having a program code embodied therewith, wherein the program code comprises: programming instructions for measuring a neural activity of the subject’s brain; and programming instructions for classifying the psychiatric disorder based on the measured neural activity.

33. The system of claim 32, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as treatment-resistant psychiatric disorder, severely symptomatic psychiatric disorder, treatment-sensitive psychiatric disorder, moderately symptomatic psychiatric disorder, or combinations thereof.

34. The system of claim 32, wherein the system is associated with one or more electrodes operable to measure the neural activity of the subject’s brain.

35. The system of claim 25, wherein the system comprises an algorithm operable to implement the programming instructions, wherein the algorithm receives the measured neural activity of the subject’s brain and classifies the psychiatric disorder based on the measured neural activity.

36. The system of claim 35, wherein the algorithm comprises a machine-learning algorithm trained on the measured neural activity.

37. The system of claim 35, wherein the machine-learning algorithm comprises an autoregressive model, wherein for each time point of neural activity, the autoregressive model makes a prediction on the value of the next time point based on the value of a previous time point and a learned coefficient.

38. The system of claim 37, wherein the autoregressive model is represented by the following formula:wherein Xt represents measured neural activity at a certain time point, etrepresents an error term, and , represents the learned coefficient of the model.

39. The system of claim 25, further comprising programming instructions for recommending or implementing a treatment based on the classification of the psychiatric disorder.

40. The system of claim 39, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as treatment-sensitive psychiatric disorder, moderately symptomatic psychiatric disorder, or combinations thereof, and wherein the treatment comprises administering a therapeutic agent to the subject.

41. The system of claim 39, wherein the classification of the psychiatric disorder comprises classification of the psychiatric disorder as treatment-resistant psychiatric disorder, severely symptomatic psychiatric disorder, or combinations thereof, and wherein the treatment comprises neuromodulation.

42. The system of claim 41, wherein the system is operable to implement the neuromodulation.

43. The system of claim 41, wherein the psychiatric disorder is selected from the group consisting of obsessive-compulsive disorder, depression, bipolar disorder, post-traumatic stress disorder, substance use disorders, or combinations thereof.

44. The system of claim 41, wherein the psychiatric disorder comprises obsessive-compulsive disorder.

Images