bioRxiv Science⌕ Search

Biology subjects

Lage-Castellanos, A.

Publications and source records attributed to Lage-Castellanos, A..

7 recordsLinked to original sources
bioRxiv2024

Unbiased estimation of the coefficient of determination in linear models: an application to fMRI encoding model comparison

Neuroscientific investigation has greatly benefited from the combination of functional Magnetic Resonance Imaging (fMRI) with linearized encoding, which allows to validate and compare computational models of neural activity based on neuroimaging data. In linearized encoding, a multidimensional feature space, usually obtained from a computational model applied to the stimuli, is related to the measured brain activity. This is often done by mapping such space to a dataset (training data, or in-sample), and validating the mapping on a separate dataset (test data, or out-of-sample), to avoid overfitting. When comparing models, the one with the highest explained variance on the test data, as indicated by the coefficient of determination (R2), is the one that better reflects the neural computations performed by the brain. An implicit assumption underlying this procedure is that the out-of-sample R2 is an unbiased estimator of the explanatory power of a computational model in the population of stimuli, and can therefore be safely used to compare models. In this work, we show that this is not the case, as the out-of-sample R2 has a negative bias, related to the amount of overfitting in the training data. This phenomenon has dramatic implications for model comparison when models of different dimensionalities are compared. To this aim, we develop an analytical framework that allows us to evaluate and correct biases in both in- and out-of-sample R2, with and without L2 regularization. Our proposed approach yields unbiased estimators of the population R2, thus enabling a valid model comparison. We validate it through illustrative simulations and with an application to a large public fMRI dataset.

neuroscience↗
bioRxiv2024

Evaluating the effect of denoising submillimeter auditory fMRI data with NORDIC

Functional magnetic resonance imaging (fMRI) has emerged as an essential tool for exploring human brain function. Submillimeter fMRI, in particular, has emerged as a tool to study mesoscopic computations. The inherently low signal-to-noise ratio (SNR) at submillimeter resolutions warrants the use of denoising approaches tailored at reducing thermal noise - the dominant contributing noise component in high resolution fMRI. NORDIC PCA is one of such approaches, and has been benchmarked against other approaches in several applications. Here, we investigate the effects that two versions of NORDIC denoising have on auditory submillimeter data. As investigating auditory functional responses poses unique challenges, we anticipated that the benefit of this technique would be especially pronounced. Our results show that NORDIC denoising improves the detection sensitivity and the reliability of estimates in submillimeter auditory fMRI data. These effects can be explained by the reduction of the noise-induced signal variability. However, we also observed a reduction in the average response amplitude (percent signal), which may suggest that a small amount of signal was also removed. We conclude that, while evaluating the effects of the signal reduction induced by NORDIC may be necessary for each application, using NORDIC in high resolution auditory fMRI studies may be advantageous because of the large reduction in variability of the estimated responses.

neuroscience↗
bioRxiv2022

Voxelwise encoding models of body stimuli reveal a representational gradient from low-level visual features to postural features in extrastriate body area.

Previous research has focused on the role of the extrastriate body area (EBA) in category-specific body representation, but the specific features that are represented in this area are not well understood. This study used ultra-high field fMRI and banded ridge regression to investigate the coding of body images by comparing the performance of three encoding models in predicting brain activity in ventral visual cortex and specifically the EBA. Our results suggest that EBA represents body stimuli based on a combination of low-level visual features and postural features. Author SummaryHistorically, research on body representation in the brain has focused on category-specific representation, using fMRI to investigate the most posterior body selective region, the extrastriate body area (EBA). However, the role of this area in body perception is still not well understood. This study aims to clarify the role of EBA, in coding information about body images. Using ultra-high field neuroimaging (fMRI) and advanced encoding techniques we tested different computational hypotheses to understand how body images are represented in EBA. Our results suggest that EBA represents bodies using a combination of low-level properties and postural information extracted from the stimulus.

neuroscience↗
bioRxiv2022

Co-fluctuations of neural activity define intra-V1 networks related to perceptual organization

Using functional resonance imaging (fMRI), we studied the relationship between perceptual organization and network topology within the primary visual cortex (V1). Twenty-six humans (male and female) were recorded during active observation of two Global and two Local Navon letters. Correlations between fMRI fluctuations from different V1 sites were measured (after removing stimulus-evoked signals) in windows specific to each condition. Intra-V1, like brain-wide networks, presented an overall decrease of correlations during stimulation compared to baseline and increased statistical dimensionality. Massive edgewise testing and network based-statistics (both corrected by FDR) identified differences between conditions of connection strengths that were mapped to the visual field. Global letters elicited long links mainly connecting V1 sites mapping the lower left/right visual quadrants. Shorter links were associated with Local letters, primarily mapped within the lower-left visual quadrant. Frequently link lengths exceeded V1 population receptive field sizes. These connections were not observed in the time-locked (feedforward) responses shared across participants. Thus, these networks reflect activity idiosyncratic to each participant, possibly generated by interactions within or feedback to V1. Perception would sculpt V1 connectivity, with specific increases in link strengths (in a background of decreases). These findings could help shed light on V1 as a "cognitive blackboard".

