Discrimination Performance of Single Neurons: Rate and Temporal-Pattern Information SUMMARY AND CONCLUSIONS 1. A new methodof measuring the performance of neurons in sensory discrimination tasks wasdeveloped and then applied to single-neuron responses recorded in the auditory nerveof chin- chilla and in the striate visualcortexof cat. 2. Most previous methods of measuring discrimination perfor- mancehavi employed decision rulesthat involvecomparing the total counts ofaction potentials (spikes) produced by two different stimuli. Such measures ignore response patternand hence may not reflect all the informationtransmitted by a neuron'The pro- posed method attempts to measure all (or most) of the transmitted information by constructing descriptive modelsof the neuron's response to eaih stimulus in the discrimination experiment; these descriptivemodels consistof measured probability distributions of the spikecountsin smalltime bins.The measured probability distribuiions are then used to define an optimal decision rule (an icleal observer) for discriminating thetwo stimuli' Finally, discnm- ination performance is measured by applying this decision rule to novel presentations of the same two stimuli. 3. Intensity andtemporal-phase discrimination were measured for three neuions in theauditory nerve of chinchilla' The discrimi- nation stimuli werelow-frequency pure tones of 70-ms duration' Intensity thresholds werefound to be 5-20 dB lowerat low inten- sities usingth e new pattern method compared with the traditional countingmethod. The pattern method led to better performance because it utilized both rate and temporal pattern information' Phase discrimination performance using the countingmethod was at chance because the average spike rate did not changewith phase. On the otherhand, usingthe patternmethod,phase discrim- ination thresholds were found to decrease with intensity,often reaching values equivalent to 30-40 ps oftemporal offset' These thresholds are as good as or better than behavioralthresholds in chinchilla. 4. Contrastand temporal-phase discrimination weremeasured for threeneurons in the striate visualcortexof cat.The discrimina- tion stimuli were drifting sine-wave gratings of 100-to 160-ms duration. Contrastdiscrimination functions measured by the pat- tern method and the countingmethod werefound to be essentially identical. Phase discrimination using the counting method wasat chance. However, using the pattern method, phasethresholds were found to decrease with contrast,reaching valuesequivalent to 7 ms of temporal offsetfor the two simple cells' 5. Our resuftssuggest that temporal response pattern carries substantial informationfor intensity and phase discrimination in theauditorynerveandforphasediscriminationinthestriatevi- sual cortex. There are likely to be other tasks,such as temporal i..qu.n.y discrimination and velocity discrimination, for which co.,sideration of only rateinformationwill substantially underes- timate discrimination performance. JoURNAL OF NEUROPHYSIOLMY Vol. 66. No. I . July- l 99 l Printed in L' S tl WILSON S. GEISLER, DUANE G. ALBRECHT, RICHARD J. SALVI, AND SAMUEL S' SAUNDERS Department of Psychotogy, university of Texas at Austin,Austin, Texas78712;and HearingResearch Lab, State University of l'{ewYork, Buffalo, New York 14214 INTRODU CTION A long-standing questionin neurophysiology concerns the natu-re of the neural code (Perkel and Bullock 1969); specifically, how is information carried in the responses o[ inAiviOuai neurons, and how is this information utilized? Assumptions made regarding the code and how it should be measured can havea significant impact on estimates of the amount of information carried and of the potential levelof neural performance (Mooreet al. 1966). Consider assessing the discriminationcapabilities of a given sensory neuron (e.g., Barlowand Levick l969a,b; Wernerand Mountcastle ieO:;. fo measure the minimum change in the stimulus required to evokea "reliably differentresponse," the sen- sory physiologist is required to define (grvena certain setof expliiit-or impticit assumptions) an index of response (e'g', the meanresponse rate)and an index of reliability (e'g', the trial-to-trial varianceassociated with the mean rate)' If the index ofresponse does not adequately reflect all ofthe rele- vant information, and if the index of reliability doesnot adequately reflect the probabilistic character of the re- sponses, fhen the estimate of performance will be erro- neous. Most previous studies of discrimination performance in single nzuronshavenot attempted to utilize all of the avail- ab[ information. In the vision literature'studies of detec- tion and discrimination performance have usually consid- eredonly the total number of action potentials (spikes) gen- erated in some fixed time period during or after the stimulus presentation (e.g.,Barlow and Levick l969a,b; Barlow etal. l97l;Cohn et al.191S;Fitzhugh 1958; Shap- ley and Victor 1986; Tolhurst et al. 1983)'A similar ap- pioach hasbeentaken by a number qflgarins researchers ie.g., Geisler et al. 1985; Relkin and -Pelli.1987; Sinex and iiit.v 1986; Young and Barta 1986). Although some of thesestudies have demonstrated good discrimination per- formance, they may have underestimated sensitivity be- causethey did not consider the temporal pattern of the ."aponr"t. Single neurons may transmitmuch information in Lrms of a tJmporal code; thus it is important to consider the temporal patiernof responses to stimuli whenassessing discrimination performance.Indeed, there are some dis- crimination tasks, such as temporal-phase, temporal-fre- quency,and velocity discrimination, for which considering .i"iv toiuf spikes wiligreatly underestimate the sensitivity of a neuron. There have been severalattempts to considerthe tem- poral structure of single-neuron responses' Siebert (1970) The American Physiological Society 334 0022-30'77191 $1.50 Copyright O l99l