Franco Pestilli

Franco Pestilli

Assistant Professor, Psychological and Brain Sciences

  • - -
  • Office Hours
    By Appointment Only


  • Research Associate, Stanford University, 2011-2014
  • Postdoc, RIKEN Brain Science Institute, 2009
  • Postdoc, Columbus University, 2008-2010
  • Ph.D., New York University, 2008
  • M.A., New York University, 2006
  • B.A., University of Rome, La Sapienza, 2001


Human systems neuroscience

Mapping the human brain requires integrating technologies that allow measuring behavior and brain signals evolving over slower and faster time scales. Behavior is organized at multiple time scales. It take years to acquire fluency in non-native languages, but car drivers can reliably pay attention to approaching danger in congested rush-hour freeways in a matter of milliseconds. Such large operational range requires distinct, dedicated brain mechanisms. The mechanisms involved in processing slow-evolving events may depend on both, changes in the activity of populations of neurons as well as changes in the properties of networks of white-matter fascicles. The mechanisms dedicated to fast information processing are believed to be coded in the spikes and synaptic activity of neuronal populations.

I study the brain and psychology of individuals from a systems neuroscience perspective. This means that I am interested in understanding how neurons in different brain areas connect together to form networks. I combine multiple neurobiological measurements to study how the mechanisms of these brain networks determine our vision of the world, values and motivate our behavior. My research relies on a model-based approach. Models are implementations of otherwise abstract theories; they allow for generating specific predictions for human behavior or brain connectivity. Errors in model prediction can be exploited to test and falsify alternative theories. I use neurobiological measurements from two magnetic resonance imaging (MRI) technologies (diffusion and functional MRI). Using these measurements in living human brains I build models that predict either the properties of the brain network of connections (the connectome) or human behavior from brain activity.

Model-based neuroanatomy

Mapping the network of white-matter connections in living human brains. Magnetic resonance diffusion imaging and computational tractography are the only technologies that enable neuroscientists to measure white matter in the living human brain. Diffusion MRI provides a way to measure the changes in brain structures and tissue properties evolving over long time frames-weeks, months and years. In the decade since their development, these technologies revolutionized our understanding of the importance of white-matter for health and disease. Prior to these measurements, the white matter was thought of as a passive cabling system. But modern measurements show that white matter axons and glia respond to experience and that the tissue properties of the white matter are transformed during development and following training. The white matter pathways comprise a set of active wires and the responses and properties of these wires predict human cognitive and emotional abilities in health and disease. We can now predict confidently that to crack the neural code in mapping the human brain, neuroscientists will have to develop an account of the connections and tissue properties of these active wires. Diffusion imaging and computational tractography enable investigators to map the network of white matter tracts. Whereas there are many impressive findings using this technology, it is widely agreed that there is an urgent need to keep developing and improving tractography methods.

Evaluation and statistical inference in living connectomes. At Stanford University, I developed a technology, called LiFE - Linear Fascicle Evaluation, to perform both tractography validation and statistical testing on the network of brain connections. Previous validation methods are extrinsic. Investigators have demonstrated that tractography algorithms generate reasonable estimates by checking in phantoms and ex-vivo tissue. These methods do not take into account the quality of a specific tractography solution (the connectome) obtained from a specific group of subjects and a specific instrument. Extrinsic validations ask us to believe a tractography solution obtained from a child with a 3T GE scanner based on a validation carried out on a ex-vivo macaque brain from a 7T Siemens scanner. LiFE can be applied to the data at hand rather than appealing to validations based on very different datasets-I call this method native validation. I utilize Big Data and machine-learning methods to treat connectomes as models of the measured white-matter signal. The model generates a prediction of the measured diffusion signal. The model prediction-error is used to identify the network of brain connections best supporting the measured diffusion signal by rejecting false connections. The technology a principled way to test hypotheses about the geometric structure of white-matter fascicles and to compare the accuracy of different connectomes-I call the method virtual lesion. These method can be applied to any type of diffusion data in healthy and diseased populations. These new methods improve current techniques in fundamental ways. A full open-source software implementation of the method is available here.

