Master Thesis: reconstructing visual image from spontaneous activity in mouse visual cortex
This Master’s thesis aims at tackling an important and fascinating question: can we decipher what mice are imagining or “seeing” when they are simply mind-wandering? Using two-photon calcium imaging and statistical modeling methods, we will develop a “mental image” reconstructor to “read” the visual images encoded by spontaneous activity in primary visual cortex
Our brain is always active. Even when we rest or sleep, our brain is occupied with spontaneous activity coordinated across the distributed network of regions. These spontaneous activities are not random, but rather occur in repetitive activity patterns that underlies recalling the past, rehearsing the future, or imagining possible or impossible events. The visual world, with its rich features and constant presence, has a strong influence on how we remember and imagine the world. As memory recall often involves a visual aspect, spontaneous activity in the brain might also reflect many aspects of the visual experience. The goal of this project is to test the hypothesis that spontaneous activity in visual cortex encodes rich details of visual experience, using mice visual cortex as a model.
Background and previous works
To understanding offline processes in the cortex, it is essential to have a reliable readout method of the encoded information. This idea is particularly attractive when it comes to the visual system, and a few human neuroimaging studies have proposed different ways to do so (Kamitani & Tong, 2005; Nishimoto et al., 2011; Thirion et al., 2006). Despite the limited spatial and temporal resolution of human imaging, it was possible to at least decode the most probable image from a predefined pool, using low-resolution visual cortex activity. Recently, one study using two-photon calcium imaging in mice has demonstrated that it is possible to construct the represented image from evoked visual activity (Yoshida & Ohki, 2020). After repetitively showing mice hundreds of natural images, this study deconstructed natural images to Gabor patches, a visual element that has been shown to evoke strong response in many V1 neurons, and estimated each V1 neuron’s responsive field and tuning properties within it. Then, the authors reconstructed the natural image shown to the mice both under anesthetized and awake conditions, using a linear combination of the preferred Gabor patches.
We will attempt to develop a spontaneous visual image reconstruction pipeline by directly using Gabor patch stimuli of different spatial scales, visual field locations and orientations and mapping V1 neuronal tuning properties to them. By modelling a linear weight vector for each neuron against all Gabor stimuli, we will make a generative model that can reconstruct the visual image using V1 population activity. Beyond validation with evoked activity, we will further use this method to investigate the information content of spontaneous activity of V1 during offline periods and sleep.
◦ Familiarize yourself with the literature and biological questions in the field
◦ Design a mapping program with Gabor filters
◦ Coupling of mapping program with image acquisition system
◦ Possibility to learn mouse surgery skills
◦ Possibility to perform in vivo mapping experiments and acquire pilot data
◦ If pilot data is acquired, explore analysis pipeline for visual image reconstruction
• Interest in neuroscience questions
• Programming experience (Python/Matlab).
• Willing to have hands-on experience
The project will take place in Prof. Fritjof Helmchen’s group at Brain Research Institute. The supervisor will be Dr. Shuting Han.
Interested students should sent an e-mail to han(at)hifo.uzh.ch and bethge(at)hifo.uzh.ch. Please attach a brief statement explaining your background/broad interests, and a copy of your CV.