Optimal decoding of correlated neural populationresponses in the primate visual cortexYuzhi Chen, Wilson S Geisler & Eyal SeidemannEven the simplest environmental stimuli elicit responses in large populations of neurons in early sensory cortical areas. How thesedistributed responses are read out by subsequent processing stages to mediate behavior remains unknown. Here we used voltage-sensitive dye imaging to measure directly population responses in the primary visual cortex (V1) of monkeys performing ademanding visual detection task. We then evaluated the ability of different decoding rules to detect the target from the measuredneural responses. We found that small visual targets elicit widespread responses in V1, and that response variability at distantsites is highly correlated. These correlations render most previously proposed decoding rules inefficient relative to one that usesspatially antagonistic center-surround summation. This optimal decoder consistently outperformed the monkey in the detectiontask, demonstrating the sensitivity of our techniques. Overall, our results suggest an unexpected role for inhibitory mechanismsin efficient decoding of neural population responses.A fundamental feature of mammalian cerebral cortex is its use oforderly topographic maps to represent sensory and motor informa-tion1–3. Because cortical neurons tend to respond to a broad range ofstimuli4or movements5, and because there are generally multipleneurons tuned to the same range of parameters within one corticalcolumn6,7, even the simplest sensory stimulus or motor response elicitsactivity that is distributed over a substantial population of neurons5,8,9.Electrophysiological studies in behaving primates suggest that percep-tual and motor responses are indeed mediated by populations ofneurons rather than by single neurons10–13. These observations raiseseveral fundamental questions: how are stimuli and movementsencoded by neural population responses, what are the optimal strate-gies for decoding (pooling) the population responses, and how efficientare different non-optimal pooling strategies?Several models of neural pooling in the brain have been pro-posed11,14–19. These include monitoring only the most sensitive neu-rons (at the extreme, a single neuron)16, simple averaging over theactive neural population11and weighted summation, where thecontribution of each neuron in the pool is proportional to itssensitivity17or proportional to the parameter value at the peak of itstuning function14,15,18,19.Importantly, evaluating these and other decoding rules has been heldback because of limited experimental techniques for reliably monitor-ing neural population responses. Optical imaging with voltage-sensitivedyes (VSD) measures neural population responses at high spatial andtemporal resolutions20. Only recently, however, has this technique beenapplied successfully to behaving animals21,22. In the current study, weuse for the first time VSD imaging in behaving monkeys to investigatepossible decoding rules for population responses in V1.RESULTSExperimental designTwo monkeys were trained to detect a small oriented visual target,indicating target presence by making a saccadic eye movement to thetarget location as soon as it was detected (Fig. 1a). While the monkeysperformed this task, VSD imaging was carried out through a cranialwindow over V1 (Fig. 1b). Performance in the detection task is likely todepend on neural signals provided by topographic maps in V1 that canbe directly identified by optical imaging23–25. Because V1 is retino-topically organized, information regarding the presence or absence ofthe target is confined to several square millimeters of cortex within V1.Optical imaging allows us to localize this cortical region precisely andvisualize the pattern of population activity within this entire region, inreal-time, as behavior unfolds. Furthermore, in primates, V1 providesthe main source of visual information to other cortical areas, and thus,optical imaging may allow us to visualize most of the information thatis potentially available to subsequent processing stages in our task.However, because VSD signals are likely to be dominated by subthres-hold synaptic activity, it is possible that some of this information is nottransmitted from V1.To evaluate the efficiency of possible decoding mechanisms andto determine the optimal Bayesian decoding strategy, we beganby analyzing in detail the statistical properties of neural popu-lation responses.Statistical properties of V1 population responsesThe major goal in this study was to determine how target-related neuralpopulation responses in V1 could be pooled by subsequent processingstages in order to mediate visual detection. The efficiency of a poolingReceived 11 September; accepted 26 September; published online 22 October 2006; doi:10.1038/nn1792Department of Psychology and Center for Perceptual Systems, 1 University Station, A8000, University of Texas at Austin, Austin, Texas 78712, USA. Correspondence shouldbe addressed to E.S. ([email protected]).1412 VOLUME 9[NUMBER 11[NOVEMBER 2006 NATURE NEUROSCIENCEARTICLES© 2006 Nature Publishing Group http://www.nature.com/natureneurosciencemethod depends on three key properties of V1 population responses:(i) the amplitude and spatial spread of the response, which determinesthe size of the neural population that could contribute to detection;(ii) the variability of the population response, which influences thequality of the signals provided by neurons at each imaging site (a singlepixel or a binned group of pixels); and (iii) the magnitude and extent ofspatial correlations in response variability, which can have a largeimpact on the gain that can be attained by pooling14,26–29. Our first stepwas to examine these three key properties of V1 responses.High-quality VSD responses were recorded in eight experiments(recording sessions) from V1 in two monkeys. We use the results fromone VSD experiment as an illustrative example (Fig. 2). TheVSD response in a small V1 region that corresponds to the targetlocation increased rapidly shortly after stimulus onset (Fig. 2a).Response amplitude decreased and response latency increased as targetcontrast was reduced (Fig. 2b, thick lines). Target-evoked responsescould easily be seen in individual trials (Fig. 2b, thin green lines),indicating that population responses in this small V1 region werehighly reliable.Spread of V1 population responseTo quantify
View Full Document
Unlocking...