Previous implementations of closed-loop peripheral electrical stimulation (PES) strategies have provided evidence about the effect of the stimulation timing on tremor reduction. However, these strategies have used traditional signal processing techniques that only consider phase prediction and might not model the non-stationary behavior of tremor. Here, we tested the use of long shortterm memory (LSTM) neural networks to predict tremor signals using kinematic data recorded from Essential Tremor (ET) patients. A dataset comprising wrist flexionextension data from 12 ET patients was pre-processed to feed the predictors. A total of 180 models resulting from the combination of network (neurons and layers of the LSTM networks, length of the input sequence and prediction horizon) and training parameters (learning rate) were trained, validated and tested. Predicted tremor signals using LSTM-based models presented high correlation values (from 0.709 to 0.998) with the expected values, with a phase delay between the predicted and real signals below 15 ms, which corresponds approximately to 7.5% of a tremor cycle. The prediction horizon was the parameter with a higher impact on the prediction performance. The proposed LSTM-based models were capable of predicting both phase and amplitude of tremor signals outperforming results from previous studies (32-56% decreased phase prediction error compared to the out-of-phase method), which might provide a more robust PES-based closed-loop control applied to PES-based tremor reduction.

Peripheral Electrical Stimulation (PES) of afferent pathways has received increased interest as a solution to reduce pathological tremors with minimal side effects. Closed-loop PES systems might present some advantages in reducing tremors, but further developments are required in order to reliably detect pathological tremors to accurately enable the stimulation only if a tremor is present. This study explores different machine learning (K-Nearest Neighbors, Random Forest and Support Vector Machines) and deep learning (Long Short-Term Memory neural networks) models in order to provide a binary (Tremor; No Tremor) classification of kinematic (angle displacement) and electromyography (EMG) signals recorded from patients diagnosed with essential tremors and healthy subjects. Three types of signal sequences without any feature extraction were used as inputs for the classifiers: kinematics (wrist flexion–extension angle), raw EMG and EMG envelopes from wrist flexor and extensor muscles. All the models showed high classification scores (Tremor vs. No Tremor) for the different input data modalities, ranging from 0.8 to 0.99 for the f1 score. The LSTM models achieved 0.98 f1 scores for the classification of raw EMG signals, showing high potential to detect tremors without any processed features or preliminary information. These models may be explored in real-time closed-loop PES strategies to detect tremors and enable stimulation with minimal signal processing steps.