Functional brain mapping of naturalistic behavior with stereoelectroencephalography

Abstract

Systems and methods for electroencephalography (EEG) mapping based on naturalistic behaviors are provided. The EEG mapping does not rely on trial-averaging and thus does not require repetitive tasks. Instead, it enables mapping of various types of naturalistic behaviors, such as conversation, which are more familiar for a patient, more ecologically valid, and useful for pediatric patients or patients with cognitive delays.

Description

Attorney Docket No. 0010872.0813187 PATENTFUNCTIONAL BRAIN MAPPING OF NATURALISTIC BEHAVIOR WITH STEREOELECTROENCEPHAEOGRAPHYCROSS-REFERENCE TO RELATED APPLICATION

[0001] The present application claims priority to U.S. Provisional Patent Application Serial No. 63 / 723,926, filed on November 22, 2024, the disclosure of which is incorporated herein in its entirety.STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

[0002] This invention was made with government support under NS 115929 awarded by the National Institutes of Health. The government has certain rights in the invention.BACKGROUND

[0001] Intracranial cognitive function mapping, particularly for language, is an essential component in the preoperative evaluation and surgical planning for patients with epilepsy. This procedure aims to identify and preserve eloquent cortex responsible for various cognitive functions, including language, memory, movement, and sensation, during resective surgery. By doing so, it minimizes the risk of post-operative deficits and optimizes surgical outcomes. Traditionally, electrical stimulation mapping (ESM) is used for such mapping. This technique involves the direct application of electrical current to specific brain regions to temporarily disrupt function, allowing for the identification of areas crucial for various cognitive processes, with a particular emphasis on language processing. However, ESM has several limitations, including the risk of after-discharges or seizures, the time-consuming nature of the procedure, and potential discomfort for the patient. Pediatric patients, and those with behavioral or cognitive impairments, frequently have limited ability to comply with ESM, further rendering them vulnerable to postsurgical cognitive deficits.

[0002] Task-related high gamma power modulation (HGM) analysis of intracranial electroencephalography (iEEG) data has emerged as a promising alternative to ESM for mapping various cognitive functions. Task-related modulations in high gamma activity, typically in theAttorney Docket No. 0010872.0813187 PATENT frequency range of 50-150 Hz, has been shown to correlate strongly with localized cortical processing and neuronal firing rates. Task-related HGM can accurately identify cortical areas related to various cognitive functions with high spatial and temporal resolution, offering a safer and potentially more efficient mapping approach. The conventional ESM as well as HGM-based mapping paradigm involves the presentation of structured tasks to the patient while recording iEEG data. For language mapping, for example, these tasks may include picture naming, auditory sentence comprehension, and verb generation. Similar task paradigms can be used for other cognitive domains. By analyzing the modulation of high gamma power during these tasks, researchers and clinicians can identify cortical regions involved in various aspects of cognitive processing.

[0003] However, the current approach to HGM-based cognitive mapping, especially for language, faces several challenges. Common tasks require sustained attention and cooperation from the patient, which can be difficult to maintain, especially in post-operative clinical settings and / or with pediatric patients. Structured tasks may not fully reflect the complexity and diversity of naturalistic cognitive processes, potentially missing important aspects of brain function. Standard tasks often focus on specific subdomains (e.g„ object naming, sentence comprehension for language) and may not engage all cognitive subdomains. The efficacy of specific tasks can vary between patients due to factors such as education level, language proficiency, attend on / behavi oral impediments, and cognitive abilities. Additionally, administering a comprehensive battery of tasks for various cognitive functions can be time-consuming, potentially limiting the practical application of ESM or HGM-based mapping in clinical settings.BRIEF SUMMARY

[0004] Some embodiments provided herein include a system for cognitive function mapping based on naturalistic behaviors. The system can include a plurality of electrodes, such as intracranial electrodes, configured to detect electrical activity of a brain of a subject and a monitoring device configured to record naturalistic behaviors of the subject. The system can further include one or more processors and a non-transitory computer-readable storage medium storing instructions that, when executed by the one or more processors, cause the system to receive intracranial electroencephalography (iEEG) data from the plurality of intracranial electrodesAttorney Docket No. 0010872.0813187 PATENT during naturalistic behaviors of the subject and receive behavior data from the monitoring device corresponding to the naturalistic behaviors. The system can synchronize the iEEG data with the behavior data and perform Time-Frequency Representation (TFR) calculations on the iEEG data to compute spectral power across multiple frequency bins as a function of time. The system can also calculate correlation coefficients between the computed spectral power across the multiple frequency bins and the behavior data and generate quantitative markers indicating statistical significance of correlation between spectral power and behavior.

[0005] Some embodiments provided herein include a method for intracranial language mapping based on naturalistic behaviors. The method can include receiving intracranial electroencephalography (iEEG) data from a plurality of intracranial electrodes implanted in a brain of a subject during naturalistic behaviors, receiving behavior data corresponding to the naturalistic behaviors of the subject and synchronizing the iEEG data with the behavior data. The method can also include performing Time-Frequency Representation (TFR) calculations on the iEEG data to compute spectral power across multiple frequency bins as a function of time and correlating the computed spectral power across the multiple frequency bins and the behavior data. A functional language map can be generated based on the correlation between spectral power and behavior.

[0006] Some other embodiments provided herein include a system for cognitive mapping based on naturalistic behaviors that includes a plurality of electrodes configured to detect electrical activity of a brain. Each of the plurality of electrodes is an intracranial electrode or an extracranial electrode. The system also includes a monitoring device configured to record naturalistic behaviors of a subject, one or more processors, and a non-transitory computer-readable storage medium storing instructions. When the instructions are executed by the one or more processors, the system can receive electroencephalography (EEG) data from one or more of the plurality of electrodes during naturalistic behaviors of the subject, receive behavior data from the monitoring device corresponding to the naturalistic behaviors, synchronize the EEG data with the behavior data. The system can also correlate changes in the spectral power across the multiple frequency bins to the behavior data and based on the correlation between spectral power and behavior, generate a map of predicted sites.Attorney Docket No. 0010872.0813187 PATENTBRIEF DESCRIPTION OF THE DRAWINGS

[0007] This application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0008] Those of skill in the art will understand that the drawings described below are for illustrative purposes only. The drawings are not intended to limit the scope of the present teachings in any way.

[0009] FIG. 1 depicts a system for intracranial language mapping based on naturalistic behaviors of a subject in accordance with a non-limiting embodiment.

[0010] FIG. 2 is a flow chart of an example method for intracranial language mapping based on naturalistic behaviors of a subject in accordance with a non-limiting embodiment.

[0011] FIG. 3 depicts a Behavior-iEEG-Spectral-Power-Correlation (BESPoC) methodology in accordance with a non-limiting embodiment.

[0012] FIG. 4 is a table showing task-wise comparison of BESPoC and HGM for localization of clinical language ESM sites.

[0013] FIG. 5 is table showing BESPoC association with HGM magnitude and localization of sites with significant task-related HGM.

[0014] FIG. 6 is a table showing the validity of BESPoC for HGM magnitude and for localization of sites with significant task-related HGM.

[0015] FIG. 7 shows locations of sites showing significant BESPoC and HGM activations on standardized brain.

[0016] FIG. 8 is a table showing a Task-wise comparison of BESPoC and HGM for localization of clinical language ESM sites.

[0017] FIG. 9 shows comparisons of BESPoC and HGM analysis of language tasks for localizing clinical language ESM.Attorney Docket No. 0010872.0813187 PATENT

[0018] FIG. 10 provides task-specific comparisons in model performance between BESPoC and HGM analysis for neuropsychological outcomes.

[0019] FIG. 11 is a table showing an association with change in NPE scores.

[0020] FIG. 12 is a table showing a comparison of BESPoC and HGM for association with neuropsychological outcomes.

[0021] FIG. 13 is a table showing a comparison of BESPoC and HGM for association with neuropsychological outcomes.

[0022] FIG. 14 is a table showing significant differences in cortical activation patterns with naturalistic conversation and conventional tasks.

[0023] FIG. 15 is a table showing differences in cortical activation patterns with naturalistic conversation and conventional tasks.

[0024] FIG. 16 depicts BESPoC localization of sites showing significant HGM.

[0025] FIG. 17 depicts locations of sites showing significant BESPoC and HGM activations on standardized brain for picture naming, story listening, and auditory naming.

[0026] FIG. 18 depicts BESPoC and HGM for localization of clinical ESM speech / language sites on standardized brain.

[0027] FIG. 19 depicts a graph comparing BESPoC analysis results from two different audio sources for patient speech and parent speech.

[0028] FIG. 20 depicts a graph comparing BESPoC analysis results from audio extracted from video-EEG recordings against task microphone recordings.

[0029] FIG. 21 depicts a graph comparing BESPoC analysis results from two different audio sources for patient speech and parent speech.

[0030] FIG. 22 depicts a graph comparing BESPoC analysis results from two different audio sources for patient speech and parent speech.Attorney Docket No. 0010872.0813187 PATENTDETAILED DESCRIPTION

[0031] Unless otherwise noted, terms are to be understood according to conventional usage by those of ordinary skill in the relevant art. In case of conflict, the present document, including definitions, will control. Exemplary methods and materials are described below, although methods and materials similar or equivalent to those described herein may be used in practice or testing of the present invention. All publications, patent applications, patents and other references mentioned herein are incorporated by reference in their entirety. The materials, methods, and examples disclosed herein are illustrative only and not intended to be limiting. The methods may comprise, consist of, or consist essentially of the elements of the methods as described herein, as well as any additional or optional element described herein or otherwise useful in the mapping of various cognitive functions, such as, without limitation, language, memory, movement, and sensation.

[0032] As used herein and in the appended claims, the singular forms “a.” “and.” and “the” include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to “a method” includes a plurality of such methods and reference to “a dose” includes reference to one or more doses and equivalents thereof known to those skilled in the art, and so forth.

[0033] The term “about” or “approximately” means within an acceptable error range for the particular value as determined by one of ordinary skill in the art, which will depend in part on how the value is measured or determined, e.g., the limitations of the measurement system. For example, “about” may mean within 1 or more than 1 standard deviation, per the practice in the art. Alternatively, “about” may mean a range of up to 20%, or up to 10%, or up to 5%, or up to 1% of a given value. Alternatively, particularly with respect to biological systems or processes, the term may mean within an order of magnitude, preferably within 5 -fold, and more preferably within 2- fold, of a value. Where particular values are described in the application and claims, unless otherwise stated the term “about” meaning within an acceptable error range for the particular value should be assumed.

[0034] The terms “individual,” “host,” “subject,” and “patient” are used interchangeably to refer to an animal that is the object of treatment, observation and / or experiment. Generally, the term refers to a human patient, but the methods and compositions may be equally applicable toAttorney Docket No. 0010872.0813187 PATENT non-human subjects such as other mammals. In some aspects, the terms refer to humans. In further aspects, the terms may refer to children.

[0035] The following non-limiting examples are provided to further illustrate embodiments of the invention disclosed herein. It should be appreciated by those of skill in the art that the techniques disclosed in the examples that follow represent approaches that have been found to function well in the practice of the invention, and thus may be considered to constitute examples of modes for its practice. However, those of skill in the art should, in light of the present disclosure, appreciate that many changes may be made in the specific embodiments that are disclosed and still obtain a like or similar result without departing from the spirit and scope of the invention.

