In a study published in Cognitive Neurodynamics, a research team led by Prof. LI Guanglin and Prof. FANG Peng from the Shenzhen Institutes of Advanced Technology (SIAT) of the Chinese Academy of Sciences proposed an optimized deep learning algorithm for detecting sensory events occurring during stimulation of the nerves using a stretchable and flexible electrocorticography (ECoG) electrode.  

This algorithm was a combination of sparse auto-encoder (SSAE) for feature extraction, and back propagation neural network (BPNN) model for pre-training and fine-tuning of weights/biases.  

In order to test the algorithm, the researchers used the flexible and stretchable microelectrode array, which is the brain interface device for recording ECoG signals, and was based on gold films and elastic substrates.  

Considering a mechanical stimulation with Von Frey filament on the plantar surface of the mouse’s foot, they implanted the flexible electrode sensors on the surface of the mouse’s brain cortex, the tactile afferent signals were obtained from the primary somatosensory cortex (S1) simultaneously.   

The paw withdrawal threshold (PWT) assessment results showed that the proposed algorithm model possessed high detection, accuracy, sensitivity, and specificity compared with other methods.   

"Additionally, early detection of sensory events would improve spontaneous sensory and motor activity," said Oluwagbenga Paul Idowu, the first author of this study.