# A Machine Learning-Based Decoder Framework for the Cortical Voltage-Sensitive Dye Responses to Retinal Neuromorphic Microstimulation: A Proof-of-Concept Simulation Study

## TL;DR

Researchers used machine learning to decode simulated brain responses to retinal microstimulation, showing it's possible to reconstruct visual images from cortical activity.

## Contribution

A novel machine learning framework was developed to decode simulated cortical responses to retinal neuromorphic microstimulation.

## Key findings

- A Wiener-system model successfully simulated VSD responses to ICMS pulse trains.
- A convolutional neural network reconstructed images from simulated cortical responses.
- The study demonstrates computational feasibility for decoding visual information from simulated cortical activity.

## Abstract

Intracortical microstimulation (ICMS) is a promising approach for visual prostheses. We recently proposed using retinal neuromorphic spike trains derived from visual images as ICMS pulse sequences, and preliminarily recorded cortical voltage-sensitive dye (VSD) responses to such stimulation. To examine whether these cortical responses contain image information, we explore the feasibility of machine-learning–based decoding. However, constructing such a decoder requires large-scale datasets linking visual images, spike trains, and cortical responses, which are not yet experimentally available. Therefore, we generated surrogate data with a Wiener-system model that simulates VSD responses of the visual cortex to ICMS pulse trains. A convolutional neural network trained on these synthetic datasets successfully reconstructed images from the simulated cortical responses. This simulation work serves as a proof-of-concept study, demonstrating the computational feasibility of estimating visual information contained in neuromorphic ICMS-evoked cortical activity and providing a foundation for future physiological validation.