[0036] The presently disclosed systems and methods are directed to a Behavior-EEG Spectral Power Correlation, referred to herein as BESPoC, which generally is an approach for iEEG functional mapping. In accordance with various embodiments. BESPoC does not rely on trialaveraging and thus does not require repetitive tasks. Instead, BESPoC enables mapping of naturalistic behavior, such as conversation or other naturalistic behaviors, which is more familiar for the patient, more ecologically valid, and could be especially useful for pediatric patients or patients with cognitive delays. In accordance with some embodiments of the present disclosure, for example. BESPoC can identify cognitive substrates underlying expressive and receptive language processing from spontaneous conversation. Moreover, beyond language-related processing, methodologies in accordance with the present disclosure can be used for mapping a variety of cognitive functions, such as, without limitation, memory, movement, and sensation, among others.

[0037] Identification of cortical areas supporting speech and language functions is critical for successful epilepsy surgery. The standard-of-care for presurgical language localization relies on electrical stimulation mapping (ESM) to prevent post-surgical speech / language deficits (1). However, it is associated with undesirable seizures and after-discharges (ADs), which threaten both patient safety and the validity (2, 3). Clinical ESM requires active, sustained, cooperation from patients, so that individual cortical sites may be evaluated sequentially through multiple stimulation settings, and using different tasks, usually over the course of several prolonged sessions (4, 5). This is necessary, partly, because no single task can recruit the entire languageAttorney Docket No. 0010872.0813187 PATENT system (6, 7). Pediatric patients, or those with behavioral or cognitive impairments, frequently have limited ability to comply with ESM, rendering them more vulnerable to postsurgical cognitive deficits (6. 8).

[0038] Another technique for language mapping relies on task-related high-gamma power augmentations in intracranial EEG (9, 10). These intracranial high-gamma modulations (HGMs) have been shown to correlate with neuronal population firing rates, hemodynamic responses on functional MRI, and glucose metabolism on positron emission tomography (11-13). HGM language mapping does not involve the risks associated with ESM and can be performed quickly using multiple tasks because it involves simultaneous recording from the entire intracranial electrode array (9. 14). Studies have also shown emergence of cognitive and linguistic deficits after lesioning of HGM language sites during epilepsy surgery, even when those sites were declared safe by ESM (15-18). However, computing HGMs requires trial- averaging spectral power between test and rest conditions (9). Hence, HGM mapping also depends on patients’ ability to perform unfamiliar, repetitive, artificial tasks, on demand and with precision, under stressful postoperative conditions, imposing limitations on some patients to perform HGM language mapping, similar to ESM (19).

[0039] Naturalistic language, such as conversation, alleviates many of the challenges posed by conventional trial-based language tasks (20). While conventional tasks may be challenging to perform and interpret in patients with speech / language impediments, conversation can conform to the patient’s abilities and stage of language development. Involving family members with the linguistic task can further improve patient engagement, compared to structured testing. Compared to conventional language tasks, naturalistic conversation is more likely to involve both expressive and receptive language functions in parallel, engage language-related visual, motor, and memory domains, and integrative or association cortices (20-22). While more complex tasks such as auditory naming can activate both expressive and receptive language to some degree, these are often too demanding for pediatric patients (16, 18). Hence, use of naturalistic language tasks in presurgical mapping is appealing, but trial-based analysis is not compatible with naturalistic behaviors.Attorney Docket No. 0010872.0813187 PATENTPrevious work on language mapping with conversation has been limited, investigating neural signatures of turn taking and response planning using event-related potentials and lower EEG frequency bands (23-27). A few studies have explored conversation for language mapping by constructing artificial epochs using a fixed time window or by splitting conversation into “idea units” (20, 21, 28-30). However, these studies lacked validation against established methods of intracranial language mapping.

[0040] In accordance with the present disclosure, systems and methods for Behavior-iEEG- Spectral-Power-Correlation (BESPoC) can offer an alternative to trial- averaged ESM and HGM. The development and validation of the BESPoC approach for intracranial language mapping in a large, diverse cohort of patients with drug-refractory epilepsy (DRE) is provided herein. As provided below, BESPoC is tested not just against ESM and HGM language maps, but its performance in predicting a comprehensive battery of postsurgical neuropsychological outcomes is also evaluated, which represent the ultimate ground truth for any functional mapping modality. Specifically, BESPoC is first validated using conventional language tasks and then it is applied to naturalistic conversation. BESPoC in accordance with the present disclosure can provided for presurgical functional mapping for a variety of patients, including those who cannot participate in the current standard-of-care trial-based testing.

[0041] Referring now to FIG. 1 , an example system 100 for cognitive mapping based on naturalistic behaviors can comprise a plurality of electrodes 102 configured to detect electrical activity of a brain of a subject 104. The electrodes 102 can be positioned anywhere in, on, and / or around the subject's brain, as the electrodes 102 can be any suitable type of intracranial electrodes, extracranial electrodes, or a combination thereof. Thus, while some examples provided in this disclosure specifically utilize intracranial electrodes for sampling intracranial EEG, it should be appreciated that the present disclosure encompasses any type of electrodes, or combination of different types of electrodes, that can be used to sample any type of physiological signal from the subject 104.

[0042] In some embodiments, the system 100 can comprise between 4 and 23 electrodes 102 that can be implanted left, right, and bilaterally into the subject 104, with 42 to 254 electrode contacts corresponding to 38 to 233 bipolar channels, for example. A monitoring device 110 canAttorney Docket No. 0010872.0813187 PATENT be configured to record naturalistic behaviors of the subject 104. By way of example, systems utilizing spontaneous conversation as a naturalistic behavior, the monitoring device 100 can be configured to record a conversation in which the subject 104 participates. More specifically, in some embodiments, the naturalistic behaviors can comprise a spontaneous conversation between the subject 104 and a partner, where a voice of the subject is recorded on a first microphone channel 112 and a voice of the partner is recorded on a second microphone channel 114.

[0043] The system 100 can further include one or more processors and a non-transitory computer-readable storage medium storing instructions, schematically depicted as computing device 116, that, when executed by the one or more processors, cause the system to perform various functions. Such functions can include, without limitation, receiving (EEG) data, such as intracranial electroencephalography data (iEEG), from the plurality of electrodes 102 during naturalistic behaviors of the subject 104 and receiving behavior data from the monitoring device 110 corresponding to the naturalistic behaviors. With regard to the naturalistic conversational behaviors, for example, conversation analysis can be performed on each of the first microphone channel 112 and the second microphone channel 114.

[0044] In some embodiments, the iEEG data is wavelet-transformed (i.e., 50-150 Hz). The iEEG data can be synchronized with the behavior data, and the synchronized data can be processed using time-frequency analysis to compute spectral power across multiple frequency bins as a function of time. Time-Frequency Representation (TFR) calculations can be performed on the iEEG data, focusing on the 50-150 Hz range for example, which encompasses high gamma activity. It is noted, however, that this disclosure is not limited to analysis of the high gamma frequencies. Further, while wavelet transformation is one example approach for TFR calculation, this disclosure is not so limited, as BESPoC can integrate various methods for TFR calculations. Regardless of the specific TFR calculation method employed, the computed spectral power across the multiple frequency bins and the behavior data can be correlated. Permutation testing can be used for determining the statistical significance of the correlation. A non-limiting approach to power calculation and behavioral trigger evaluation are provided in Example 1, below.

[0045] The association between log-transformed power on each stereoelectroencephalography (SEEG) channel and each behavioral channel can be quantified withAttorney Docket No. 0010872.0813187 PATENTPearson’s correlation coefficient, although any suitable correlation calculation can be utilized. In some embodiments, this correlation value is assigned a probability based on the sampling distribution of con-elation coefficients obtained by randomizing behavioral channel data. Due to the large dataset (often >100 channels / patient with several minutes of SEEG recordings), the sampling distribution can be obtained by, for example: (1) taking a random 5s power segment and randomly sampling a 5s segment of the behavioral channel with replacement. (2) computing the correlation between these two 5s segments, and (3) scaling the correlation coefficient by m / n, where m is the length of the subsample (5s), and n is the length of the total recording (47). Because the power distributions are normal and the behavioral channels are binary, the distribution of correlation values within each frequency bin on each channel also approximated a normal distribution. Therefore, a gaussian curve can be fit to the distribution of correlation values, and the resulting mean and standard deviation for each frequency on each channel can be used to convert the correlation values to Z-scores. Finally, corresponding p-values can be obtained from the normal cumulative distribution function. These p-values (and Z-scores) can serve as a quantitative marker of the correlation between SEEG power and patient behavior channels.

[0046] Some embodiments can additionally include positive or negative time lag to the trigger channel and repeat the process again such that changes in correlation can be examined. A map of predicted sites can be generated and, in some implementations, the generated map can be provided to a user interface for display. In accordance with the present disclosure, the map can be generated without trial-averaging repetitive tasks performed by the subject 104.

[0047] Referring now to FIG. 2, a flow chart 200 of an example method for functional m mapping of naturalistic behaviors is provided. In this example embodiment, intracranial language mapping is depicted, although this disclosure is not so limited. In accordance with various embodiments, a method for intracranial language mapping based on naturalistic behaviors can include, at 202, receiving intracranial electroencephalography (iEEG) data from a plurality of intracranial electrodes implanted in a brain of a subject during naturalistic behaviors. While the method of FIG. 2 utilizes intracranial electrodes, other embodiments may utilize EEG data from extracranial electrodes, for example. At 204, behavior data can be received corresponding to the naturalistic behaviors of the subject and, at 206, the iEEG data can be synchronized with the behavior data. The naturalistic behaviors can include, without limitation, spontaneousAttorney Docket No. 0010872.0813187 PATENT conversation between the subject and a partner. By way of example a voice of the subject can be recorded on a first microphone channel and a voice of the partner can be recorded on a second microphone channel such that each of the first microphone channel and the second microphone channel can be independently analyzed. Other approaches, however, can use a variety of other types of naturalistic behaviors. Time-Frequency Representation (TFR) calculations can be performed on the iEEG data to compute spectral power across multiple frequency bins as a function of time at 208. At 210 the computed spectral power across the multiple frequency bins and the behavior data can be correlated. At 212, based on a statistically significant correlation between spectral power and behavior, a functional language map can be generated in accordance with various embodiments.Example 1

[0048] Objective. Intracranial language localization with electrical stimulation mapping (ESM) and high-gamma modulation (HGM) mapping relies on artificial, repetitive tasks, requiring sustained cooperation from patients. Herein the validity of unstructured, interpersonal, naturalistic conversation for language localization, using a novel methodology was tested: Behavior-iEEG- Spectral-Power-Correlation (BESPoC). BESPoC was first validated against ESM, HGM, and neuropsychological outcomes using well-established language tasks, then demonstrated the validity of naturalistic conversation.

[0049] Methods. The testing included 134 patients (59 females), aged 2-29 years, undergoing standard-of-care stereo-EEG monitoring who engaged in picture naming, auditory naming, story listening, and conversed with a family member. ESM and HGM analysis were performed using established methods. BESPoC methodology quantified correlation between stereo-EEG spectral power from task recordings, obviating the need for any trial-based epochs, and behavioral markers. The large sample size allowed mixed-effects modeling to compare BESPoC with HGM and ESM.

