{"title": "Dual Inhibitory Mechanisms for Definition of Receptive Field Characteristics in a Cat Striate Cortex", "book": "Advances in Neural Information Processing Systems", "page_first": 75, "page_last": 82, "abstract": null, "full_text": "Dual Inhibitory Mechanisms for  Definition of \nReceptive Field Characteristics in Cat  Striate \n\nCortex \n\nA.  B.  Bonds \n\nDept.  of Electrical  Engineering \n\nVanderbilt  University \nN ashville,  TN 37235 \n\nAbstract \n\nIn  single  cells  of the  cat  striate  cortex,  lateral  inhibition  across  orienta(cid:173)\ntion  and/or spatial frequency  is  found  to enhance  pre-existing  biases.  A \ncontrast-dependent but spatially non-selective inhibitory component is also \nfound.  Stimulation with  ascending  and  descending  contrasts  reveals  the \nlatter  as  a  response  hysteresis  that  is  sensitive,  powerful  and  rapid,  sug(cid:173)\ngesting that it is  active in day-to-day vision.  Both forms of inhibition are \nnot  recurrent  but  are  rather  network  properties.  These  findings  suggest \ntwo  fundamental  inhibitory mechanisms:  a  global mechanism that limits \ndynamic range  and  creates  spatial selectivity  through  thresholding  and a \nlocal mechanism that specifically refines  spatial filter  properties.  Analysis \nof burst  patterns in spike  trains demonstrates that these  two mechanisms \nhave  unique  physiological origins. \n\n1 \n\nINFORMATION PROCESSING IN STRIATE \nCORTICAL  CELLS \n\nThe  most  popular  current  model  of single  cells  in  the  striate  cortex  casts  them \nin  terms  of spatial  and  temporal  filters.  The  input  to  visual  cortical  cells  from \nlower  visual areas,  primarily the LGN, is fairly broadband (e.g., Soodak, Shapley & \nKaplan,  1987;  Maffei  & Fiorentini,  1973).  Cortical  cells  perform significant  band(cid:173)\nwidth restrictions on this information in at least three domains:  orientation, spatial \nfrequency  and  temporal frequency.  The  most  interesting  quality  of these  cells  is \n75 \n\n\f76 \n\nBonds \n\ntherefore what they reject from the broadband input signal, rather than what they \npass, since the mere passage of the signal adds no information.  Visual cortical cells \nalso  show  contrast-transfer,  or  amplitude-dependent,  nonlinearities  which  are  not \nseen  at lower levels in the visual pathway.  The primary focus  of our lab is study of \nthe  cortical mechanisms that support both the band limitations and nonlinearities \nthat are imposed on the  relatively unsullied signals incoming from the LGN. All of \nour work  is  done on the cat. \n\n2  THE ROLE  OF  INHIBITION IN  ORIENTATION \n\nSELECTIVITY \n\nOrientation selectivity  is  one  of the most dramatic demonstrations of the filtering \nability of cortical  cells.  Cells  in  the  LGN  are  only  mildly biased  for  stimulus ori(cid:173)\nentation, but cells  in cortex are  completely unresponsive  to orthogonal stimuli and \nhave  tuning bandwidths  that average  only  about  40-50\u00b0  (e.g.,  Rose  &  Blakemore, \n1974).  How  this happens remains controversial, but there is  general consensus  that \ninhibition  helps  to  define  orientation selectivity  although  the  schemes  vary.  The \nconcept  of cross-orientation  inhibition suggests  that the inhibition is  itself orienta(cid:173)\ntion selective and tuned in a complimentary way to the excitatory tuning of the cell, \nbeing smallest at the optimal orientation and greatest at the orthogonal orientation. \nMore recent  results,  including those from our own lab, suggests that this is  not the \ncase. \nWe  studied  the  orientation  dependence  of inhibition  by  presenting  two  superim(cid:173)\nposed  gratings,  a  base  grating at the  optimal orientation to provide  a steady  level \nof background  response  activity,  and  a  mask grating  of varying orientation  which \nyielded  either excitation or inhibition that could supplement or suppress  the  base(cid:173)\ngenerated  response.  