Friday Lunch Seminar (English)
December 8, 2023　　
12:15 〜 13:00 (JST)
Conference Room, CiNet bldg. / On-line

Talk Title：A state estimation method of unmeasurable neurons from local spike recordings

Takahiro Asahina
Guest Researcher
Brain Networks and Communication Laboratory
Center for Information and Neural Networks (CiNet)
National Institutes of Information and Communications Technology (NICT)

Host PI :  Takafumi Suzuki

Abstract:
In this talk, I will present my research in my doctoral course at the University of Tokyo investigating estimation of neuronal ensemble from spike recordings.

Since neural recording technologies such as Utah array and ECoG have limitation on recording area, estimation of neuronal state which is outside the recording area is said to be effective. We developed a method for estimating the state of many neurons that are synchronously active outside the recording area. The state of a neuronal population was estimated from only local spike recording and was used as feature values in BMI decoding.

The estimation method is based on mathematical model of a system including recorded and unrecorded neurons, with some simplifications. By assuming a mean field approximation that activity of many neurons can be handled in an averaged manner, we constructed a mathematical model to estimate synaptic transmission and synaptic connectivity, then derived a maximum likelihood estimation.

We confirmed the estimation abilities of the constructed method by simulating neuronal activity and conducting multiple experiments using cultured neurons on microelectrode arrays. In particular, experiments using optogenetics to control neuronal activities of cultured neurons demonstrated that multiple information expression could be discriminated using spike recordings.

Also, we evaluated the improvement of BMI decoding accuracy with the estimation method. By using the estimated synaptic connectivity as additional feature values, motor BMI decoding accuracy from public spike data improved by 11% in average. The possibility that the estimation of neuronal states can be applied clinically was demonstrated from in vivo data.