[0050] Results. BESPoC showed high specificity (0.79-0.83) and sensitivity (0.64-0.86) for localizing HGM language sites across the tasks. BESPoC also compared well with HGM across all language tasks for localizing ESM speech / language sites. With conventional tasks, BESPoC was superior to HGM for modeling neuropsychological deficits seen despite preserving ESM speech / language sites. Naturalistic conversation compared well with standard tasks for localizationAttorney Docket No. 0010872.0813187 PATENT of HGM and ESM language sites, and determined neuropsychological outcomes better than conventional tasks.

[0051] Interpretation. Using BESPoC methodology naturalistic conversation is shown to produce valid cortical language maps of both expressive and receptive language, and determine neuropsychological outcomes after epilepsy surgery.1.0 METHODS1.1 Patients and SEEG acquisition

[0052] All DRE patients undergoing stereo-electroencephalography (SEEG)monitoring at Cincinnati Children’s Hospital who were native English speakers were included. SEEG signals were sampled at 2048 Hz with Natus Quantum amplifiers using Neuroworks 8.5software (Natus, Middleton WI). SEEG was recorded for the study with a referential montage, using the electrode contact farthest from the presumptive seizure-onset zone, free of artifact by clinician analysis, and outside the canonical peri-sylvian language cortex as the reference. The study was approved by the Cincinnati Children’s Institutional Review Board. Informed consent from patients >18 years- of-age and parental permission in others was obtained.1.2 Language Tasks1.2.1 Picture Naming (PN)

[0053] A series of 40 pictures from the Snodgrass set were presented on a monitor using E- Prime 3.0 (Psychology Software Tools, Sharpsburg PA), for 2-3 s based on patient comfort with the task (14, 31). Each image display was preceded by a Is inter- stimulus interval having a black screen. Patients were instructed to name the picture aloud immediately after the display. Images were pre-screened to ensure patients were able to identify the images, and the subset shown during the task were presented in a random order. Precise timing of image display was captured by a digital pulse input into the SEEG data stream from a photodiode. Patients’ audio output was collected with a lapel microphone and digital pulses corresponding to the onset and termination of patients’ voice were recorded on a separate channel synchronized to the SEEG data. Trials were aligned at image onset, with Is inter-trial baseline preceding the stimulus and 2s for patientAttorney Docket No. 0010872.0813187 PATENT response. 2sTrials with microphone activation during baseline or without microphone activation in the response window were excluded.1.2.2 Auditory Naming (AN)

[0054] A series of 50 pre-recorded 5 s modified dictionary definitions from the Hamberger- Seidel task were played through a sound bar (32). Each prompt was followed by 5s of silence for the patient to respond. Patients were instructed to name aloud the object being described (e.g. “An instrument you beat with sticks”, “Drum”), but to remain silent if they could not answer the prompt. While all the audio epochs were 5s long, the verbal descriptions had different lengths (mean ± standard deviation 3.60 ± 0.67s). The patient was free to respond immediately following the prompt, even before the 5s response window. Patient audio and prompt start times were synchronized to the SEEG data using automated digital triggers. Trials were aligned on the termination of auditory cues. Trials with microphone activation during the description or inter-trial baseline, or without microphone activation during the response window, were excluded.1.2.3 Story Listening (SL)

[0055] A series of 40 pre-recorded 5s segments of children’s stories were played through a sound bar (19). Each segment was preceded by I s silence, 3s of broadband noise, and I s silence. The spectral envelope of the noise was identical to the preceding story segment. Patients were instructed to listen attentively. To reduce the task burden, only 20 trials were presented after late 2021 (57 patients). The timing of story and noise were recorded on separate channels synchronized to SEEG data. Trials were aligned on story onset so that the 5s silence-noise-silence cycle preceded each 5s stimulus. The 3 s of noise was treated as inter- trial baseline.1.2.4 Naturalistic Conversation (NC)

[0056] Approximately 5 minutes of conversation was captured between patient and a family member. Subjects were instructed to avoid speaking simultaneously, but otherwise the conversation was unstructured with free choices. Digital pulses for onset and termination of speech were recorded, and synchronized to the SEEG data on separate channels for the patient and the family member.Attorney Docket No. 0010872.0813187 PATENT1.3 ESM

[0057] Patients underwent standard-of-care language ESM with 50 Hz, biphasic, 250-300 ps, square wave pulses, in trains lasting 4-6s, such that 3-5 trials of the PN task could be performed during each stimulation (33, 34). The initial current of 0.5-1 mA was increased in 0.5-2 mA steps, until a language response, evolving AD’s, or an arbitrary maximum of 10 mA was reached. Electrode contacts on an individual SEEG electrode were stimulated in a bipolar manner from deep to superficial, and those showing / not showing reproducible speech / language interference were regarded as ESM+ and ESM- respectively.1.4 Neuropsychological Evaluations

[0058] Pre- and 1-year postsurgical neuropsychological evaluations (NPE) were analyzed to determine the cognitive effects of lesioning sites with significant HGM or BESPoC activation (6, 17). No ESM language site was lesioned in any patient. The tests included: (1) Wechsler Preschool and Primary Scale of Intelligence (4thedition) / Wechsler Intelligence Scale for Children (5thedition) / Wechsler Adult Intelligence Scale (4thedition) for Verbal Comprehension, Visual Spatial, Working Memory, and Processing Speed (Table 1); (2) Woodcock-Johnson Tests of Achievement IV, for Letter-Word Identification, Calculation, Spelling, and Passage Comprehension; (3) Wide Range Assessment of Memory and Learning (2ndedition), for Immediate, Delayed, and Recognition subtests of Story Memory and Verbal Learning respectively; (4) Peabody Picture Vocabulary Test (4th / 5theditions); (5) Visual Motor Integration (6th edition): and (6) Boston Naming Test. Reliable change indices (RCIs) were obtained for each NPE by subtracting presurgical scores from postsurgical scores and dividing by the standard error of mean for the presurgical scores. RCIs < -1.96 were regarded as significant declines. Additional details are found in Table 1.Attorney Docket No. 0010872.0813187 PATENTTable 11.5 Resected Electrode Contacts

[0059] Post-implantation computed tomographic (CT) scan and post-operative T1W-MRI were co-registered to the pre-operative T1 W-MRI via 6-parameter rigid body transformation using the SPM12 toolbox in MATLAB. Within the co-registered CT image, electrode contacts were localized using the FASCILE software (35). Electrode contacts within resected or ablated tissue were identified manually.Attorney Docket No. 0010872.0813187 PATENT1.6 HGM Analysis

[0060] HGMs during PN, SL, and AN were calculated using the standard trial-averaging approach (14, 17). Artifactual channels were first rejected using the previously published computational methods and visual review (14). The remaining channels were bipolar re-referenced along SEEG electrodes such that the deeper electrode contact was anode and the more superficial contact was cathode, and then notch-filtered at harmonics of 60 Hz. High-gamma power from the squared envelope of the 50-150 Hz bandpass filtered signal was calculated for the test condition (from visual or verbal prompt until after response) and rest condition (inter-trial rest or noise) separately. The log-transformed power between test and rest conditions was compared for each channel using independent samples t-tests with p-values Bonferroni corrected for the total number of SEEG channels in each patient.1.7 Methodology of BESPoC

[0061] Referring to FIG. 3, an example BESPoC methodology is schematically depicted. A channel localized to the left superior temporal gyrus is shown as an example during the picture naming task. Synchronized patient microphone and picture display channels are illustrated. StereoEEG recordings are wavelet-transformed, then correlated with patient microphone channel. Distributions of correlation coefficients are calculated for each frequency bin, and tested for statistical significance.1.7.1 Power Calculation

[0062] Channels were bipolar re-referenced and artifactual channels were rejected as described for HGM. An anti-aliasing filter was applied and channels were down-sampled to 512 Hz. Then, SEEG data or the entire duration of the task, without any epochs, was convolved with a series of wavelets to obtain a broad-band time-frequency representation (TFR) of power, as shown in FIG. 3. These wavelets were centered at 16 logarithmically spaced frequencies between 5-200 Hz, with a Gaussian window of 7 / cycles, where / is the central frequency of each wavelet. A single TFR of log-transformed power for each SEEG channel with 16 frequency bins was obtained for the entire recording.1.7.2 Behavioral ChannelsAttorney Docket No. 0010872.0813187 PATENT

[0063] Each task was represented by two behavioral channels: patient microphone and picture display for PN; patient microphone and audio playback for AN; story and noise for SL; and patient microphone and partner microphone for NC. Behavioral channels were resampled to 512 Hz and binarized to values of 0 or 1.1.7.3 Correlation Between SEEG Spectral Power and Observed Behavior

[0064] The association between log-transformed power on each SEEG channel and each behavioral channel was quantified with Pearson’s correlation coefficient. This correlation value was assigned a probability based on the sampling distribution of correlation coefficients obtained by randomizing behavioral channel data. Due to the large dataset (often >100 channels / patient with several minutes of SEEG recordings), the sampling distribution was obtained by: (1) taking a random 5s power segment and randomly sampling a 5s segment of the behavioral channel with replacement, (2) computing the correlation between these two 5s segments, and (3) scaling the correlation coefficient by (m / n), where m is the length of the subsample (5s), and n is the length of the total recording (36). This distribution of correlation coefficients was also approximately normal, because the power distributions were approximately normal and the behavioral channels were binary Therefore, a gaussian curve was fit to this distribution and the resulting means and standard deviations for each frequency on each channel were used to convert the correlation values to z-scores and to obtain p-values from the normal cumulative distribution function. These p- values quantified the correlation between SEEG power and behavioral channels.

[0065] By adding positive or negative lag to the trigger channel and repeating this process again, changes in correlation can be examined in time. For example, adding a -200 ms lag to the patient microphone during picture naming may identify channels experiencing unusually high power in the high gamma range for the 200 ms (approximate duration of a one-syllable word) before vocalization. Similarly, a +200 ms lag in story listening may highlight channels involved in receptive language processing. Correlation values with time lags were calculated by clipping the start of the trigger channel and the end of the power timeseries for negative lags, or the end of the trigger and start of the power for positive lags. The subsampling distribution was retained, but the sampling rate and L isAttorney Docket No. 0010872.0813187 PATENT the lag time in seconds. Correlation coefficients were found for all channels at lags between -1.5 and 1.0 seconds at 100 ms intervals.1.8 Combining Tasks

[0066] Tasks were combined for analysis (e.g. PN+AN) by including intracranial sites “active” during any task in a combination for patients who participated in all tasks in a given combination. Note that AN and SL, both relying on auditory processing, were performed in a mutually exclusive fashion. We performed AN in patients >9 years-of-age and SL in younger patients and those who were unable to perform AN. For group-level analysis, the audio prompt and patient microphone channels of AN were regarded as equivalent to story and noise channels of SL respectively.1.9 Validation of BESPoC

[0067] The performance of BESPoC was evaluated against HGM, ESM, and NPEs, respectively, and BESPoC and HGM for determining ESM results and NPEs. For each analysis, only patients with all relevant measures were included.1.9.1 Comparison with HGM

[0068] BESPoC was validated both against statistically significant HGM language sites (binary variable) and the magnitude of HGM (continuous variable). For each task (PN, AN / SL, and NC) inferential generalized linear mixed-effects models (GLMMs) were fitted with significant HGM (binary) as the dependent variable, BESPoC (log-transformed p-values of correlations) as fixed effect and “patient” as the random effect. A receiver operating characteristic (ROC) curve was fitted and thresholds were selected to optimize sensitivity and specificity. 95% confidence intervals (Cis) were computed for sensitivity and specificity, and using DeLong’s method for area under the ROC curve (AUC).

