Complexity of fractal dimension patterns and machine learning-based classification of altered motor cortical oscillatory activity in rodent models of Parkinson disease

Background In Parkinson disease (PD), enhanced beta frequency band activity in cortico-basal ganglia networks has been proposed as a possible biomarker for adaptive deep brain stimulation (DBS). Previous studies demonstrated enhanced beta frequency peaks both in the acute Haloperidol (HALO) and the chronic 6-hydroxydopamine (6-OHDA) rat models of parkinsonism. Beta peaks decreased after apomorphine (APO) injection or with DBS of the subthalamic nucleus (STN). Objective We investigate changes in motor cortical oscillatory activity using fractal dimension (FD) in the HALO and the 6-OHDA rat models of PD. Additionally, we test a support vector machine (SVM) model to predict neuronal dynamics in the 6-OHDA PD model which has been used earlier in an acute rat model of PD. Methods In the HALO model, electrocorticogram (ECoG) was recorded from the motor cortex (MCtx) (1) during basal activity, (2) after injection of HALO (0.5 mg/kg), and (3) after subsequent APO injection (1 mg/kg). For the chronic 6-OHDA model, MCtx-ECoG recordings were obtained (1) during basal activity, and (2) during STN DBS. Higuchi's FD algorithm and SVM-based classification were utilized for analysis. Results Average FD values in the MCtx were higher in both PD models compared to controls ( P 0.001). APO injection ( P 0.001) and STN DBS ( P 0.05) reduced average FD values in both models. The SVM model achieved 80% classification accuracy and an AUC of 0.86 in the 6-OHDA rat model. Conclusion The non-linear analysis of FD reveals changes in cortical oscillatory patterns in rodent models of PD. SVM-based predictions demonstrate potential for classifying altered neural activity, which may offer future strategies for adaptive DBS.