There  is  some  confusion  when  both  base  and  mask  generate \nexcitation.  In order to separate the response components from each of these stimuli, \nthe  two  gratings were  drifted  at  differing  temporal frequencies.  At  least  in simple \ncells,  the individual contributions to the  response  from  each  grating could  then be \nresolved  by  performing Fourier analysis on the  response  histograms. \nExperiments were done on 52 cells, of which about 2/3 showed organized suppression \nfrom  the  mask grating  (Bonds,  1989).  Fig.  1 shows  that while  the  mask-generated \nresponse suppression is somewhat orientation selective, it is  by and large much flat(cid:173)\nter than would be required  to account for  the tuning of the cell.  There is thus  some \norientation dependence of inhibition, but not specifically at the orthogonal orienta(cid:173)\ntion as might be expected.  Instead,  the predominant component of the suppression \nis  constant  with  mask  orientation,  or  global.  This  suggests  that  virtually  any \nstimulus  can  result  in inhibition,  whether  or  not  the  recorded  cell  actually  \"sees\" \nit.  What orientation-dependent  component of inhibition that might appear is  ex(cid:173)\npressed  in suppressive  side-bands near the limits of the excitatory tuning function, \nwhich  have the effect  of enhancing any pre-existing orientation bias. \nThus  the  concept  of cross-orientation  inhibition  is  not  particularly  correct,  since \nthe  inhibition is  found  not just  at  the  \"cross\"  orientation  but  rather  at  all  orien(cid:173)\ntations.  Even  without  orientation-selective  inhibition,  a  scheme  for  establishment \nof true  orientation selectivity  from  orientation-biased  LGN  input  can  be  derived \n\n\fDual Inhibitory Mechanisms \n\n77 \n\n70 . - - - - - - - - - - - - ,  ~ 0 , . . - - - - - - - - - - - ,  \n\nA  ~ \n15. \nE  -5 \n\n14%mask __ \n\n28%1NIIk--\n\nB \n\nNo INIIk (luning) \u00b7\u00b70\u00b7\u00b7 \n~ 80  14% mask  - -\nIi \n28% mask  ___ \na.  50 \nE \n:~ \n! \n&~ \nI  --~~---~~r_ \n~20 \nu \nci  10 \n\nc;. \n\n:I \u00b710 \n\n& \n\nI~ \n~ \n\no~~-~-~-~-~ \n\n80 \n\n80 \n\n100 \n\n120 \n\n0.------------------------, \n\n2 Hz (*e) rwp.  -----(cid:173)\nSupprWIion \n\nNo mask (luning) \nExc:iWtion \n\nc \n\u00b7\u00b7\u00b70\u00b7\u00b7\u00b7\u00b7\u00b7\u00b7 \n\nSv12:A4.D11 \nSimple \n\n330 \n\n~ \n\n90 \n210 \nMask Orientation (deg) \n\n150 \n\n270 \n\n330 \n\nFigure 1:  Response suppression by mask gratings of varying orientation.  A.  Impact \nof masks  of 2  different  contrasts  on  2  Hz  (base-generated)  response,  expressed  by \ndecrease  (negative  imp/sec)  from  response  level  arising from  base  stimulus alone. \nB.  Similar example for  mask orientations spanning a full  3600 \u2022 \n\nby assuming that the nonselective inhibition is  graded and contrast-dependent  and \nthat it acts as a  thresholding device  (Bonds,  1989). \n\n3  THE ROLE  OF INHIBITION IN SPATIAL \n\nFREQUENCYSELECTnnTY \n\nWhile most retinal and  LGN  cells  are  broadly tuned  and predominantly low-pass, \ncortical cells generally have spatial frequency bandpasses of about 1.5-2 octaves (e.g., \nMaffei &  Fiorentini, 1973).  We  have examined the influence of inhibition on spatial \nfrequency selectivity using the same strategy as the previous experiment (Bauman & \nBonds, 1991).  A base grating, at the optimal orientation and spatial frequency,  drove \nthe cell, and a superimposed mask grating, at the optimal orientation but at different \nspatial and temporal frequencies,  provided response  facilitation or suppression. \n\nWe  defined  three broad categories of spatial frequency  tuning functions:  Low  pass, \nwith no  discernible  low-frequency fall-off,  band-pass,  with  a  peak  between  0.4  and \n0.9  c/deg,  and  high  pass,  with  a  peak  above  1  c/deg.  About  75%  of the  cells \nshowed  response  suppression  organized  with  respect  to  the  spatial  frequency  of \nmask gratings.  