Closed-Loop Deep Brain Stimulation Based on a Dual Machine Learning System Analyzing Brain Functional Networks

Session Number

MEDH(ai) 42

Advisor(s)

James Kragel, University of Chicago

Discipline

Medical and Health Sciences

Start Date

17-4-2025 11:10 AM

End Date

17-4-2025 11:25 AM

Abstract

Neurodegeneration impairs memory, with treatments mainly addressing symptoms, not underlying dysfunction. Deep brain stimulation (DBS) shows promise by directly modulating memory. This study hypothesizes that DBS improves memory when targeted towards poor encoding states by influencing semantic organization. Data from 38 epilepsy patients with implanted electrodes were analyzed and it was discovered that LTC stimulation improved semantic organization during free recall, a critical factor for overall memory accuracy. LTC stimulation had a 27.54% success rate, higher than other memory-related regions (e.g. hippocampus, medial temporal lobe).

Next, 38 patient-specific convolutional neural network (CNN) models were trained using various brain functional networks including envelope correlation and directed phase lag index during the first 200 milliseconds of memory encoding. These models identified states of poor memory encoding, achieving an AUC of 72 ± 13%. Connecting these model outputs with logistic regression, LTC stimulation during poor-memory encoding states achieved an AUC of 89%, outperforming the 57%–67% scores in other areas.

These findings support the potential of a dual-system machine learning approach targeting the LTC to improve memory.

Share

COinS
 
Apr 17th, 11:10 AM Apr 17th, 11:25 AM

Closed-Loop Deep Brain Stimulation Based on a Dual Machine Learning System Analyzing Brain Functional Networks

Neurodegeneration impairs memory, with treatments mainly addressing symptoms, not underlying dysfunction. Deep brain stimulation (DBS) shows promise by directly modulating memory. This study hypothesizes that DBS improves memory when targeted towards poor encoding states by influencing semantic organization. Data from 38 epilepsy patients with implanted electrodes were analyzed and it was discovered that LTC stimulation improved semantic organization during free recall, a critical factor for overall memory accuracy. LTC stimulation had a 27.54% success rate, higher than other memory-related regions (e.g. hippocampus, medial temporal lobe).

Next, 38 patient-specific convolutional neural network (CNN) models were trained using various brain functional networks including envelope correlation and directed phase lag index during the first 200 milliseconds of memory encoding. These models identified states of poor memory encoding, achieving an AUC of 72 ± 13%. Connecting these model outputs with logistic regression, LTC stimulation during poor-memory encoding states achieved an AUC of 89%, outperforming the 57%–67% scores in other areas.

These findings support the potential of a dual-system machine learning approach targeting the LTC to improve memory.