SeeMe: Using Automated Facial Tracking to Detect Voluntary Behavior in Brain Injury

SeeMe: A Multimodal Behavioral-Electrophysiological Tool for Real-Time Detection of Motor Behavior in Brain Injury Patients

Objective: This prospective interventional study introduces "SeeMe," an automated, high-resolution computer vision platform designed to objectively quantify microscopic, auditory command-evoked movements in patients with Traumatic Brain Injury (TBI). Current clinical assessments, such as the Glasgow Coma Scale (GCS) and Coma Recovery Scale-Revised (CRS-R), rely on subjective human observation and often fail to detect low-amplitude motor responses, potentially misclassifying up to 25% of patients as unresponsive.

Methodology: SeeMe utilizes vector analysis, cross-correlation, and deep neural networks (DNNs) to track individual facial pores and hand movements with sub-millimeter precision (0.5 mm) and high temporal resolution (0.03s). The study will enroll a cohort of 60-80 TBI patients, alongside healthy controls and pharmacologically paralyzed subjects, to validate SeeMe's sensitivity and specificity.

Primary Goals:

1. Validation: Compare SeeMe's detection of voluntary motor recovery against gold-standard clinical examinations (CRS-R).
2. Synchronization: Simultaneously record and time-lock electroencephalography (EEG) and electrocorticography (ECoG) with SeeMe-detected movements.
3. Biomarker Identification: Characterize neural signatures (specifically Beta-band oscillations) associated with the return of voluntary behavior.

Impact: By providing a real-time, objective measure of motor intention and execution, SeeMe aims to identify "Cognitive-Motor Dissociation" (CMD) earlier than current methods, facilitating more accurate prognostications and laying the framework for future closed-loop neuromodulation (e.g., Vagus Nerve Stimulation) to accelerate TBI recovery.

Study Overview

• Status: Recruiting
• Conditions: Severe Traumatic Brain Injury; Traumatic Brain Injury (TBI) Patients
• Intervention / Treatment: Diagnostic test: SeeMe Multimodal Auditory Command Protocol

Detailed Description

1. Study Rationale and Scientific Gap Standard clinical assessments for Traumatic Brain Injury (TBI), such as the Glasgow Coma Scale (GCS) and Coma Recovery Scale-Revised (CRS-R), are limited by human subjectivity, examiner bias, and low spatiotemporal resolution. Recent multicenter studies indicate that up to 25% of patients who appear "unresponsive" at the bedside may exhibit "Cognitive-Motor Dissociation" (CMD)-a state of covert awareness where the brain intends to move, but motor output is too microscopic for the naked eye to detect. Misclassifying these patients as unresponsive negatively impacts clinical outcomes due to the potential withholding of rehabilitative resources. This study validates SeeMe, a multimodal platform designed to bridge this diagnostic gap by objectively quantifying microscopic behavior in real-time.

2. The SeeMe Technological Framework SeeMe is an innovative computer-vision sensing platform that utilizes vector analysis, cross-correlation, and deep neural networks (DNNs) to track individual facial pores and hand landmarks.

   Sub-millimeter Precision: The system offers high spatial precision (0.5 mm) and high temporal resolution (0.03s), allowing it to capture "micro-expressions" and low-amplitude motor responses (e.g., vertical eye movements or subtle mouth twitches) that escape clinical detection.

   Advanced Classification: The investigators utilize a bidirectional long short-term memory (LSTM) network to classify heatmaps of facial movement. This ensures that detected responses are specific to the auditory command provided (e.g., "Show me a smile" vs. "Open your eyes") rather than generalized arousal or non-specific reflexive movements.

3. Multimodal Brain-Behavior Synchronization The primary objective of the R61 phase is the seamless integration of SeeMe with electrophysiological activity (EEG/ECoG). By time-locking microscopic motor initiation with neural oscillations, the investigators aim to identify objective biomarkers of recovery.

   Beta-Band Signatures: The investigators focus on Beta-band Event-Related Desynchronization (ERD) and Synchronization (ERS) as signatures of motor planning and execution.

   One-to-One Mapping: This multimodal approach provides a robust computational framework to synchronize brain activity with behavior, allowing us to map the "bigger picture" of recovery during the critical transition from coma to consciousness.

4. Study Design and Methodology

   This prospective validation study will utilize three distinct cohorts (Total N = 80-100):

   Cohort 1: Healthy Controls (n=10): Establishes the "ground truth" for robust, voluntary motor movements and baseline algorithm performance.

   Cohort 2: Sedated/Anesthetized Controls (n=10): Patients undergoing spine surgery under general anesthesia/pharmacological paralysis provide a "zero-movement" baseline to quantify the algorithm's noise floor and establish specificity.

   Cohort 3: Recovering TBI Patients (n=60-80): The primary clinical cohort.

