Tomas Knapen

pRF fitting toolbox wish list With this form we are taking stock of the field's wishes when it comes to pRF fitting software implementations. We will be presenting the results from this form in our kick-off meeting, and will use t...

Hey everyone at @vssmtg.bsky.social! If you’re interested in pRF fitting, go visit Garikoitz Lerma-Usabiaga’s poster on pRF fitting methods!

For our development of these tools, we’re very interested to hear you want in these tools. Please fill out our questionnaire:

forms.gle/fx5UMs1362jv...

𝘗𝘦𝘳𝘴𝘰𝘯, 𝘴𝘵𝘢𝘯𝘥𝘪𝘯𝘨 𝘰𝘯 𝘵𝘩𝘦 𝘣𝘦𝘢𝘤𝘩 𝘥𝘰𝘪𝘯𝘨 𝘺𝘰𝘨𝘢, 𝘴𝘵𝘢𝘯𝘥𝘪𝘯𝘨 𝘰𝘯 𝘰𝘯𝘦 𝘧𝘰𝘰𝘵, 𝘸𝘪𝘵𝘩 𝘰𝘯𝘦 𝘩𝘢𝘯𝘥 𝘰𝘯 𝘵𝘩𝘦 𝘨𝘳𝘰𝘶𝘯𝘥, 𝘤𝘰𝘯𝘵𝘰𝘳𝘵𝘦𝘥 𝘪𝘯 𝘢𝘯 𝘶𝘯𝘶𝘴𝘶𝘢𝘭 𝘢𝘯𝘥 𝘤𝘩𝘢𝘭𝘭𝘦𝘯𝘨𝘪𝘯𝘨 𝘱𝘰𝘴𝘦, 𝘸𝘩𝘪𝘭𝘦 𝘤𝘰𝘯𝘵𝘦𝘮𝘱𝘭𝘢𝘵𝘪𝘯𝘨 𝘵𝘩𝘦 𝘪𝘥𝘦𝘢 𝘵𝘩𝘢𝘵 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘈𝘐 𝘩𝘢𝘴 “𝘸𝘰𝘳𝘭𝘥 𝘮𝘰𝘥𝘦𝘭𝘴”

A thread motivated by a new paper on body representations in the human brain at a fine-grained (multi-unit) level, spearheaded by J Garcia Ramirez, T Theys, and P Janssen, where I was a small part of a bigger collaboration that also included S Bracci, R Murty and @nancykanwisher.bsky.social. 1/n

And people sampling the videos with their eyes allows them to shape their own brain responses. This will likely generate an additional level of “individuality” to brain responses, lowering ISC

Our results indicate the brain uses aligned, 'multiplexed' topographic maps to structure connections between vision and somatosensation. The computational machinery classically attributed to the somatosensory system is embedded within/aligned with that of the "visual" system. 🧵

These findings complement recent work indicating that dorsolateral visual cortex is a fundamentally multi-sensory part of the brain whose role extends beyond passive visual analysis to encompass semantic and bodily information relevant to interactions with the world. 20/n

These encoding model fits revealed a new map of visual body-part selectivity, which overlapped with somatotopic tuning across the FBA, EBA and, strikingly, the visual word form area (VWFA). 19/n

To address this, we combined the Natural Scenes Dataset with a pose-detection algorithm fit a body-part tuning encoding model. This allowed us to generate a map of visual body part preference, organised along a similar toe-to-tongue axis as the somatotopic connectivity maps. 18/n

Much of visual cortex is body-part selective. If this tuning relates to our somatotopic connectivity, we should also be able to predict visual body part selectivity from somatotopic tuning and reveal multi-modal body-referenced alignment playing out at more semantic levels. 17/n

We did indeed find evidence for an alignment between visual field tuning and body part tuning beyond that expected by chance. We found this mostly dorsally and in the superior portion of EBA. 16/n

But do these bodily maps predict anything about visual function? For instance, could lower body part tuning (e.g. toes) predict lower visual field tuning? Such an alignment might facilitate interactions with the environment. 15/n

Yes! Throughout dorsolateral visual cortex, we see several body-part gradients separated by reversals. These maps were consistent across hemispheres and subject splits. 14/n

But what about the body part tuning of these somatotopic activations? Do these dorsolateral regions exhibit orderly gradients, as found in 'core' somatosensory regions around the central sulcus? The answer is... 13/n

We then repeated our somatosensory connectivity analyses separately on a movie section involving human agents and another without any humans. This demonstrated that somatotopic responses are not generic, but driven by movie content, specifically that featuring human action. 12/n

Our analysis allows us to contrast somatotopic and retinotopic explained variance. All dorsolateral (but not ventral!) visual regions were characterised by multimodal topographic connectivity. These regions care as much or more about the body as they do the visual scene! 11/n

We find that movie watching led to increased somatotopic connectivity in the somatosensory network outlined above. But strikingly, we now also find that dorsolateral visual cortex has structured connectivity with S1. Look at that red band across visual cortex! 10/n

So, we turned to the HCP movie watching experiment. This dataset allows us to investigate the relation of somatosensory connectivity to naturalistic visual experiences, where mental content is yoked to a visual stimulus. 9/n

So, during resting state, endogenous activations throughout frontal, parietal, and insular cortex resonate along scaffolding provided by the somatotopic structure of bodily sensations. But how this resonance relates to mental content in resting state is unclear... 8/n

Body part tuning revealed multiple somatotopic gradients and body-part tuning biases that are typically only revealed by exogenous stimulation (e.g. brushing people). Critically, we show that these same detailed principles can be revealed in the absence of sensory input! 7/n

Since this method uses one part of the brain to explain another, we can fit these models to any data - even with no stimulus! Analysing 7T resting state data from the HCP, we found signals in a large cortical network were predicted by somatotopic connectivity with S1. 6/n

This allows us to 'project' the visual field or body part tuning of V1 and S1 onto the rest of the brain. We’re performing connectivity-derived retinotopic and somatotopic mapping, which allows us to find higher-level sensory maps throughout the brain. 5/n

Our computational model of functional connectivity explains target voxels’ timecourses as 'connective fields' located on the surfaces of both V1 and S1. Due to their specific locations on our primary sensory cortices, connective fields inherit visual and somatotopic tuning. 4/n

These findings raise the question how the brain connects the computational machinery of vision and touch. Here, we developed a model for measuring joint visual and somatosensory tuning throughout the brain and applied it to resting state and movie watching data. 3/n

When seeing bodily experiences, our brain responds 'as if' it simulates the observed tactile experience as our own. We know that e.g. viewing fingers being touched can activate finger-selective regions of primary somatosensory cortex (S1) (tinyurl.com/vissom3b) 2/n

Since I need a first bsky post, here's a rehash of @nickhedger.bsky.social's thread announcing our preprint with Kendrick Kay & Thomas Naselaris. TLDR; High-level visual cortex is tiled with maps 'multiplexing' vision and touch. tinyurl.com/seesoma 🧵 1/n

I can confirm your #10: lowering the sampling rate from 1 or 2 kHz to 500 or even 250 can drastically reduce pixel noise!