neuroscience↗
bioRxiv2019

Objects seen as scenes: neural circuitry for attending whole or parts

Depending on our goals, we pay attention to the global shape of an object or to the local shape of its parts, since its difficult to do both at once. This typically effortless process can be impaired in disease. However, it is not clear which cortical regions carry the information needed to constrain shape processing to a chosen global/local level. Here, novel stimuli were used to dissociate functional MRI responses to global and local shapes. This allowed identification of cortical regions containing information about level (independent from shape). Crucially, these regions overlapped part of the cortical network implicated in scene processing. As expected, shape information (independent of level) was mainly located in category-selective areas specialized for object- and face-processing. Regions with the same informational profile were strongly linked (as measured by functional connectivity), but were weak when the profiles diverged. Specifically, in the ventral-temporal-cortex (VTC) regions favoring level and shape were consistently separated by the mid-fusiform sulcus (MFS). These regions also had limited crosstalk despite their spatial proximity, thus defining two functional pathways within VTC. We hypothesize that object hierarchical level is processed by neural circuitry that also analyses spatial layout in scenes, contributing to the control of the spatial-scale used for shape recognition. Use of level information tolerant to shape changes could guide whole/part attentional selection but facilitate illusory shape/level conjunctions under impoverished vision. Significance statementOne daily engages hierarchically organized objects (e.g. face-eyes-eyelashes). Their perception is commonly studied with global shapes composed by of local shapes. Seeing shape at one level is easy, but difficult for both at once. How can the brain guide attention to one level? Here using novel stimuli that dissociate different levels over time and examining local patterns of brain-activity, we found that the level and shape of visual objects were represented into segregated sets of cortical regions, each connected into their own pathway. Level information was found in part of the cortical network known to process scenes. Coding of object-level independently from shape could participate in guiding sustained attention within objects, eliminating interference from irrelevant levels. It could also help produce "illusory conjunctions" (perceptual migration of a shape to the wrong level) when attention is limited. HighlightsO_LIModified Navon figures allow dissociation in time of fMRI responses for the global/local levels. C_LIO_LIShape-invariant hierarchical level information was found in scenes selective areas, whereas level-invariant shape information was found in object- and faces- selective areas. C_LIO_LILevel and shape regions were divided by the mid-fusiform sulcus (MFS) in VTC cortex, and each type of region connected into its own pathway. C_LIO_LIHaving separate level/shape pathways could facilitate selective-attention, but foster illusory conjunctions. C_LI

neuroscience↗
bioRxiv2018

Methods for computing the maximum performance of computational models of fMRI responses

Computational neuroimaging methods aim to predict brain responses (measured e.g. with functional magnetic resonance imaging [fMRI]) on the basis of stimulus features obtained through computational models. The accuracy of such prediction is used as an indicator of how well the model describes the computations underlying the brain function that is being considered. However, the prediction accuracy is bounded by the proportion of the variance of the brain response which is related to the measurement noise and not with the stimuli (or cognitive functions). The bound to the performance of a computational model to the prediction of brain responses has been referred to as the noise ceiling. In previous neuroimaging applications two methods have been proposed for estimating the noise ceiling based on either a split-half procedure or Monte Carlo simulations. These methods make different assumptions over the nature of the effects underlying the data, and, importantly, their relation has not been clarified yet. Here, we use a two-level generative framework to formally describe the partition between the variance of measurement noise and the stimulus related variance. In this framework we derive an analytical form for the noise ceiling that does not require computationally expensive simulations or a splitting procedure that reduce the amount of data. We describe the relation between the newly introduced noise ceiling estimator and the previous methods for variable levels of measurements noise using simulated data. Additionally, as the relation to the noise ceiling is used to make conclusions on the validity of a model with respect to others, we evaluate the effect the interplay between regularization (often used to estimate model fits to the data when the number of computational features in the model is large) and model complexity on the performance with respect to the noise ceiling. Finally, we show the differences between the methods on real fMRI data acquired at 7 Tesla. We demonstrate that while the split half estimator provides a pessimistic estimate of the noise ceiling due to the small amount of data available in conventional fMRI datasets, the parametric nature of the Monte Carlo estimator results in overly optimistic estimates. For this reason, for real data, we propose a robust procedure to the estimation of the noise ceiling based on bootstraps.\n\nAuthor SummaryEncoding computational models in brain responses measured with fMRI allows testing the algorithmic representations carried out by the neural population within voxels. The accuracy of a model in predicting new responses is used as a measure of the brain validity of this model, but the result of this analysis is determined not only by how precisely the model describes the responses but also by the quality of the data. In this article, we validate existing approaches to estimate the best possible accuracy that any computational model can achieve conditioned to the amount of measurement noise that is present in the experimental data (i.e. the noise ceiling). Additionally we introduce a close form estimation of the noise ceiling that does not require computationally or data expensive procedures. All the methods are compared using simulated and real fMRI data. We draw conclusions over the impact of regularisation procedures and model complexity and make practical recommendations on how to report the results of computational models in neuroimaging.

neuroscience↗