Prediction of Parkinsonian Clinical State During Deep Brain Stimulation With AI

Stereotactic Neurosurgery operation, Pasteur 2 Hospital, Nice, France, Treating Parkinsons disease through deep brain stimulation, by implanting electrodes in brain, modulating abnormal cerebral electrical activity, The room is equipped with a pre-op O-arm scanner and NeuroMate, a stereotactic robot. The robot enables surgeons to optimize and increase reliability of their operating technique, and the O-arm enables real time 3D images to be obtained to improve the precision, safety and efficiency of the operation. The start of the operation. The surgeon and his team prepare locations using the O-arm. (Photo by: BSIP/Universal Images Group via Getty Images)
Presenting at the 2022 AAN Annual Meeting, researchers identified patient-specific spectral biomarkers of parkinsonian clinical state during deep brain stimulation using artificial intelligence.

The following article is part of conference coverage from the 2022 American Academy of Neurology (AAN) Annual Meeting. Neurology Advisor’s staff will be reporting breaking news associated with research conducted by leading experts in neurology. Check back for the latest news from the 2022 AAN Annual Meeting.


Measuring beta band neural oscillations during deep brain stimulation (DBS), despite its relation to motor activity, is not always the most useful biomarker of a parkinsonian clinical state, according to research findings presented at the 2022 American Academy of Neurology (AAN) Annual Meeting, held from April 2 to April 7 in Seattle, Washington, and virtually from April 24-26, 2022.

Beta frequency band neural oscillations are usually related to motor activity, showing a decrease prior to movement, then a rebound after movement is finished. DBS has been used for many years to help manage tremors produced by Parkinson disease (PD). Adaptive DBS (aDBS), a more recent approach to regulate stimulation moment by moment, dependent on measuring the patient’s current parkinsonian condition, may be enhanced by machine learning to estimate the clinical state.

Researchers conducted a prospective study of 3 patients with PD, implanted with neurostimulators to record subcortical local field potentials (LFPs).

Electrocorticography measured activity during a medication cycle of the high and low-levodopa states. Different stimulation settings were used: low amplitude for the high-medication state anticipating movement, and, high amplitude for the low-medication state, anticipating rebound. “Partitioning the subcortical and cortical power spectra and using forward-feature selection with linear discriminant analysis,” helped researchers identify spectral biomarkers that would best predict the medication state.

Researchers found no consistency patient-to-patient for which biomarker predicted the clinical state. Even though they measured beta neural oscillation peaks during OFF-stimulation, they discovered that beta power is not the most useful distinguishing biomarker.

“Subcortical beta power is not always the most discriminative biomarker of parkinsonian clinical state during DBS stimulation, even if beta peaks are seen during stimulation-OFF conditions,” they concluded. “Machine learning may be useful to identify data-driven patient-specific frequency bands that better predict clinical state during active stimulation.”


Hammer L, Oehrn C, Smyth C, et al. Artificial intelligence identifies spectral biomarkers for use in adaptive deep brain stimulation in Parkinson’s disease. Presented at: the 2022 AAN Annual Meeting; April 2-7, 2022; Seattle, Washington; April 24-26, 2022; Virtual Meeting. Abstract S16.004.