The purpose of this study is to look at how signals in the brain, body, and behavior relate to anxiety and memory function. This project seeks to develop the CAMERA (Context-Aware Multimodal Ecological Research and Assessment) platform, a state-of-the-art open multimodal hardware/software system for measuring human brain-behavior relationships. The R61 portion of the project is designed to develop the CAMERA platform, which will use multimodal, passive sensor data to predict anxiety-memory state in patients undergoing inpatient monitoring with intracranial electrodes for clinical epilepsy, as well as to build CAMERA's passive data framework and active data framework.
CAMERA will record neural, physiological, behavioral, and environmental signals, as well as measurements from ecological momentary assessments (EMAs), to develop a continuous high-resolution prediction of a person's level of anxiety and cognitive performance. CAMERA will provide a significant advance over current methods for human behavioral measurement because it leverages the complementary features of multimodal data sources and combines them with interpretable machine learning to predict human behavior. A further distinctive aspect of CAMERA is that it incorporates context-aware, adaptive EMA, where the timing of assessments depends on the subject's physiology and behavior to improve response rates and model learning. In this study, CAMERA focuses on predicting anxiety state and concurrent memory performance, but the platform is flexible for use in various domains. Currently, it is challenging to study complex, longitudinal relationships between the brain, body, and environment in humans. Most existent tools do not allow the investigator to measure transient internal states or cognitive functions comprehensively or continuously. Instead the investigators typically rely on sparsely collected and constrained self-reports or experimental constructs, including EMA.
Study Type
INTERVENTIONAL
Allocation
NA
Purpose
SCREENING
Masking
NONE
Enrollment
40
The CAMERA platform is a multimodal, hardware-software framework for measuring brain-behavior interactions in an unstructured environment and predict ecological states. CAMERA will use multimodal, passive sensor data to predict anxiety-memory state in patients undergoing inpatient monitoring with intracranial electrodes for clinical epilepsy. CAMERA consists of: Wristband sensors of autonomic physiologic signals, emphasizing heart rate metrics and electrodermal activity; Smartphone usage, emphasizing natural language processing of text input for linguistic features; Subject-tracking audiovisual array, emphasizing subject vocal activity; Intracranial neural recordings, emphasizing hippocampal theta power and high-frequency activity (\~70-200 Hz).
Columbia University Irving Medical Center
New York, New York, United States
RECRUITINGMean absolute error between predicted and actual ecological momentary assessment (EMA) scores
Use a multimodal machine learning model (EMANet ) to predict ≥1 EMA anxiety-memory state outcome (target) in held-out data at the population level. Mean absolute error will be the mean difference in absolute value of predicted EMA and actual EMA scores. A higher mean error represents a less accurate prediction. Prediction must use ≥2 different passive modalities, showing significantly better prediction accuracy than either of the modalities alone.
Time frame: 1-30 days
Percent of subjects demonstrating improvement in the EMANet prediction over time.
Use EMANet to predict ≥1 ecological momentary assessment (EMA) anxiety-memory state outcome (target) demonstrating improvement over time as measured with a linear regression applied to the mean absolute error between predicted and actual EMA values measured over days. Prediction must use ≥2 different passive modalities, showing significantly better prediction accuracy than either of the modalities alone.
Time frame: 1-30 days
Mean absolute error between predicted and actual absolute error on a daily basis
Use a multimodal machine learning model of prediction uncertainty (UncertaintyNet) to predict the mean absolute prediction error of ecological momentary assessment (EMA) predictions in held-out data, at single-subject level on each day. Mean absolute error will measure the difference between the predicted error (based on all available data) and the actual error.
Time frame: 1-30 days
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.