Mechanisms of motivation and attention

Predicting perceptual decision-making from human brain activity. Functional MRI provides neurobiological measurements on the neuronal, synaptic and neuromodulatory mechanisms evolving over fast time scales-seconds, minutes and days. Fast information processing is important for goal-oriented behavior and decision-making. Healthy brains implement several mechanisms to quickly and reliably associate sensory inputs with internal goals and behavioral choices. Attention is an umbrella-term we use to address experimentally the set of brain mechanisms that deal with the selection of information. There are several types of attention. For example, primates can attend to a location in space, an object or feature-such as a color. During graduate school I studied the behavioral changes in contrast and visual acuity thresholds elicited by exogenous attention. This is the type of attention car drivers use to detect approaching danger. Contrast and acuity are fundamental processes that happen early on in the visual analysis. They are necessary for most subsequent perceptual processes; stimuli below acuity- or contrast-threshold are invisible to us. In a series of articles I showed that when attention is attracted to a location in the sensory scene, visibility improves. Concurrently to improved vision at attended locations, both contrast sensitivity and acuity are impaired at locations away from attended ones (Pestilli & Carrasco Vision Research 2005; Pestilli, Viera & Carrasco Journal of Vision 2007; Montagna, Pestilli & Carrasco Vision Research 2009; Pestilli, Ling & Carrasco Vision Research 2009). These behavioral trade-offs have matched effects on cortical activity, whereby the fMRI response in early visual cortex increases at attended and decreases at unattended locations (Liu, Pestilli & Carrasco Neuron 2005).

Attention and visual selection. A fundamental endeavor of neuroscience is to identify the neural mechanisms supporting healthy human decision-making. A successful way to embark in such an effort is by using computational methods to predict decision-making from brain activity. The challenge of such an approach is that it requires concurrent measurements of cortical activity and behavior in tasks for which models can be formulated to quantitatively predict the two measurements. When Gustav Fechner initiated what we call Psychophysics, he had specifically this problem in mind. Fechner had no access to measurements of brain activity, which he replaced with the psychological quantity just-noticeable-difference (JND). Recently, I used the very idea of JND in combination with measurements of brain activity (fMRI) and computational modeling to test the predictions of three alternative models of attentional enhancement. We demonstrated that improvements in behavioral sensitivity with attention are fully accounted for by a model of visual selection (a gating mechanism that pools sensory signals across the visual scene) that does not requires a change in quality of the sensory representation (Pestilli et al., Neuron 2011; Gardner, Hara and Pestilli, Frontiers in Computational Neuroscience in review).

Mechanisms of value processing and motivated behavior. Many situations require selecting one among several options of different value. In these situations valuation and preference drive motivation and behavior. Although attention allows for selecting task-relevant information out of the sensory scene, many decisions depend on the subjective value assigned to alternative options. About one hundred years before Fechner, Daniel Bernoulli (1700-1782) suggested that valuation conforms to a psychological rule akin to the JND proposed by Fechner for sensory Psychophysics. Bernoulli proposed that humans decide among alternatives by using internal utility not extrinsic value. Just like for the JND, utility is equal for alternatives with values of equal ratio. Bernoulli's Utility Theory held still for more than two-hundred years until Amos Tversky and Daniel Kahneman in the 1980s showed that the theory has flaws. They proposed the Prospect Theory, suggesting that utility is not absolute, but depends on the prospective of the decision-maker. Critically, they demonstrated that humans are loss averse; utility is higher for losses than gains of identical absolute value [11]. We investigated the effects of loss aversion, expecting gains versus losses, on perceptual decision-making and cortical activity in early visual cortex (Pestilli, Khan & Ferrera, Society for Neuroscience 2011). We measured fMRI response in early visual cortex concurrently with contrast discrimination thresholds for stimuli of different expected value. We used a computational model to manipulate task difficulty and expected value on a trial-by-trial base. The model determined that on half the trials observers had to perform a perceptual judgment expecting to win. On the other half of the trials subjects performed an identical judgment but expecting to lose. Both perceptual judgments, stimuli and absolute values were identical for gain- and loss-trials. We found that behavioral sensitivity improved and cortical response increased with value in early visual cortex (V1, V2, V3, hV4). The increase followed the log-expected absolute value but was, independent on whether the expected outcome was a loss or a gain. This is predicted by the Utility Theory. Most importantly, cortical responses in early visual cortex increase more for loss-predicting stimuli than for gain-predicting stimuli of equal value: an effect predicted only by the Prospect Theory (Pestilli, Khan and Ferrera in preparation).