For  example,  Fig.  2A  shows  a  low-pass  cells  with  high-frequency \nsuppression  and  Fig.  2B  shows  a  band-pass  cell  with  mixed suppression,  flanking \nthe tuning curve at both low and high frequencies.  In each case response suppression \nwas graded with mask contrast and some suppression was found even at the optimal \nspatial frequency.  Some cells showed no suppression, indicating that the suppression \nwas  not  merely  a  stimulus artifact.  In all  but  2  of 42  cases,  the  suppression  was \nappropriate to the enhancement of the tuning function (e.g., low-pass cells had high(cid:173)\nfrequency  response  suppression),  suggesting  that the  design  of the  system  is  more \n\n\f78 \n\nBonds \n\nthan  coincidental.  No  similar spatial-frequency-dependent  suppression  was  found \nin LGN  cells. \n\n~.-----------------------, \n\nNo mask (tuning) ......  A \n14'lCtmask  - -\n2O'lCt  mask  ---\n28'lCt mask  ---\n\n40 o 0)30 \n..!!? \n~20 \nc. \n0)  10 \na 0 ~---------\" -,'  __  ~ ___ -e--I \n(I) \n110 \n\nSimple \nLV9:R4.05 \n\n-20 \n\n40  10'lCt mask ---\n\nNo mask (tuning)  ....... . \n\n~.-----------------------, \n\n20 === :// .. ... \\'\\'\" =\" \n\nB \n\n.. \n\nO~~==~----------~==~~ \n\n\u2022. . 0  .\u2022 . -- 0 ' \n\n-20 \n\n~~~--~--~----~----~~ 4D~~--~--~----~----~~ \n\n0.2 \n\n0.3 \n\n0.5 \n\n1 \n\n2 \n\n0.2 \n\n0.3 \n\n0.5 \n\n1 \n\n2 \n\nSpatial Frequency (cyc/deg) \n\nFigure  2:  Examples  of spatial frequency-dependent  response  suppression.  Upper \nbroken  lines  show  excitatory  tuning  functions  and  solid  lines  below  zero  indicate \nresponse reduction at three different contrasts.  A.  Low-pass cell with high frequency \ninhibition.  B.  Band-pass cell with mixed (low and high frequency)  inhibition.  Note \nsuppression  at optimal spatial frequency  in both cases. \n\n4  NON-STATIONARITY OF CONTRAST TRANSFER \n\nPROPERTIES \n\nThe two experiments described above demonstrate the existence of intrinsic cortical \nmechanisms that refine  the spatial filter  properties  of the  cells.  They also  reveal  a \nglobal form  of inhibition that is  spatially non-specific.  Since  it  is  found  even  with \nspatially optimal stimuli, it can influence  the form of the cortical contrast-response \nfunction  (usually measured with optimal stimuli).  This function is  essentially loga(cid:173)\nrithmic, with saturation or even super-saturation at higher contrasts (e.g., Albrecht \n&  Hamilton,  1982),  as  opposed  to  the more  linear  response  behavior seen  in  cells \nearlier in the visual pathway.  Cortical cells also show some degree of contrast adap(cid:173)\ntation; when  exposed  to high mean contrasts for  long periods of time, the response \nvs  contrast  curves  move  rightward  (e.g.,  Ohzawa,  Sclar  &  Freeman,  1985).  We \naddressed  the  question  of whether  contrast-response  nonlinearity  and  adaptation \nmight be causally related. \n\nIn  order  to compensate  for  \"intrinsic  response  variability\"  in  visual  cortical  cells, \nexperimental stimulation  has  historically  involved  presentation  of randomized  se(cid:173)\nquences  of pattern  parameters,  the so-called  multiple histogram technique  (Henry, \nBishop,  Tupper  & Dreher,  1973).  Scrambling presentation  order  distributes  time(cid:173)\ndependent response variability across all stimulus conditions, but this procedure can \nbe self-defeating by masking any stimulus-dependent response  variation.  We  there(cid:173)\nfore presented cortical cells with ordered sequences of contrasts, first  ascending then \ndescending in a stepwise manner (Bonds,  1991).  This revealed a  clear and powerful \nresponse  hysteresis.  Fig.  3A shows  a  solid line  representing  the  contrast-response \n\n\fDual Inhibitory Mechanisms \n\n79 \n\nfunction measured in the usual way,  with randomized parameter presentation, over(cid:173)\nlaid on an envelope outlining responses  to sequentially increasing or decreasing 3-sec \ncontrast  epochs;  one  sequential  presentation  set  required  54  secs.  