Optimal deep brain stimulation therapy with Q learning
First Claim
1. A closed loop Deep Brain Stimulation (DBS) system comprising:
- a physiological sensor, said physiological sensor comprising at least one of a brain activity sensor and a mechanical body motion sensor;
an adaptive control system in communication with said physiological sensor; and
an implantable pulse generator (IPG), said IPG arranged to provide pulses selectively to one or more electrodes of a multi-electrode DBS lead responsive to stimulation parameters provided by said adaptive control system,said adaptive control system comprising;
a learning module arranged to determine stimulation parameters to be output to said IPG responsive to feedback received from said physiological sensor wherein said stimulation parameters comprise a selection of a particular set of said one or more electrodes of said multi-electrode DBS lead; and
a deterministic algorithmic module arranged to maintain safety limits for determined stimulation parameters provided to said IPG,wherein said adaptive control system is arranged to operate in a plurality of states, the states comprising;
a non-adaptive state wherein said adaptive control system controls said IPG to deliver stimulation in accordance with received input stimulation parameters;
an adaptive gradient ascent (GA) learning state wherein said input stimulation parameters are modified by said adaptive control system and optimal parameters for varying patient conditions are learned responsive to feedback received from said physiological sensor, said learning comprising storing reward function values responsive to said modified stimulation parameters and resultant inputs from said physiological sensor in a Q-learning (QL) look up table; and
an adaptive QL state in which said adaptive control system operates said IPG alternately responsive to inputs from said physiologic sensor and inputs stored in said QL look up table, said alternate selection responsive to a probabilistic scheme.
0 Assignments
0 Petitions
Accused Products
Abstract
A closed loop Deep Brain Stimulation (DBS) system constituted of: a physiological sensor; a multi-electrode DBS lead; an adaptive control system in communication with the physiological sensor; and an implantable pulse generator (IPG) responsive to the adaptive control system, the adaptive control system comprising a learning module operable to learn to find the optimal stimulation parameters, classify and associate patient conditions responsive to the physiological sensor with optimal stimulation parameters in a plurality of patient conditions. The adaptive DBS device control system learns to deliver the optimal stimulation parameters based on Watkins and Dayan Q learning recursive formula, the closed loop adaptive DBS control system thus finds the optimal stimulation parameters online.
-
Citations
17 Claims
-
1. A closed loop Deep Brain Stimulation (DBS) system comprising:
-
a physiological sensor, said physiological sensor comprising at least one of a brain activity sensor and a mechanical body motion sensor; an adaptive control system in communication with said physiological sensor; and an implantable pulse generator (IPG), said IPG arranged to provide pulses selectively to one or more electrodes of a multi-electrode DBS lead responsive to stimulation parameters provided by said adaptive control system, said adaptive control system comprising; a learning module arranged to determine stimulation parameters to be output to said IPG responsive to feedback received from said physiological sensor wherein said stimulation parameters comprise a selection of a particular set of said one or more electrodes of said multi-electrode DBS lead; and a deterministic algorithmic module arranged to maintain safety limits for determined stimulation parameters provided to said IPG, wherein said adaptive control system is arranged to operate in a plurality of states, the states comprising; a non-adaptive state wherein said adaptive control system controls said IPG to deliver stimulation in accordance with received input stimulation parameters; an adaptive gradient ascent (GA) learning state wherein said input stimulation parameters are modified by said adaptive control system and optimal parameters for varying patient conditions are learned responsive to feedback received from said physiological sensor, said learning comprising storing reward function values responsive to said modified stimulation parameters and resultant inputs from said physiological sensor in a Q-learning (QL) look up table; and an adaptive QL state in which said adaptive control system operates said IPG alternately responsive to inputs from said physiologic sensor and inputs stored in said QL look up table, said alternate selection responsive to a probabilistic scheme. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10)
-
-
11. A method for deep brain stimulation (DBS) delivered to a patient in a closed loop system, said method comprising:
-
providing a physiological sensor; providing a multi-electrode DBS lead; determining stimulation parameters to be provided via said provided multi-electrode DBS lead responsive to feedback received from said provided physiological sensor wherein said stimulation parameters comprise a selection of a particular set of said one or more electrodes of said provided multi-electrode DBS lead; and maintaining safety limits for said stimulation parameters provided to said multi-electrode DBS lead, wherein said determining stimulation parameters comprises; delivering brain stimulation via said provided multi-electrode DBS lead according to programmed stimulation parameters; machine learning the programmed stimulation parameters; switching to an adaptive gradient ascent (GA) learning state when the machine learning converges; modifying the programmed stimulation parameters thereby learning optimal stimulation parameters responsive to feedback from said provided physiological sensor; storing reward function values responsive to said modified stimulation parameters and resultant inputs from said physiological sensor in a Q-learning (QL) look up table; and in the event that said learning optimal stimulation parameters converges within predetermined parameters, switching to an adaptive QL state in which brain stimulation is provided via said provided multi-electrode DBS lead alternately responsive to inputs from said physiologic sensor and inputs stored in said QL look up table, said alternate selection responsive to a probabilistic scheme. - View Dependent Claims (12, 13, 14, 15, 16, 17)
-
Specification