[2025] A Hardware-Efficient Framework for DNN-Based Neural Spike Classification

The cerebellum is a brain region responsible for motor control. A wide range of behaviours can be investigated
through animal experiments that record electrophysiological signals from Purkinje cells in the cerebellum. Typi-
cally, these experiments are conducted in mice and involve wired connections between an implanted head stage or recording device and an external acquisition device, where the recorded neural data is processed offline. The wired setup limits the mice’s freedom of movement, compromising the realism of the experiments. To eliminate the need for wires, a dimensionality reduction step should be performed on the recording device, allowing only the relevant data to be stored on the head stage and retrieved for later offline analysis. However, a data compression algorithm that is low-power and lightweight enough to be implemented on the same head stage, without compromising the device’s battery life or making its weight unsuitable for placement on the mouse’s head, is yet to be developed.
 

Assignment

The neuronal spikes from Purkinje cells can be divided into simple and complex spikes. This distinctive spike
characteristic can be leveraged to achieve compression of the recorded cerebellum data. Thus, the aim of this
thesis is to develop a hardware-friendly algorithm that detects the spikes, classifies them based on their morphology, and compresses the data, enabling longer and wireless recordings. The algorithm will be deployed in a microcontroller. This project will be carried out in collaboration with Erasmus MC in Rotterdam, where the animal experiments are conducted.

Requirements

MSc student in Biomedical Engineering, Electrical Engineering, or Computer Engineering with an interest in signal processing, Machine Learning, and hardware design (microcontroller/firmware).

Contact

Last modified: 2025-03-10