[0069] Similar linear mixed-effects models were also fitted with log-transformed p-values of HGMs between the task (test) and baseline (rest) conditions as the (continuous) dependent variable.1.9.2 Comparison with ESMAttorney Docket No. 0010872.0813187 PATENT

[0070] Inferential GLMMs were again fitted for each task with ESM naming sites as the dependent variable, either BESPoC or HGM (log-transformed p-values) as fixed effects and patient as the random effect. Sensitivity, specificity, AUC, and their 95% Cis were obtained.1.9.3 Determining NPEs

[0071] It was hypothesized that post-surgical changes in NPEs would be driven by resected channels. Hence, linear regression models were fitted with NPE difference scores (postsurgical - presurgical) as dependent variables and BESPoC / HGM log-transformed p-values for channels recorded from resected / ablated tissue, as inputs, for every language task. Additionally, — log(p) was relied on, where p is the p-value of F-statistic for comparisons among different tasks / modalities and computed its 95% CI using patient-level leave-2-out cross-validation.

[0072] Logistic regressions were also fitted with significant NPE declines (RCI< -1.96) as dependent variables and BESPoC / HGM log-transformed p-values as inputs, for every language task. The inputs were scaled and models weighted for class imbalance. Sensitivity, specificity, AUC, and their 95% Cis were obtained. For each sub-analyses, p-values were Bonferroni corrected for the number of independent tests.2.0 RESULTS2.1 Cohort Details

[0073] 134 patients (59 female) were included in the study, aged 2.0 to 29.0 years (12.7 ± 5.2).Between 4 and 23 electrodes (13 ± 4) were implanted into left, right, and both hemispheres in 59, 34, and 41 patients, respectively, with 42 to 254 electrode contacts (145 ± 44) per patient, corresponding to 38 to 233 bipolar channels (132 ± 41). A total of 19,463 contacts (17,714 bipolar channels) were analyzed. Picture naming (PN), auditory naming (AN), story listening (SL), and naturalistic conversation (NC) tasks were performed by 122 (91%), 50 (37%), 78 (58%), and 72 (54%) patients, respectively. ESM was performed in 47 / 134 (35%) patients, with 891 stimulated contact pairs. NPE data were available for 36 patients with 17 ablations, 18 resections, and 1 patient with both resection and ablation, having 406 bipolar channels within the lesioned tissue. Additional details regarding patient demographics and involvement are provided in Table 2:Attorney Docket No. 0010872.0813187 PATENTTable 2Abbreviations in Table 2: ID identification number, PN picture naming, AN auditory naming, SL story listening, NC naturalistic conversation, ESM electrical stimulation mapping, NPE neuropsychological evaluation, R right, L left, B bilateral, A ambidextrous, U unassigned, M male, F female.2.2 Comparison of HGM

[0074] FIG. 4 is a table showing task-wise comparison of BESPoC and HGM for localization of clinical language ESM sites. FIG. 4 uses generalized linear mixed effects models with given variable(s) as fixed effects and patient as the random effect. FIG. 5 is table showing BESPoCAttorney Docket No. 0010872.0813187 PATENT association with HGM magnitude and localization of sites with significant task-related HGM. The data in FIG. 5 is based on linear and generalized linear (logistic) mixed effects models with given variables as fixed effects and patient as random effect. FIG. 16 depicts BESPoC localization of sites showing significant HGM, with ranges signify 95% CI. The generalized linear model confidence intervals calculated using DeLong's method (AUC) and Wilson's method (Sensitivity / Specificity). P- Values <0.0001 are significant after Bonferroni correction. FIG. 6 is a table showing the validity of BESPoC for HGM magnitude and for localization of sites with significant task-related HGM. BESPoC was validated against the magnitude of HGM at a given site (using linear mixed effects models) and for localization of sites with significant task-related HGM (using generalized linear mixed effects models) both fitted with patient as the random effect. P values were Bonferroni corrected. Note that model quality measures for BESPoC-NC are never significantly worse than across-task metric for the same HGM task.

[0075] For each language task, BESPoC localized HGM language sites with high sensitivity and high specificity (FIGS. 4-5). BESPoC was best at modeling HGM results for SL with 0.92 (95% CI 0.92-0.93) AUC, 0.86 (0.85-0.88) sensitivity, and 0.82 (0.81-0.83) specificity (FIGS. 5- 6, FIG. 16). BESPoC localized HGM for AN with AUC 0.83 (0.82-0.84), sensitivity 0.67 (0.65- 0.69), and specificity 0.83 (0.81-0.84); and for PN with AUC 0.78 (0.77-0.79), sensitivity 0.64 (0.62-0.65), and specificity 0.79 (0.78-0.79).

[0076] Significant linear relationships were also seen for HGM magnitude and BESPoC correlations (both continuous variables)). For all three tasks, both behavioral channels analyzed for BESPoC were significant regressors of HGM (p<0.0001), however, all three models had moderate R2values (0.30<R2<0.50). Among the three modalities for intracranial language mapping (ESM, HGM, and BESPoC), only BESPoC lends itself to analysis of NC. Hence, to evaluate BESPoC analysis of NC, its ability to localize HGM maps of conventional tasks was compared. BESPoC analysis of conversation localized HGM during conventional tasks with a 0.70-0.74 AUC, 0.63-0.77 specificity and 0.60-0.65 sensitivity (FIGS. 4-5, FIG. 16). FIG.7 displays anatomical locations of task-related activations, while comparisons between BESPoC prediction of significant HGM via logistic regression. Abbreviations in FIG. 7: Marker indicates mean, ranges signify standard deviation. PN picture naming, AN auditory naming, SL storyAttorney Docket No. 0010872.0813187 PATENT listening, NC naturalistic conversation, HGM high gamma power modulation, BESPoC Behavior- iEEG-Spectral-Power-Correlation, ROC receiver operating characteristic.

[0077] FIG. 7 illustrates anatomical locations of task-related activations, while comparisons between BESPoC and HGM are illustrated in FIG. 17, with FIG. 17 showing locations of sites showing significant BESPoC and HGM activations on standardized brain. More specifically, anatomic locations of significant BESPoC and HGM during PN, SL, AN, and NC are shown. Similar task-specific activation patterns emerge from both methods, suggesting BESPoC and HGM are extracting similar features of the data. Likewise, similar spatial patterns emerge across tasks identifying primarily expressive and receptive language sites. BESPoC PN evaluates patient microphone channel, BESPoC SL evaluates story audio channel, BESPoC AN evaluates both prompt audio and patient microphone, and BESPoC NC evaluates family member and patient microphones.2.3 Comparison with ESM

[0078] FIG. 8 is a table showing a Task-wise comparison of BESPoC and HGM for localization of clinical language ESM sites. FIG. 8 uses generalized linear mixed effects models with given variable(s) as fixed effects and patient as the random effect. Confidence intervals calculated using DeLong's method for AUC and Wilson's method for sensitivity / specificity. SL and AN are merged for "Combined" task. NC performs within 95% Cis for all models except BESPoC analysis of PN+SL and BESPoC-SL. FIG. 9 shows comparisons of BESPoC and HGM analysis of language tasks for localizing clinical language ESM. Point estimate indicates mean area under the receiver operating characteristic curve for the generalized linear mixed models while ranges signify 95% confidence intervals.

[0079] For conventional language tasks (PN, AN / SL) and combinations thereof, HGM and BESPoC results were within 95% Cis for localization of ESM naming sites when evaluating the same task (FIG. 4, FIG. 8, FIG, 9). AUCs ranged from 0.70-0.81 for BESPoC analysis and 0.69- 0.79 for HGM, showing no significant differences between the two methods. However, task selection did show some significant differences: BESPoC analysis of SL performed better than AN (BESPoC and HGM), NC (BESPoC), and AN+NC task (BESPoC), for localizing ESM. Also, PN+SL task (BESPoC) performed better than AN, PN+AN, NC, and AN+NC (BESPoC).Attorney Docket No. 0010872.0813187 PATENTGenerally, AN alone and in combination with other tasks was a poor localizer of language ESM, despite being performed in older and more cooperative patients. Rather, the best localization of ESM typically incorporated SL (BESPoC-PN+SL, BESPoC-SL, HGM-PN+SL, HGM-SL, BESPoC-SL+NC), a surprising result because ESM mainly relies on interference with expressive language including speech motor responses (33, 34) Anatomical locations of these comparisons are shown in FIG. 18 depicts BESPoC and HGM for localization of clinical ESM speech / language sites on standardized brain.

[0080] Although BESPoC and HGM showed good agreement between their respective abilities to localize ESM naming sites, there were important spatial differences among language tasks (FIGS. 5 and 7). PN showed frequent activation in the anterior language area compared to SL which was predominantly active around the superior temporal sulcus. Although NC activation, especially during partner speech, favored posterior language area, BESPoC-NC+ / ESM- sites extended into the anterior temporal lobe.2.4 Association with Neuropsychological Outcomes

[0081] FIG. 10 provides task-specific comparisons in model performance between BESPoC and HGM analysis for neuropsychological outcomes. The horizontal lines represent the 95% confidence intervals and the darker point within each range indicates the point estimate of —log p, where p is the p-value obtained from linear regression models, using BESPoC and HGM respectively, such that location towards right side of the figure indicates better performance. The vertical lines represent Bonferroni corrected (solid) and uncorrected (dashed) p<0.05. Line width indicates the number of channels included for each analyses (key in bottom right). FIG. 11 is a table showing the an association with change in NPE scores. Because no ESM sites were lesioned in any patients, the ability of each task analyzed with HGM and BESPoC to model changes across 17 NPEs (FIGS. 10-11). Data regarding patients and electrode contacts for each NPEare summarized in Table 3. Table 3 is a sampling of neuropsychological domains, which identifies the number of patients with pre- and post-operative evaluation of each domain, and the number of contacts which were resected or ablated in those patients.Attorney Docket No. 0010872.0813187 PATENTTable 3Abbreviations for Table 3: NPE neuropsychological evaluation, Pts number of patients, ECs number of electrode contacts, VC Verbal Comprehension, VS Visual Spatial, WM Working Memory, PS Processing Speed. LWI Letter-Word Identification, Calc Calculation, Spel Spelling, PC Passage Comprehension, SM Story Memory, VL Verbal Learning, I Immediate, D Delayed, R Recognition, BNT Boston Naming Test, VMI Visual Motor Integration, PPVT Peabody Picture Vocabulary Test.