   Procedures: Participants listen to simple auditory commands (e.g., "Stick out your tongue," "Open your eyes," "Close your hands") alongside a control command ("Today is a sunny day") to distinguish auditory-evoked movements from true command following.

   Comparison: SeeMe data will be collected daily and compared against blinded CRS-R assessments to determine the "detection lead time"-the number of days SeeMe identifies a response before the clinical team notes command following.

5. Clinical and Neuroethical Oversight

   Acknowledging the vulnerability of the TBI population, the study incorporates two novel neuroethical safeguards:

   Patient Advocate: Modeled on pioneering work in organ donation, a dedicated liaison will be appointed to act as a bridge between the research team and families, ensuring participant safety and well-being remain the central focus.

   TBI-Affected Family Council: A consultative council of family members who have experienced TBI will review findings and provide guidance on the ethical communication of results to families.

6. Future Directions The data gathered during this R61 phase-specifically the synchronization of SeeMe and EEG-will define the physiological triggers and milestones for Phase R33. In the subsequent phase, the SeeMe-EEG platform will be integrated with Vagus Nerve Stimulation (VNS). This closed-loop system will test the hypothesis that precisely-timed VNS pulses, paired with SeeMe-detected motor intentions, can facilitate the reorganization of motor circuits and accelerate the recovery of goal-directed behavior after TBI.

• Study Type: Interventional
• Enrollment (Estimated): 80
• Phase: Not Applicable

Contacts and Locations

• Study Contact: Name: Sima Mofakham, PhD; Phone Number: 631-444-1278; Email: sima.mofakham@stonybrookmedicine.edu

Study Locations

• United States
  • New York
    • Stony Brook, New York, United States, 11794
      • Recruiting
      • Stony Brook University Hospital
      • Contact: Sima Mofakham, PhD; Phone Number: 631-444-1278; Email: sima.mofakham@stonybrookmedicine.edu

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: Yes

Description

Group 1: Traumatic Brain Injury (TBI) Cohort

Inclusion Criteria:

• Adults (22+) with a history of acute traumatic brain injury
• Documented loss of consciousness with a Glasgow Coma Scale (GCS) score less than or equal to 8 upon hospitalization
• Clinically stable as determined by the primary neurosurgery or ICU team
• Intact auditory pathways as confirmed by BAERs
• Family consent for study participation

Exclusion Criteria:

• Hearing Impairment confirmed via absence of Brainstem Auditory Evoked Responses (BAERs) that would prevent the patient from hearing the auditory commands
• No legal authorized representative (LAR) available to provide informed consent for the patients in a comatose state
• Any other medical condition that, in the judgment of the investigator, makes participation in the study unsafe.
• Pregnant women
• Any previous history of traumatic brain injury
• Any neurodegenerative disease such as dementia

Group 2: Healthy Control Cohort

Inclusion Criteria:

• Adults 22+ with no history of neurological or psychiatric disorders
• Normal baseline neurological examination
• Intact auditory pathways
• Ability to provide informed consent
• Ability to follow simple auditory commands in English

Exclusion Criteria:

• Hearing Impairment that would prevent the participant from hearing the auditory commands
• Any previous history of severe traumatic brain injury (TBI)
• Any neurodegenerative disease (e.g., dementia)
• Any motor impairment (e.g., facial palsy, carpal tunnel syndrome) that would interfere with facial or hand movement tracking
• Any other medical condition that, in the judgment of the investigator, makes participation in the study unsafe
• Pregnant women

Group 3: Sedated/Anesthetized Cohort

Inclusion Criteria:

• Adults (22+) undergoing elective spine surgery
• Requirement of general anesthesia and pharmacological paralysis (neuromuscular blockade) as part of the standard surgical procedure
• Clinically stable for study procedures as determined by the anesthesia and surgical teams.
• Intact auditory pathways
• Ability to provide informed pre-operative consent

Exclusion Criteria:

• Hearing Impairment that would prevent the patient from hearing the auditory commands
• Any previous history of severe traumatic brain injury
• Any neurodegenerative disease such as dementia
• Significant baseline facial or hand motor deficits prior to the administration of anesthesia
• Any other medical condition that, in the judgment of the investigator, makes participation in the study unsafe
• Pregnant women

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Diagnostic
• Allocation: Non-Randomized
• Interventional Model: Parallel Assignment
• Masking: Single

