فهرست نشریات علمی دسترسی آزاد ایران

Brain-machine interfaces (BMIs) have emerged as promising technologies for restoring motor functions by translating neural signals into control commands for external devices. Local field potentials (LFPs), representing aggregate neuronal activity, provide a rich source of information for decoding motor intentions, especially in freely moving animals where naturalistic behavior is preserved. This study presents a comprehensive approach to decoding force-related neural patterns from LFP signals recorded in the primary motor cortex (M1) of freely moving rats during an active lever-press task. Employing advanced deep learning architectures, including Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and a hybrid CNN-LSTM model, we systematically evaluate the efficacy of each in capturing both spatial and temporal features of LFP signals. Our models leverage multi-band frequency components of LFPs, focusing on gamma-band activity, known for its correlation with motor output. Using a dataset derived from established experimental protocols, we demonstrate that the hybrid CNN-LSTM architecture outperforms standalone models, achieving a Pearson correlation coefficient exceeding 0.89 ± 0.02 and an R² score above 0.81 ± 0.03 in force prediction tasks. The integration of deep learning with LFP-based decoding offers a robust framework for next-generation BMIs, potentially enhancing neuroprosthetic control fidelity in real-world scenarios.