[0082] Referring to FIG. 10, task-specific differences in model quality between BESPoC and HGM analysis of high gamma power on resected electrodes when predicting changes in NPE scores are shown. Endpoints of each horizontal line represent the predictive ability of models trained using BESPoC and HGM. Line width scales with the size of the training set (key in bottom right shows the line width associated with 100 channels). Model quality is calculated as log- transformed P-value of model F-statistic of ordinary least squares linear regression. HGM highAttorney Docket No. 0010872.0813187 PATENT gamma power modulation, BESPoC Behavior-iEEG-Spectral-Power-Correlation, VC Verbal Comprehension, VS Visual Spatial, WM Working Memory, PS Processing Speed, LWI Letter- Word Identification, Calc Calculation, Spel Spelling, PC Passage Comprehension, SM Story Memory, VL Verbal Learning, I Immediate, D Delayed, R Recognition, BNT Boston Naming Test, VMI Visual Motor Integration, PPVT Peabody Picture Vocabulary Test.

[0083] As FIG. 10 shows, overall BESPoC performed better than HGM for modeling postsurgical neuropsychological outcomes for all domains except story memory recall (with PN task) and processing speed (with AN task). For most models, the two methods were within 95% Cis. Crucially, HGM models were never significantly better than BESPoC for any task, but BESPoC was significantly better than HGM in 35 of 100 direct comparisons (same task, same NPE). Further, only one HGM model (PN+AN for SM-R) achieved statistical significance after Bonferroni correction, compared to 34 BESPoC models (Table S6).

[0084] Combining tasks further improved the performance of BESPoC, with the composite task (PN, AN / SL, and NC) significantly associated with 6 / 17 NPEs, and PN+SL task associated with 8 / 17 NPEs. Letter-word index (6) and passage comprehension (5) were modeled by the most tasks, all BESPoC models. Six NPEs: visual spatial, calculation, spelling, immediate story memory, immediate verbal learning, and Boston naming test, were not associated with any task.

[0085] FIGS. 12-13 are tables showing a comparison of BESPoC and HGM for association with neuropsychological outcomes. BESPoC analyses of task combinations were consistently able to predict significant deficits (RCI< -1.96) across nearly all neuropsychological domains, generally with higher AUC compared to HGM analysis, though both methods were always within 95% Cis of each other for the same task / NPE (FIG. 12-13). BESPoC analysis of the composite task (PN, SL / AN, and NC) produced the best models (lower 95% CI for AUC>0.70), modeling significant declines across 11 / 17 NPEs. FIGS. 12-13 shows the predictive significance per neuropsychological test. F-statistic (top of cell) and P- value (bottom of cell) of ordinary least squares linear regression of changes in NPE scores on BESPoC and HGM analysis of language tasks on resected electrodes is shown. Abbreviations in FIGS. 12-13: PN picture naming, AN auditory naming, SL story listening, NC naturalistic conversation, NPE neuropsychological evaluation, HGM high gamma power modulation, BESPoC Behavior-iEEG-Spectral-Power-Attorney Docket No. 0010872.0813187 PATENTCorrelation, AUC area under receiver operating characteristic curve, Sens sensitivity, Spec specificity, VC Verbal Comprehension, VS Visual Spatial, WM Working Memory, PS Processing Speed, LWI Letter-Word Identification, Calc Calculation, Spel Spelling, PC Passage Comprehension, SM Story Memory, VL Verbal Learning, I Immediate, D Delayed, R Recognition, BNT Boston Naming Test, VMI Visual Motor Integration, PPVT Peabody Picture Vocabulary Test.2.5 Naturalistic Conversation

[0086] HGM and ESM are incompatible with NC, hence, head-to-head comparisons between the modalities were not possible. BESPoC analysis of naturalistic conversation (BESPoC-NC) localized significant HGM sites with good AUC for all language tasks, including SL (0.74), PN (0.71), and AN (0.70) (FIGS. 5-6). When BESPoC and HGM were compared using different tasks (“across tasks” in FIGS. 5-6), NC was never worse and sometimes significantly better than these models (e.g. BESPoC-SL compared to HGM-PN or BESPoC-PN compared to HGL-SL), suggesting that NC maps language more completely than either task. BESPoC-NC was also a good determinant of the magnitude of HGM during structured tasks similar to other across task comparisons of BESPoC and HGM.

[0087] NC also well localized ESM naming sites by itself (AUC 0.72, 95% CI 0.67-0.76) and when combined with SL (AUC 0.77, 95% CI 0.71-0.84) or PN (AUC 0.73, 95% CI 0.68-0.77, FIGS. 4 and 8). Although BESPoC-PN+SL and BESPoC-SL were significantly better for localizing ESM, NC overlapped with the 95% Cis of all other models.

[0088] BESPoC-NC was significantly associated with 3 NPEs (which survived Bonferroni correction): working memory and processing speed (Wechsler) and passage comprehension (Woodcock-Johnson), however, significant F-statistics were seen for 9 / 17 NPEs (FIGS. 10-11). BESPoC-NC also modeled clinically relevant decline (RCI< -1.96) in Passage Comprehension, Immediate and Delayed subtests of Verbal Learning, and Visual Motor Integration, with the lower 95% CI of AUC>0.70 (Tables 3, S7). Combinations of NC with other language tasks produced high-quality models across nearly all (14 / 17) NPEs.Attorney Docket No. 0010872.0813187 PATENT

[0089] FIG. 14 is a table showing significant differences in cortical activation patterns with naturalistic conversation and conventional tasks, where f indicates ignificant difference (Bonferroni corrected), and * indicates significant difference (uncorrected), as determined by Fisher's exact test. Only parcels with > 25 channels for at least one task are shown. Percentage of significantly correlated channels (per BESPoC analysis) within MICCAI parcels compared to expressive and receptive language components of conventional tasks. FIG. 15 is another table showing differences in cortical activation patterns with naturalistic conversation and conventional tasks.

[0090] Cortical parcels showing preferential activation during NC rather than structured tasks (FIG. 14-15) were also analyzed. For both expressive and receptive language across PN and AN / SE, NC had significantly less activation in left central operculum. On comparing NC and PN, many of the parcels preferentially active during PN were probably involved in visual processing, including bilateral precuneus, left inferior occipital gyrus, left fusiform gyrus, and left lingual gyrus. Similarly, patient microphone during AN showed significantly more activation in bilateral middle temporal gyri and bilateral pars triangularis compared to NC. In comparison, receptive processing during NC was associated with significantly higher activation in left frontal lobe: left middle frontal gyrus when compared to SE and left superior frontal gyrus when compared to AN. Also, left superior and middle temporal gyri and right angular gyrus were more active during receptive processing in NC (partner microphone) than AN (audio prompt).3.0 Discussion

[0091] BESPoC is a novel approach for task-free mapping of neural substrates of naturalistic behavior. BESPoC was validated against recognized intracranial functional mapping methods including electrical stimulation mapping (ESM) and high-gamma modulation (HGM), and against postsurgical neuropsychological outcomes, using well-established tasks including picture naming (PN), auditory naming (AN), and story listening (SL). BESPoC was then applied to define the cortical substrates of unconstrained, naturalistic conversation (NC) between patient(s) and family member(s).

[0092] BESPoC produced highly specific models of HGM language sites (FIGS. 5-6) and compared well with HGM for localizing ESM naming sites (FIGS. 4 and 8-9). BESPoC wasAttorney Docket No. 0010872.0813187 PATENT superior to HGM for modeling neuropsychological outcomes across several cognitive domains, identifying deficits observed despite preserving ESM naming sites during epilepsy surgery (FIGS. 10 andl2-13). FIG. 7 shows locations of sites showing significant BESPoC and HGM activations on standardized brain. BESPoC analysis of NC compared favorably with conventional tasks for cerebral localization (FIGS. 7 and 14-15). Moreover, adding BESPoC analysis of NC to conventional tasks improved models of postsurgical neuropsychological outcomes across many domains (FIGS. 10 and 12-13).3.1 BESPoC: Defining Neural Substrates of Naturalistic Conversation

[0093] Cortical activation from BESPoC analysis of NC compared well with that from conventional language tasks analyzed using BESPoC itself or HGM. NC engaged both receptive and expressive language, and relied on verbal working memory, identifying a more comprehensive cortical language map compared to conventional tasks (20. 21, 27). Statistical models of association between neuropsychological outcomes and BESPoC-NC consistently performed well with AUC >0.70 for 7 / 17 measures, and lower 95% CI of AUC >0.70 for 4 / 17 (FIGS. 12-13).

[0094] Comparisons of parcel- wise activations showed a more focused activation pattern withNC compared to expressive language processing during conventional tasks (FIGS. 14-15). Compared to PN, left precentral and anterior cingulate gyri showed higher activation during NC, suggesting higher integration between cognition and vocalization required for conversation than naming pictures (37). Not surprisingly, parcels showing significantly higher activation with PN than NC were involved in visual processing. Further, left hippocampus was significantly more active during AN than NC, suggesting higher reliance on long-term memory for answering unfamiliar auditory cues than conversing about personal topics. For receptive language processing, the trend for more focused activation with NC was not so clear. With receptive processing during NC, significantly higher activation was seen in left middle frontal gyrus and right frontal operculum compared to SL, and in left superior temporal, middle temporal, and superior frontal gyri compared to AN. This suggests broader involvement of association cortices involved in contextual interpretation, insight, judgement, and executive functioning for receptive processing during a conversational dialogue compared to artificial tasks. Interestingly, left central operculum was less active during NC for both expressive and receptive processing compared to all other tasks,Attorney Docket No. 0010872.0813187 PATENT leading us to hypothesize that frontoparietal opercular cortex probably monitors performance during structured tasks, which is less required during NC (38).4.0 BESPoC: A Novel Modality for Presurgical Language Mapping

[0095] Despite several shortcomings, ESM remains the standard-of-care for presurgical language mapping (1, 34). HGM mapping is quicker, safer, and affords improved protection against postsurgical neuropsychological deficits (15-18. 39). However, both rely on repetitive trialbased testing; ESM for sequentially testing contact pairs at different settings and using different tasks and HGM to boost signal-to-noise ratio (9, 40, 41). Hence, many patients, including children and those with developmental or behavioral impairments, cannot undergo language mapping, rendering them more vulnerable to surgically imposed deficits. BESPoC shares the strengths of HGM, including simultaneous data acquisition from entire implanted array and lack of risks associated with electrical stimulation. More importantly, it does not require trial-based testing or the need to epoch SEEG data, making it uniquely suitable for studying neural basis of naturalistic behaviors. In the study, BESPoC localized HGM language sites with high specificity (FIGS. 5-6). BESPoC was better at modeling HGM magnitudes of SL and AN than PN, probably due to longer task windows with 8s of story-noise cycle (SL) and 3-5s of auditory cues (AN) compared to 3s of picture display (PN) and Is window for neural activation including lexical / semantic processing, speech motor planning, and vocalization42. BESPoC was comparable to HGM for modeling ESM naming sites. Because ESM probably does not represent ground truth for language mapping, the conclusions from ESM localizations should focus on relative performance of BESPoC and HGM, rather than their individual accuracy.