Number of Arms

3

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: TBI Patients; Patients with Traumatic Brain Injury receiving the SeeMe Auditory Stimulation protocol daily to detect covert motor responses	Diagnostic test: SeeMe Multimodal Auditory Command Protocol; A standardized, computer-controlled auditory stimulation (AS) protocol designed to elicit and quantify microscopic motor responses. Protocol Details: Stimuli: Participants are presented with five distinct auditory commands: 1) 'Stick out your tongue,' 2) 'Open your eyes,' 3) 'Show me a smile,' 4) 'Close your hands,' and 5) a neutral control command ('Today is a sunny day'). Timing: Each command is presented 10 times via single-use headphones with a randomized 30-45 second jittered interval between trials to distinguish stimulus-evoked responses from spontaneous arousal. Data Capture: Responses are captured using high-resolution video (Panasonic HC-2000X) at 0.03s temporal resolution and synchronized millisecond-level EEG/ECoG. Analysis: Displacement heatmaps are generated via facial pore vector analysis and classified using a bidirectional long short-term memory (LSTM) neural network to determine the statistical significance of motor initiation compared to a 15-minute resting base; Other Names: Computer-Vision Based Consciousness Assessment
Active Comparator: Healthy Control Cohort; Awake, healthy volunteers receiving the SeeMe protocol to establish "ground truth" for normal voluntary motor signatures and algorithm sensitivity.	Diagnostic test: SeeMe Multimodal Auditory Command Protocol; A standardized, computer-controlled auditory stimulation (AS) protocol designed to elicit and quantify microscopic motor responses. Protocol Details: Stimuli: Participants are presented with five distinct auditory commands: 1) 'Stick out your tongue,' 2) 'Open your eyes,' 3) 'Show me a smile,' 4) 'Close your hands,' and 5) a neutral control command ('Today is a sunny day'). Timing: Each command is presented 10 times via single-use headphones with a randomized 30-45 second jittered interval between trials to distinguish stimulus-evoked responses from spontaneous arousal. Data Capture: Responses are captured using high-resolution video (Panasonic HC-2000X) at 0.03s temporal resolution and synchronized millisecond-level EEG/ECoG. Analysis: Displacement heatmaps are generated via facial pore vector analysis and classified using a bidirectional long short-term memory (LSTM) neural network to determine the statistical significance of motor initiation compared to a 15-minute resting base; Other Names: Computer-Vision Based Consciousness Assessment
Sham Comparator: Sedated/Paralyzed Patients; Patients undergoing general anesthesia and pharmacological paralysis receiving the SeeMe protocol to establish the algorithm's specificity and "noise floor."	Diagnostic test: SeeMe Multimodal Auditory Command Protocol; A standardized, computer-controlled auditory stimulation (AS) protocol designed to elicit and quantify microscopic motor responses. Protocol Details: Stimuli: Participants are presented with five distinct auditory commands: 1) 'Stick out your tongue,' 2) 'Open your eyes,' 3) 'Show me a smile,' 4) 'Close your hands,' and 5) a neutral control command ('Today is a sunny day'). Timing: Each command is presented 10 times via single-use headphones with a randomized 30-45 second jittered interval between trials to distinguish stimulus-evoked responses from spontaneous arousal. Data Capture: Responses are captured using high-resolution video (Panasonic HC-2000X) at 0.03s temporal resolution and synchronized millisecond-level EEG/ECoG. Analysis: Displacement heatmaps are generated via facial pore vector analysis and classified using a bidirectional long short-term memory (LSTM) neural network to determine the statistical significance of motor initiation compared to a 15-minute resting base; Other Names: Computer-Vision Based Consciousness Assessment

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Lead Time to Detection of Command-Following by SeeMe Compared With Blinded CRS-R Assessment	Number of days between the first SeeMe-detected significant stimulus-evoked motor response and the first blinded Coma Recovery Scale-Revised (CRS-R) assessment demonstrating command-following. A SeeMe-positive detection is defined as a stimulus-evoked movement that meets the prespecified criteria of a Kolmogorov-Smirnov statistic >0.1 and a pixel displacement >400. This stimulus-evoked movement must be detected reliably (at least 3 out of 10 trials). CRS-R examiners will be blinded to SeeMe outputs.	From Day 1 until hospital discharge, typically within 45 days.

