FPGA-Embedded Serial SVM Classifier for Neuromuscular Disorders Assessment

The paper proposes a portable clinical tool that works on both electromyographic (EMG) and electroencephalographic (EEG) signals. The core is a digital processor architecture, which aims to bridge the clinical gap between the muscular syndromes detection and typically disjointed differential neurodegenerative evaluation. The architecture has been implemented on Altera Cyclone V FPGA interfaced with a wireless sensing system made up by 8 EMGs (lower limbs) and 7 EEGs (motor-cortex) electrodes. The signals are acquired and processed on-board following the workflow: (i) Signal digitization and pre-processing (ii) time-frequency features extraction (FE) (iii) Serial Support Vector Machine (SSVM) realtime classification. The system has been validated on 4 subjects affected by mild and severe PD, 2 subjects for each pathology stage. The results showed that overall processing chain is able to categorizes and distinguish among each other two subsequent pathology stages with an accuracy of about 94%. The accuracy in healthy subjects’ recognition is near the 100%. All the evaluations referred to subjects involved in a clinical walking protocol. The FPGA resources utilization results in: 31.04% adaptive logic modules, 15.87% registers, 5.85% memory blocks and the 82.75% DSP blocks of an Altera Cyclone V FPGA.