[0096] BESPoC outperformed HGM for modeling many neuropsychological outcomes, noted despite sparing ESM naming sites (FIG. 10). More importantly, BESPoC compared to HGM, was better able to model clinically relevant declines across neuropsychological domains (RCIs> -1.96), especially with task combinations (FIGS. 12-13). We acknowledge that the cohort with neuropsychological data was smaller, however, all neuropsychological measures (except one) were represented by over 180 electrode contacts localized to surgically lesioned tissue (Table 3). Hence, with the possible exception of the Woodcock-Johnson Spelling subtest, our results are reliable.Attorney Docket No. 0010872.0813187 PATENT

[0097] Furthermore, the analysis does not reflect how well BESPoC or HGM might independently protect against neuropsychological deficits; rather what ESM has potentially missed, because no ESM naming sites were lesioned in the patients.4.1 Additional Example Neuroscientific Applications

[0098] BESPoC lends itself to analysis of naturalistic behaviors such as reading, playing video games, and watching television. These behaviors can generate opportunities for functional mapping in patients unable to perform structured tasks. Audiovisual recordings, synchronized to SEEG, can allow functional maps to be generated without entering patients’room. Patients who do not speak the primary language of the medical team, including sign language; communication with family members, in whatever form it may take, can still produce valid language maps Thus, while the examples of BESPoC provided herein focus on using BESPoC as a clinical functional mapping modality, this disclosure is not limited, as however, it has immense potential for neuroscientific applications. Further, correlation-based analysis of frequency-specific connectivity measures, broad-band power analysis, and cross-frequency coupling can provide insights into neural network activity at different time-frequency scales (43, 44).

[0099] In accordance with various embodiments, the BESPoC methodology provided herein can enable task-free presurgical language mapping through analysis of naturalistic conversation extracted from routine video-EEG recordings during o intracranial EEG monitoring o. This embodiment may eliminate the need for structured language tasks or dedicated recording equipment by utilizing spontaneous conversational interactions that occur naturally during clinical monitoring periods. The approach may leverage automated conversation detection and speaker diarization techniques to identify and separate patient and conversation partner speech from the mixed audio captured during video-EEG monitoring sessions.

[0100] The task-free implementation may provide particular advantages for patient populations who may have difficulty participating in conventional structured language testing, including pediatric patients, patients with cognitive impairments, or patients experiencing postoperative limitations. By analyzing naturally occurring conversational behaviors, the methodology may capture language processing patterns in their ecological context while maintaining the analytical rigor necessary for presurgical planning. The following figures demonstrate theAttorney Docket No. 0010872.0813187 PATENT validation and performance characteristics of this task-free approach across multiple patients and recording conditions.

[0101] The patient speech analysis may show temporal correspondence between the patient microphone channel, and the diarization results from video-EEG audio extraction. The binary timeseries may represent periods of speech activity as elevated signal levels and periods of silence as baseline signal levels, creating a temporal map of when the patient engages in vocalization during the recording session. The diarization results may demonstrate substantial agreement with the conventional microphone recordings, indicating that automated speaker separation techniques may accurately identify periods of patient speech activity from the mixed audio signal captured during video-EEG monitoring.

[0102] The parent speech analysis may reveal similar temporal correspondence between the parent microphone channel and the diarization results from video-EEG audio extraction. The parent speech patterns may show distinct temporal characteristics compared to patient speech, with different durations and timing of vocalization periods that reflect the conversational dynamics between the patient and family member. The diarization algorithm may successfully distinguish between patient and parent speech contributions, enabling separate analysis of each speaker's vocal activity patterns.

[0103] The visual inspection of the temporal alignment may reveal that while the diarization results may not achieve perfect correspondence with the conventional microphone recordings, the automated approach may recreate the microphone activity patterns with high accuracy. The temporal discrepancies between the two approaches may be minimal, with the majority of speech onset and offset times showing close agreement between the conventional microphone channels and the diarization-derived speech activity markers.

[0104] The high accuracy of the diarization approach indicates that video-EEG recordings can contain sufficient audio quality and temporal resolution to support automated extraction of speech activity patterns for BESPoC analysis. Thus, the successful recreation of microphone activity patterns can provide and preserve the temporal precision required for correlation analysis between neural activity and behavioral markers when using diarization techniques applied to video-EEG audio streams.Attorney Docket No. 0010872.0813187 PATENT

[0105] The validation results demonstrates that diarization software can effectively separate overlapping speech signals and ambient noise from the mixed audio captured during video-EEG monitoring. The ability to distinguish between patient and parent speech contributions can enable independent analysis of expressive and receptive language processing during naturalistic conversation, similar to the analysis capabilities provided by dedicated microphone recording systems.

[0106] The temporal fidelity of the diarization results may support the implementation of BESPoC methodology using audio extracted from routine video-EEG monitoring without requiring additional recording equipment or patient cooperation for structured task paradigms. The preservation of speech timing information may enable correlation analysis between neural activity patterns and naturalistic speech behaviors that occur spontaneously during clinical monitoring periods.

[0107] Referring now to FIG. 19, an example graph comparing BESPoC analysis results from two different audio sources for patient speech and parent speech is depicted. The diarization validation results for the example patient (Patient 1) demonstrates the feasibility of extracting speech activity patterns from video-EEG recordings using automated speaker diarization techniques. The analysis may compare binary speaking and not-speaking timeseries obtained from dedicated microphone recordings during conventional task recording against diarization results derived from audio extracted from video-EEG recordings. The comparison may provide validation of the accuracy with which diarization software can recreate the temporal patterns of speech activity that are captured through conventional microphone-based recording systems.

[0108] Referring to FIG. 20, the diarization validation results for Patient 2 may demonstrate similar performance characteristics as Patient 1, providing consistent validation of automated speaker separation techniques when applied to video-EEG audio recordings. Like the analysis shown in FIG. 19, the comparison for Patient 2 may evaluate the temporal patterns of speech activity obtained through conventional microphone-based recording systems against the results produced by diarization software processing audio extracted from video-EEG monitoring sessions. The comparison may provide additional validation of the reliability and accuracy of diarizationAttorney Docket No. 0010872.0813187 PATENT approaches for identifying distinct speaker contributions during naturalistic conversation recordings.

[0109] The patient speech activity analysis for both patients show temporal alignment between the patient microphone channel from the conventional task recording and the diarization results derived from video-EEG audio extraction. The binary timeseries representation displays periods of patient vocalization as distinct temporal segments separated by intervals of silence, creating a detailed map of when the patient engages in speech production during the conversational interaction. The diarization algorithm may therefore successfully identify the majority of patient speech episodes across both patients, demonstrating the ability to distinguish patient vocalizations from other audio components present in the mixed signal captured during video-EEG monitoring.

[0110] The parent speech activity patterns for both patients may exhibit similar temporal correspondence between the conventional parent microphone channel and the diarization-derived speech activity markers. The parent vocalization periods may show distinct temporal characteristics that complement the patient speech patterns, reflecting the turn-taking dynamics and conversational flow that characterize naturalistic dialogue between the patient and family member. The diarization software may effectively separate parent speech contributions from the combined audio signal for both patients, enabling independent tracking of each speaker's vocal activity throughout the recording session.

[0111] The visual inspection of the temporal alignment for both Patients confirms that the diarization approach recreates the speech activity patterns with substantial accuracy when compared to the conventional microphone recordings. The automated speaker separation technique can capture the onset and offset timing of speech episodes with precision sufficient to support correlation analysis between neural activity and behavioral markers across different individuals. The temporal fidelity of the diarization results indicates that the approach preserves the timing information necessary for BESPoC methodology implementation.

[0112] The consistency of diarization performance across both Patient 1 and Patient 2, as demonstrated in FIGS. 19 and 20 respectively, demonstrates the robustness of automated speaker separation techniques when applied to video-EEG audio recordings from different individuals. The successful validation across multiple patients suggests that the diarization approach can beAttorney Docket No. 0010872.0813187 PATENT generalized across diverse patient populations and recording conditions, supporting the broader applicability of the technique for extracting speech activity patterns from routine clinical monitoring sessions.

[0113] The accuracy of speaker identification and temporal segmentation demonstrated across both Patients may further support the feasibility of implementing task-free presurgical language mapping using audio extracted from video-EEG recordings. The preservation of speech timing information through the diarization process can therefore enable correlation analysis between neural activity patterns and naturalistic conversational behaviors without requiring dedicated recording equipment or structured task paradigms that may be challenging for certain patient populations.

[0114] Referring to FIG. 21, the BESPoC correlation analysis results may demonstrate strong relationships between conventional task microphone recordings and audio extracted from video- EEG recordings for Patient 1. The scatter plot analysis compares log-transformed p-values multiplied by direction of effect obtained from BESPoC analysis using task microphone recordings against corresponding values obtained from BESPoC analysis using audio extracted from video- EEG recordings. The linear regression analysis reveals distinct correlation patterns for patient speech and parent speech components of the naturalistic conversation.

[0115] The patient speech correlation analysis may achieve an R2value of 0.1393, indicating that approximately 13.93% of the variance in BESPoC analysis results from video-EEG audio extraction may be explained by the corresponding results from conventional task microphone recordings. The positive linear relationship demonstrates that the two analytical approaches produce consistent identification of neural activity patterns associated with patient speech production during naturalistic conversation. The correlation strength suggests that automated diarization of video-EEG audio preserves the temporal and spectral characteristics necessary for accurate BESPoC analysis of expressive language processing.

[0116] The parent speech correlation analysis yielded an R2value of 0.0074, representing approximately 0.74% of explained variance between the two analytical approaches. The lower correlation value for parent speech reflects the challenges associated with capturing and processing audio from speakers who may be positioned at varying distances from the video-EEG recordingAttorney Docket No. 0010872.0813187 PATENT equipment. The positive linear relationship nevertheless indicates that diarization techniques identifies parent speech contributions with sufficient accuracy to support correlation analysis between neural activity and receptive language processing during conversational interactions.

[0117] Referring to FIG. 22, the BESPoC correlation analysis results for Patient 2 demonstrates enhanced performance compared to Patient 1, particularly for parent speech processing. The scatter plot visualization displays the relationship between log-transformed p- values multiplied by direction of effect from conventional task microphone recordings and corresponding values from video-EEG audio extraction, revealing improved correlation patterns across both patient and parent speech components.

[0118] The patient speech correlation analysis for Patient 2 achieved an R2value of 0.1807, indicating that approximately 18.07% of the variance in video-EEG-derived BESPoC results can be explained by conventional microphone-based analysis. The enhanced correlation strength compared to Patient 1 reflects individual differences in speech characteristics, recording conditions, or diarization algorithm performance. The positive linear relationship provides additional validation that automated speaker separation techniques preserves the neural-behavioral correlation patterns detected through conventional recording approaches.

[0119] The parent speech correlation analysis for Patient 2 demonstrates substantially improved performance with an R2value of 0.0463, representing approximately 4.63% of explained variance between the two analytical methods. The six-fold improvement in correlation strength compared to Patient 1 parent speech analysis indicates that recording conditions, speaker positioning, or audio quality factors influence the accuracy of diarization-based speech extraction. The positive linear relationship suggests that parent speech contributions can be successfully identified and analyzed using video-EEG audio extraction when optimal recording conditions are present.

[0120] The positive linear trendlines observed across both patients and both speech components also indicate strong relationships between conventional microphone-based BESPoC analysis and video-EEG audio extraction approaches. The consistent positive correlations indicates that task-free language mapping using patient and parent speech patterns extracted from video-EEG recordings are feasible for clinical implementation. The con-elation patternsAttorney Docket No. 0010872.0813187 PATENT demonstrate that automated diarization techniques preserve the temporal precision and spectral characteristics required for accurate neural-behavioral correlation analysis.