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Diagnostic Performance of SeeMe for Subsequent Confirmed Command-Following	Sensitivity, specificity, positive predictive value, and negative predictive value of SeeMe-detected responses for subsequent blinded CRS-R-confirmed command-following.	From Day 1 until hospital discharge, typically within 45 days.
Rate of SeeMe-Detected Responsiveness in Participants Without Observable Command-Following at Enrollment	The percentage of patients labeled as "unresponsive" (GCS 3-8) who demonstrate statistically significant stimulus-evoked movements via SeeMe that are invisible to the naked eye.	From Day 1 until hospital discharge, typically within 45 days.
False Positive Rate of SeeMe in Pharmacologically Paralyzed Controls	The rate of "significant" movements detected by the SeeMe algorithm in the cohort of patients undergoing general anesthesia and pharmacological paralysis. This establishes the "noise floor" for the computer vision system.	During the localized 60-minute window of the surgical anesthesia procedure.
Correlation Between SeeMe Motor Initiation and Beta-Band Event-Related Desynchronization (ERD)	The Pearson correlation coefficient (r) used to assess the temporal relationship between the onset of facial/hand motor movements (as quantified by the SeeMe algorithm in pixels/second) and the magnitude of Beta-band (13-30 Hz) power suppression (measured in dB or muV^2 on synchronized EEG or ECoG. This confirms the neurophysiological validity of the computer-vision detection. Unit of Measure: Correlation Coefficient (r)	At each synchronized recording session from Day 1 through hospital discharge, typically within 45 days.
Multi-Class Command Classification Performance of the Bidirectional LSTM Model	Multiclass classification accuracy of the LSTM model for distinguishing responses to prespecified auditory commands and neutral control statements. More specifically, we will measure the accuracy percentage of correctly classifying facial/hand responses to five distinct auditory commands (e.g., "Show me a smile" vs. "Open your eyes") versus neutral control commands ("Today is a sunny day").	Assessed at the end of R61 model development and validation, approximately Year 3.
Long-term Functional Outcome via Glasgow Outcome Scale-Extended (GOS-E)	The GOS-E is a standardized assessment of the patient's global functional recovery and level of independence. Scores range from 1 (Death) to 8 (Upper Good Recovery). This measure will be used to determine if the "Lead Time" or "Sensitivity" of early SeeMe motor detection in the acute phase is a significant predictor of long-term functional independence.	6 months and 12 months post-injury.
Longitudinal Consciousness Recovery via Coma Recovery Scale-Revised (CRS-R)	The CRS-R is the clinical gold standard for assessing consciousness. By performing this assessment at long-term follow-up, we can track the trajectory of patients who were "SeeMe positive" (detected by the algorithm) but "clinically negative" (undetected by humans) during their initial ICU stay to confirm if they eventually reached higher states of overt consciousness.	6 months and 12 months post-injury.

Collaborators and Investigators

• Sponsor: Stony Brook University
• Collaborators: National Institute of Mental Health (NIMH)

Study record dates

Study Major Dates

• Study Start (Actual): March 30, 2026
• Primary Completion (Estimated): December 1, 2029
• Study Completion (Estimated): December 1, 2029

Study Registration Dates

• First Submitted: April 17, 2026
• First Submitted That Met QC Criteria: April 23, 2026
• First Posted (Actual): May 1, 2026

Study Record Updates

• Last Update Posted (Actual): May 5, 2026
• Last Update Submitted That Met QC Criteria: April 30, 2026
• Last Verified: April 1, 2026

More Information

Terms related to this study

• Keywords: Artificial Intelligence; Traumatic Brain Injury; Deep Learning; Coma; Disorders of Consciousness; Computer Vision; Cognitive-Motor Dissociation; Covert Awareness
• Additional Relevant MeSH Terms: Neurologic Manifestations; Brain Diseases; Central Nervous System Diseases; Nervous System Diseases; Mental Disorders; Wounds and Injuries; Neurobehavioral Manifestations; Neurocognitive Disorders; Craniocerebral Trauma; Trauma, Nervous System; Brain Injuries; Unconsciousness; Pathological Conditions, Signs and Symptoms; Signs and Symptoms; Brain Injuries, Traumatic; Consciousness Disorders; Coma
• Other Study ID Numbers: IRB2019--00199; 1R61MH138612-01 (U.S. NIH Grant/Contract)

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: YES
• IPD Plan Description: De-identified individual participant data (IPD) that underlie the results reported in the primary publication will be made available. This includes numerical vector data from SeeMe facial/hand tracking heatmaps, synchronized EEG/ECoG power spectra (Beta, Alpha, Delta, Theta bands), and blinded CRS-R clinical scores. To protect participant privacy, raw high-resolution video files will not be shared; however, processed, anonymized displacement time-series and algorithmic classification outputs will be provided. The Study Protocol and Statistical Analysis Plan will also be available.
• IPD Sharing Time Frame: Data will be made available beginning 6 months after the primary manuscript is published and will remain accessible for a period of 5 years. This allows the primary research team time to complete secondary analyses while ensuring the data remains useful for the scientific community during the R33 transition phase.
• IPD Sharing Access Criteria: Data will be shared with researchers who provide a methodologically sound proposal and have obtained IRB approval from their home institution. Proposals must be directed to the Principal Investigator (Sima Mofakham). To gain access, data requestors will need to sign a Data Use Agreement (DUA) that strictly prohibits any attempt to re-identify participants or use raw signal data for non-academic purposes.
• IPD Sharing Supporting Information Type: STUDY_PROTOCOL; SAP; ICF; ANALYTIC_CODE

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: Yes