A Computationally-Efficient, Online-Learning Algorithm for Detecting High-Voltage Spindles in the Parkinsonian Rats
- PDF / 1,316,461 Bytes
- 12 Pages / 593.972 x 792 pts Page_size
- 53 Downloads / 148 Views
Annals of Biomedical Engineering ( 2020) https://doi.org/10.1007/s10439-020-02680-0
Original Article
A Computationally-Efficient, Online-Learning Algorithm for Detecting High-Voltage Spindles in the Parkinsonian Rats RAMESH PERUMAL,1 VINCENT VIGNERON,2,3 CHI-FEN CHUANG,4 YEN-CHUNG CHANG,4 SHIH-RUNG YEH,4 and HSIN CHEN1,5 1
Department of Electrical Engineering, National Tsing Hua University, No.101, Sec.2 Kuang-Fu Road, Hsinchu 30013, Taiwan, R.O.C.; 2IBISC, EA 4526, Universite Evry, Universite Paris-Saclay, Saint-Aubin, France; 3DSPCom, FAC/UNICAMP, Limeira, SP, Brazil; 4Institute of Molecular Medicine, National Tsing Hua University, No.101, Sec.2 Kuang-Fu Road, Hsinchu 30013, Taiwan, R.O.C.; and 5Biopro Scientific, Unit 312, Center of Innovative Incubator, No.101, Sec.2, Kuangfu Rd., East Dist., Hsinchu 30013, Taiwan, R.O.C. (Received 27 May 2020; accepted 22 October 2020) Associate Editor Xiaoxiang Zheng oversaw the review of this article.
Abstract—Abnormally-synchronized, high-voltage spindles (HVSs) are associated with motor deficits in 6-hydroxydopamine-lesioned parkinsonian rats. The non-stationary, spike-and-wave HVSs (5-13 Hz) represent the cardinal parkinsonian state in the local field potentials (LFPs). Although deep brain stimulation (DBS) is an effective treatment for the Parkinson’s disease, continuous stimulation results in cognitive and neuropsychiatric side effects. Therefore, an adaptive stimulator able to stimulate the brain only upon the occurrence of HVSs is demanded. This paper proposes an algorithm not only able to detect the HVSs with low latency but also friendly for hardware realization of an adaptive stimulator. The algorithm is based on autoregressive modeling at interval, whose parameters are learnt online by an adaptive Kalman filter. In the LFPs containing 1131 HVS episodes from different brain regions of four parkinsonian rats, the algorithm detects all HVSs with 100% sensitivity. The algorithm also achieves higher precision (96%) and lower latency (61 ms), while requiring less computation time than the continuous wavelet transform method. As the latency is much shorter than the mean duration of an HVS episode (4.3 s), the proposed algorithm is suitable for realization of a smart neuromodulator for mitigating HVSs effectively by closed-loop DBS. Keywords—Parkinson’s disease, Autoregressive modeling, Adaptive Kalman filter, Hilbert-Huang transform, Smart neuromodulator, Closed-loop deep brain stimulation.
ABBREVIATIONS AKF AR cDBS CWT DBS FN FP FPGA HHT HVS LFP ML PACF PD PSD SNR TP TR 6-OHDA
Adaptive Kalman filter Autoregressive model Closed-loop deep brain stimulation Continuous wavelet transform Deep brain stimulation False negative False positive Field-programmable gate array Hilbert-Huang transform High-voltage spindle Local field potential Machine-learning Partial autocorrelation function Parkinson’s disease Power spectral density Signal-to-noise ratio True positive Detection threshold 6-Hydroxydopamine
INTRODUCTION Address correspondence to Hsin Chen, Department of Electrical En
Data Loading...