[0121] The task-free presurgical language mapping capability provided herein can enabled through the combination of automated conversation detection and speaker diarization applied to routine video-EEG monitoring recordings. The system can identify instances of naturalistic conversation occurring spontaneously during clinical monitoring periods without requiring structured task presentations or patient cooperation with artificial language paradigms. The diarization process may partition the mixed audio signal into separate speaker channels, enabling independent analysis of patient and conversation partner speech contributions.

[0122] The conversation detection component may utilize automated algorithms to identify periods of vocal activity within the continuous video-EEG audio stream, distinguishing conversational speech from other audio components such as medical equipment sounds, ambient noise, or non-conversational vocalizations. The detection algorithms may analyze acoustic features such as spectral characteristics, temporal patterns, and amplitude variations to identify segments of the recording that contain naturalistic conversational interactions suitable for BESPoC analysis.

[0123] The speaker diarization process may employ machine learning algorithms trained to distinguish between different speakers based on acoustic characteristics such as fundamental frequency, spectral envelope, and temporal speech patterns. The diarization algorithms may assign speech segments to individual speakers, creating separate audio channels for the patient and conversation partners that may be analyzed independently using BESPoC methodology. The automated speaker separation may enable analysis of both expressive language processing during patient speech and receptive language processing during partner speech without requiring manual audio segmentation or dedicated recording equipment.

[0124] 4.2 Conclusions

[0125] Naturalistic conversation is an ecologically valid cognitive task for presurgical localization of language function. It provides both expressive and receptive language maps comparable to conventional tasks and models neuropsychological deficits. It is more suitable forAttorney Docket No. 0010872.0813187 PATENT patients with behavioral or developmental limitations and can be performed at practically any age, even in patients with limited vocabulary.

[0126] BESPoC is a novel method for identifying significant changes in spectral power in intracranial signals, correlated to specified behaviors. It does not require parsing or epoching SEEG data for structured trials and is therefore eminently suitable to study naturistic behaviors, as shown here for unconstrained conversations. BESPoC uses only the speech envelope, not the content, allowing brain mapping to be performed without compromising patient privacy. BESPoC analysis of conventional language tasks can identify ESM and HGM language sites and model diverse neuropsychological outcomes. The simplicity and flexibility of BESPoC analysis allow it to be widely applied to a variety of naturalistic behaviors offering new insights into their neural substrates.REFERENCES

[0127] 1. Jayakar P, Gotman J, Harvey AS, Palmini A, Tassi L, Schomer D, et al. Diagnostic utility of invasive EEG for epilepsy surgery: Indications, modalities, and techniques. Epilepsia. 2016; 57(11):1735 47.

[0128] 2. Aungaroon G, Zea Vera A, Hom PS, Byars AW, Greiner HM, Tenney JR, et al.After-discharges and seizures during pediatric extra- operative electrical cortical stimulation functional brain mapping: Incidence, thresholds, and determinants. Clin Neurophysiol. 2017; 128(10):2078— 86.

[0129] 3. Corley JA, Nazari P, Rossi VJ, Kim NC, Fogg LF, Hoeppner TJ, et al. Cortical stimulation parameters for functional mapping. Seizure. 2017; 45:36-41.

[0130] 4. Hamberger MJ. Cortical language mapping in epilepsy: a critical review.Neuropsychol Rev. 2007; 17(4):477-89.

[0131] 5. Ojemann GA. Individual variability in cortical localization of language. JNeurosurg. 1979; 50(2): 164-9.Attorney Docket No. 0010872.0813187 PATENT

[0132] 6. Sakpichaisakul K, Byars AW, Horn PS, Aungaroon G, Greiner HM, Mangano FT, et al. Neuropsychological outcomes after pediatric epilepsy surgery: Role of electrical stimulation language mapping. Seizure. 2020; 80:183-91.

[0133] 7. Hermann BP, Perrine K, Chelune GJ, Barr W, Loring DW, Strauss E, et al. Visual confrontation naming following left anterior temporal lobectomy: a comparison of surgical approaches. Neuropsychology. 1999; 13(1):3— 9.

[0134] 8. Hyslop A, Duchowny M. Electrical stimulation mapping in children. Seizure. 2020;77:59-63.

[0135] 9. Arya R, Hom PS, Crone NE. ECoG high-gamma modulation versus electrical stimulation for presurgical language mapping. Epilepsy Behav. 2018; 79:26-33.

[0136] 10. Sinai A, Bowers CW, Crainiceanu CM, Boatman D, Gordon B, Lesser RP, et al. Electrocorticographic high gamma activity versus electrical cortical stimulation mapping of naming. Brain. 2005; 128(Pt 7): 1556-70.

[0137] 11. Ray S, Crone NE, Niebur E, Franaszczuk PJ, Hsiao SS. Neural correlates of high-gamma oscillations (60-200 Hz) in macaque local field potentials and their potential implications in electrocorticography. J Neurosci. 2008; 28(45): 11526-36.

[0138] 12. Lachaux J-P, Fonlupt P, Kahane P, Minotti L, Hoffmann D, Bertrand O, et al. Relationship between task-related gamma oscillations and BOLD signal: new insights from combined fMRI and intracranial EEG. Hum Brain Mapp. 2007; 28(12): 1368-75.

[0139] 13. Nishida M, Juhasz C, Sood S, Chugani HT, Asano E. Cortical glucose metabolism positively correlates with gamma-oscillations in nonlesional focal epilepsy. Neuroimage. 2008; 42(4): 1275-84.

[0140] 14. Ervin B, Buroker J, Rozhkov L, Holloway T, Horn PS, Scholle C, et al.High-gamma modulation language mapping with stereo-EEG: A novel analytic approach and diagnostic validation. Clin Neurophysiol. 2020; 131(12):2851- 60.Attorney Docket No. 0010872.0813187 PATENT

[0141] 15. Arya R, Roth C, Leach JL, Middeler D, Wilson JA, Vannest J, et al.Neuropsychological outcomes after resection of cortical sites with visual naming associated electrocorticographic high-gamma modulation. Epilepsy Res. 2019; 151:17-23.

[0142] 16. Sonoda M, Rothermel R, Carlson A, Jeong J-W, Lee M-H, Hayashi T, et al.Naming-related spectral responses predict neuropsychological outcome after epilepsy surgery. Brain. 2022; 145(2):517-30.

[0143] 17. Ervin B, Kargol C, Byars AW, Buroker J, Rozhkov L, Skoch J, et al. High- gamma modulation language mapping and cognitive outcomes after epilepsy surgery. Epilepsia. 2024; 65(10):3052-63.

[0144] 18. Kojima K, Brown EC, Matsuzaki N, Rothermel R, Fuerst D, Shah A, et al.Gamma activity modulated by picture and auditory naming tasks: intracranial recording in patients with focal epilepsy. Clin Neurophysiol. 2013; 124(9); 1737-44.

[0145] 19. Arya R, Wilson JA, Fujiwara H, Vannest J, Byars AW, Rozhkov L, et al.Electrocorticographic high-gamma modulation with passive listening paradigm for pediatric extraoperative language mapping. Epilepsia. 2018; 59(4):792- 801.

[0146] 20. Arya R, Wilson JA, Vannest J, Byars AW, Greiner HM, Buroker J, et al.Electrocorticographic language mapping in children by high-gamma synchronization during spontaneous conversation: comparison with conventional electrical cortical stimulation. Epilepsy Res. 2015; 110:78-87.

[0147] 21. Towle VL, Yoon H-A, Castelie M, Edgar JC, Biassou NM, Frim DM, et al.ECoG gamma activity during a language task: differentiating expressive and receptive speech areas. Brain. 2008; 13 l(Pt 8):2013-27.

[0148] 22. Hmamouche Y, Ochs M, Prevot L, Chaminade T. Interpretable prediction of brain activity during conversations from multimodal behavioral signals. PLoS One. 2024; 19(3):e0284342.Attorney Docket No. 0010872.0813187 PATENT

[0149] 23. Hidalgo C, Mohamed I, Zielinski C, Schon D. The effect of speech degradation on the ability to track and predict turn structure in conversation. Cortex. 2022; 151:105-15.

[0150] 24. Nieuwland MS. Do “early” brain responses reveal word form prediction during language comprehension? A critical review. Neurosci Biobehav Rev. 2019; 96:367-400.

[0151] 25. Bogels S, Magyari L, Levinson SC. Neural signatures of response planning occur midway through an incoming question in conversation. Sci Rep. 2015; 5:12881.

[0152] 26. Levinson SC. Turn-taking in Human Communication-Origins andImplications for Language Processing. Trends Cogn Sci. 2016; 20(l):6-14.

[0153] 27. Bogels S. Neural correlates of turn-taking in the wild: Response planning starts early in free interviews. Cognition. 2020; 203:104347.

[0154] 28. Bauer PR, Vansteensel MJ, Bleichner MG, Hermes D, Femer CH,Aamoutse EJ, et al. Mismatch between electrocortical stimulation and electrocorticography frequency mapping of language. Brain Stimul. 2013; 6(4):524-31.

[0155] 29. Derix J, Iljina O, Schulze-Bonhage A, Aertsen A, Ball T. “Doctor” or“darling”? Decoding the communication partner from ECoG of the anterior temporal lobe during non-experimental, real-life social interaction. Front Hum Neurosci. 2012; 6:251.

[0156] 30. Derix J, Iljina O, Weiske J, Schulze-Bonhage A, Aertsen A, Ball T. From speech to thought: the neuronal basis of cognitive units in non-experimental, real-life communication investigated using ECoG. Front Hum Neurosci. 2014; 8:383.

[0157] 31. Snodgrass JG, Vanderwart M. A standardized set of 260 pictures: norms for name agreement, image agreement, familiarity, and visual complexity. J Exp Psychol Hum Learn. 1980; 6(2): 174— 215.

[0158] 32. Hamberger MJ, Seidel WT. Auditory and visual naming tests: normative and patient data for accuracy, response time, and tip-of-the-tongue. J Int Neuropsychol Soc. 2003; 9(3):479-89.Attorney Docket No. 0010872.0813187 PATENT

[0159] 33. Arya R, Ervin B, Dudley J, Buroker J, Rozhkov L, Scholle C, et al.Electrical stimulation mapping of language with stereo-EEG. Epilepsy Behav. 2019; 99:106395.

[0160] 34. Arya R, Baumer FM, Chauvel P, Frauscher B, Jayakar P, Kheder A, et al.American Clinical Neurophysiology Society Technical Standards for Electrical Stimulation With Intracranial Electrodes for Functional Brain Mapping and Seizure Induction. J Clin Neurophysiol. 2025; 42(3): 190-200.

[0161] 35. Ervin B, Rozhkov L, Buroker J, Leach JL, Mangano FT, Greiner HM, et al.Fast Automated Stereo-EEG Electrode Contact Identification and Labeling Ensemble. Stereotact Funct Neurosurg. 2021; 99(5):393-404.

[0162] 36. Politis DN, Romano JP. Large sample confidence regions based on subsamples under minimal assumptions. The Annals of Statistics. 1994; :2031-50.

[0163] 37. Oane I, Barborica A, Mindruta IR. Cingulate Cortex: Anatomy, Structural and Functional Connectivity. J Clin Neurophysiol. 2023; 40(6):482-90.

[0164] 38. Quirmbach F, Limanowski J. A Crucial Role of the Frontal Operculum inTask-Set Dependent Visuomotor Performance Monitoring. eNeuro. 2022; 9(2):ENEURO.0524- 21.2021.

[0165] 39. Babajani-Feremi A, Holder CM, Narayana S, Fulton SP, Choudhri AF,Boop FA, et al. Predicting postoperative language outcome using presurgical fMRI, MEG, TMS, and high gamma ECoG. Clin Neurophysiol. 2018; 129(3):560-71.

[0166] 40. Lachaux J-P, Axmacher N, Mormann F, Halgren E, Crone NE. High- frequency neural activity and human cognition: past, present and possible future of intracranial EEG research. Prog Neurobiol. 2012; 98(3):279-301.

[0167] 41. Kambara T, Sood S, Alqatan Z, Klingert C, Ratnam D, Hayakawa A, et al.Presurgical language mapping using event-related high-gamma activity: The Detroit procedure. Clin Neurophysiol. 2018; 129(1): 145-54.Attorney Docket No. 0010872.0813187 PATENT

[0168] 42. Ervin B, Buroker J, Byars AW, Rozhkov L, Leach JL, Hom PS, et al. A distributed network supports spatiotemporal cerebral dynamics of visual naming. Clin Neurophysiol. 2021; 132(12):2948— 58.

[0169] 43. Luhmann HJ, Sinning A, Yang J-W, Reyes-Puerta V, Stuttgen MC,Kirischuk S, et al. Spontaneous Neuronal Activity in Developing Neocortical Networks: From Single Cells to Large-Scale Interactions. Front Neural Circuits. 2016; 10:40.

[0170] 44. Higley MJ, Cardin JA. Spatiotemporal dynamics in large-scale cortical networks. Curr Opin Neurobiol. 2022; 77:102627.

[0171] 45. Wechsler, D. (2012). Wechsler Preschool and Primary Scale of IntelligenceFourth Edition (WPPSI-IV). San Antonio, TX: Pearson.

[0172] 46. Wechsler, D. (2014). Wechsler Intelligence Scale for Children Fifth Edition(WISC-V). San Antonio, TX: Pearson.

[0173] 47. Wechsler, D. (2008). Wechsler Adult Intelligence Scale Fourth Edition(WAIS-IV). San Antonio. TX: Pearson.

[0174] 48. Schrank, F. A., Mather, N., & McGrew, K. S. (2014). Woodcock-JohnsonIV Tests of Achievement. Rolling Meadows, IL: Riverside.

[0175] 49. Sheslow, D., & Adams, W. (2003). Wide Range Assessment of Memory and Learning-Second Edition (WRAML-2). Wilmington, DE: Wide Range, Inc.

[0176] 50. Dunn, L. M., & Dunn, D. M. (2007). Peabody Picture Vocabulary Test-Fourth Edition (PPVT-4). Bloomington, MN: NCS Pearson.

[0177] 51. Dunn, D. M. (2019). Peabody Picture Vocabulary Test-Fifth Edition(PPVT-5). Bloomington, MN: NCS Pearson.

[0178] 52. Beery, K. E., & Beery, N. A. (2010). The Beery-Buktenica DevelopmentalTest of Visual-Motor Integration: Administration, Scoring, and Teaching Manual (6th ed.). Minneapolis, MN: NCS Pearson.Attorney Docket No. 0010872.0813187 PATENT

[0179] 53. Kaplan, E., Goodglass, H., & Weintraub, S. (1983). Boston Naming Test.Philadelphia, PA: Lea & Febiger.

[0180] All percentages and ratios are calculated by weight unless otherwise indicated.

[0181] All percentages and ratios are calculated based on the total composition unless otherwise indicated.

[0182] It should be understood that every maximum numerical limitation given throughout this specification includes every lower numerical limitation, as if such lower numerical limitations were expressly written herein. Every minimum numerical limitation given throughout this specification will include every higher numerical limitation, as if such higher numerical limitations were expressly written herein. Every numerical range given throughout this specification will include every narrower numerical range that falls within such broader numerical range, as if such narrower numerical ranges were all expressly written herein.

[0183] The dimensions and values disclosed herein are not to be understood as being strictly limited to the exact numerical values recited. Instead, unless otherwise specified, each such dimension is intended to mean both the recited value and a functionally equivalent range surrounding that value. For example, a dimension disclosed as “20 mm” is intended to mean “about 20 mm.”

[0184] Every document cited herein, including any cross referenced or related patent or application, is hereby incorporated herein by reference in its entirety unless expressly excluded or otherwise limited. All accessioned information (e.g., as identified by PUBMED, PUBCHEM, NCBI, UNIPROT, or EBI accession numbers) and publications in their entireties are incorporated into this disclosure by reference in order to more fully describe the state of the art as known to those skilled therein as of the date of this disclosure. The citation of any document is not an admission that it is prior art with respect to any invention disclosed or claimed herein or that it alone, or in any combination with any other reference or references, teaches, suggests or discloses any such invention. Further, to the extent that any meaning or definition of a term in this document conflicts with any meaning or definition of the same term in a document incorporated by reference, the meaning or definition assigned to that term in this document shall govern.Attorney Docket No. 0010872.0813187 PATENT

[0185] While particular embodiments of the present invention have been illustrated and described, it would be obvious to those skilled in the art that various other changes and modifications may be made without departing from the spirit and scope of the invention. It is therefore intended to cover in the appended claims all such changes and modifications that are within the scope of this invention.

Claims

Attorney Docket No. 0010872.0813187 PATENTCLAIMS1. A system for cognitive function mapping based on naturalistic behaviors, comprising: a plurality of intracranial electrodes configured to detect electrical activity of a brain of a subject; a monitoring device configured to record naturalistic behaviors of the subject; one or more processors and a non-transitory computer- readable storage medium storing instructions that, when executed by the one or more processors, cause the system to: receive intracranial electroencephalography (iEEG) data from the plurality of intracranial electrodes during naturalistic behaviors of the subject: receive behavior data from the monitoring device corresponding to the naturalistic behaviors; synchronize the iEEG data with the behavior data; perform Time-Frequency Representation (TFR) calculations on the iEEG data to compute spectral power across multiple frequency bins as a function of time: calculate correlation coefficients between the computed spectral power across the multiple frequency bins and the behavior data; and generate quantitative markers indicating statistical significance of correlation between spectral power and behavior.2.A system for cognitive function mapping based on naturalistic behaviors according to any of the preceding claims, wherein the Time-Frequency Representation (TFR) calculations comprise a wavelet-transformation of the iEEG data.

3. A system for cognitive function mapping based on naturalistic behaviors according to any of the preceding claims, wherein generating quantitative markers comprises converting the correlation coefficients to Z-scores and corresponding p-values.Attorney Docket No. 0010872.0813187 PATENT4. A system for cognitive function mapping based on naturalistic behaviors according to any of the preceding claims, wherein generating quantitative markers comprises performing permutation testing on the correlation coefficients5. A system for cognitive function mapping based on naturalistic behaviors according to any of the preceding claims, wherein the naturalistic behaviors comprise a conversation between the subject and a partner.

6. A system for cognitive function mapping based on naturalistic behaviors according to claim 3, wherein a voice of the subject is recorded on a first microphone channel and a voice of the partner is recorded on a second microphone channel, and wherein the instructions further cause the system to perform conversation analysis on each of the first microphone channel and the second microphone channel.

7. A system for cognitive function mapping based on naturalistic behaviors according to any of the preceding claims, wherein the instructions further cause the system to, based on the correlation between spectral power and behavior, generate a map of predicted sites, wherein the map is generated without trial- averaging repetitive tasks performed by the subject.

8. A method for intracranial language mapping based on naturalistic behaviors, comprising: receiving intracranial electroencephalography (iEEG) data from a plurality of intracranial electrodes implanted in a brain of a subject during naturalistic behaviors; receiving behavior data corresponding to the naturalistic behaviors of the subject; synchronizing the iEEG data with the behavior data; performing Time-Frequency Representation (TFR) calculations on the iEEG data to compute spectral power across multiple frequency bins as a function of time; correlating the computed spectral power across the multiple frequency bins and the behavior data; andAttorney Docket No. 0010872.0813187 PATENT generating a functional language map based on the correlation between spectral power and behavior.

9. A method for intracranial language mapping based on naturalistic behaviors according to claim 8. wherein performing the TFR calculations comprises wavelet- transforming the iEEG data.

10. A method for intracranial language mapping based on naturalistic behaviors according to any of claims 8 to 9, wherein the naturalistic behaviors comprise a conversation between the subject and a partner.

11. A method for intracranial language mapping based on naturalistic behaviors according to any of the preceding claims, further comprising: recording a voice of the subject on a first microphone channel and recording a voice of the partner on a second microphone channel; and analyzing each of the first microphone channel and the second microphone channel.

12. A method for intracranial language mapping based on naturalistic behaviors according to any of claims 8 to 11, wherein the functional language map is generated without trial- averaging repetitive tasks performed by the subject.

13. A system for cognitive mapping based on naturalistic behaviors, comprising: a plurality electrodes configured to detect electrical activity of a brain, wherein each of the plurality of electrodes is any of an intracranial electrode and an extracranial electrode; a monitoring device configured to record naturalistic behaviors of a subject; one or more processors and a non-transitory computer-readable storage medium storing instructions that, when executed by the one or more processors, cause the system to: receive electroencephalography (EEG) data from one or more of the plurality of electrodes during naturalistic behaviors of the subject;Attorney Docket No. 0010872.0813187 PATENT receive behavior data from the monitoring device corresponding to the naturalistic behaviors; synchronize the EEG data with the behavior data; correlate changes in the spectral power across the multiple frequency bins to the behavior data; based on the correlation between spectral power and behavior, generate a map of predicted sites.

14. A system for intracranial language mapping based on naturalistic behaviors according to claim 13, wherein the map is generated without trial-averaging repetitive tasks performed by the subject.

15. A system for intracranial language mapping based on naturalistic behaviors according to any of claims 13 to 14, wherein at least one of the plurality electrodes is a scalp electrode.

16. A system for intracranial language mapping based on naturalistic behaviors according to any of claims 14 to 15, wherein the naturalistic behaviors comprise a conversation by the subject.

17. A system for intracranial language mapping based on naturalistic behaviors according to claim 16, wherein the naturalistic behaviors comprise a spontaneous conversation between the subject and a partner; and wherein a voice of the subject is recorded on a first microphone channel and a voice of the partner is recorded on a second microphone channel.

19. A system for intracranial language mapping based on naturalistic behaviors according to claim 18, wherein the instructions further cause the system to perform conversation analysis on each of the first microphone channel and the second microphone channel.

20. A system for intracranial language mapping based on naturalistic behaviors according to any of claims 13 to 19, wherein the instructions further cause the system to generate quantitative markers indicating statistical significance of correlation between spectral power and behavior; wherein generating quantitative markers comprises converting the correlation coefficients to Z- scores and corresponding p- values.

Images