Across  36  cells \nmeasured in this same way, the average response hysteresis corresponded  to 0.36 log \nunits of contrast.  Some hysteresis  was found  in every  cortical  cell  and  in no LGN \ncells,  so this phenomenon is  intrinsically cortical. \n\n50 \n\n-\n-CD \n\n0 \nCD  40 \n.!e \na. \nE \n._  30 \n\n~20 \n0 a. \nen  10 \nCD \na: \n\n8 \n\nComplex \n\n14% peak contr. \n\n100 \n\n80 \n\n60 \n\n40 \n\n20 \n\n0 \n\n3 \n\n5 \n\n10 \n\n2030 \n\n0 \n\n50 \n\n100 \n\n3 \nContrast (%) \n\n5 \n\n10 \n\nFigure  3:  Dynamic response  hysteresis.  A.  A  response  function  measured  in  the \nusual  way,  with  randomized  stimulus  sequences  (filled  circles)  is  overlaid  on  the \nfunction  resulting  from  stimulation  with  sequential  ascending  (upper  level)  and \ndescending  (lower  level)  contrasts.  Each  contrast  was  presented  for  3 seconds.  B. \nHysteresis  resulting from peak contrast of 14%;  3 secs  per datum. \n\nHysteresis  demonstrates  a  clear  dependence  of response  amplitude on  the  history \nof stimulation:  at a  given contrast, the amplitude is always less  if a higher contrast \nwas  shown  first.  This  is  one  manifestation of cortical  contrast  adaptation,  which \nis well-known.  However,  adaptation is  usually measured after long periods of stim(cid:173)\nulation with high  contrasts,  and may not  be  relevant  to normal behavioral vision. \nFig.  3B  shows  hysteresis  at a  modest  response  level  and low  peak  contrast  (14%), \nsuggesting  that it  can serve  a  major function  in  day-to-day visual processing.  The \nspeed  of hysteresis  also addresses  this issue,  but it is  not so easily measured.  Some \nresponse  histogram  waveforms show  consistent  amplitude loss  over  a  few  seconds \nof stimulation (see  also Albrecht,  Farrar &  Hamilton,  1984),  but other  histograms \ncan  be  flat  or  even  show  a  slight  rise  over  time  despite  clear  contrast  adaptation \n(Bonds,  1991) .  This  suggests  the  possibility  that,  in  the  classical  pattern  of any \nwell-designed  automatic gain control,  gain reduction  takes  place  quite rapidly,  but \nits effects  linger for  some time. \nThe  speed  of reaction  of gain  change  is  illustrated  in  the  experiment  of Fig.  4. \nA  \"pedestal\"  grating  of 14%  contrast  is  introduced.  After  500  msec,  a  contrast \nincrement of 14% is added to the pedestal for a variable length of time.  The response \nduring  the  first  and last 500  msec  of the  pedestal  presentation  is  counted  and  the \nratio is  taken.  In the absence of the increment, this ratio is  about 0.8, reflecting  the \nadaptive nature of the  pedestal  itself.  For  an increment of even  50  msec  duration, \nthis  ratio  is  reduced,  and  it  is  reduced  monotonically-by  up  to  half the  control \n\n\f80 \n\nBonds \n\nlevel-for increments lasting less  than a second.  The gain control mechanism is  thus \nboth sensitive  and rapid. \n\n1 \n\n0.8 \n\n0.6 \n\n0 \ne. \nCD \n't:J \n::l s:: \nQ. \nE \n0( \n't:J \nCD \n.!::! \nIii \n~  0.2 \n0 z \n\n0.4 \n\n\"Blip\u00b7 \n\n28%~ \n\n14'li.l.1 \n\n\"Probe\" \n\nt2  L \n\n0.0  0.5  1.0  1.5  2.0 \n\nTime (sec) \n\nCV9:L11.06-7 \n\nNorm. Ampl. = spikes (t2)/spikes(t1) \n\n0 \n\n0 \n\n0.2 \n\n0.4 \n\n0.6 \nBlip Duration (sec) \n\n0.8 \n\nFigure 4:  Speed of gain reduction.  The ratio of spikes generated during the last and \nfirst  500  msec of a  2 sec  pedestal  presentation  can  be  modified  by  a  brief contrast \nincrement  (see  text). \n\n5  PHYSIOLOGICAL INDEPENDENCE OF TWO \n\nINHIBITORY MECHANISMS \n\nThe  experimental  observations  presented  above  support  two  basic  phenomena: \nspatially-dependent  and  spatially-independent  inhibition.  The  question  remains \nwhether  these  two  types  of inhibition are  fundamentally different,  or  if they  stem \nfrom the same physiological mechanisms.  This question can be addressed  by exam(cid:173)\nining  the  structure  of a  serial  spike  train  generated  by  a  cortical  cell.  In general, \nrather than being distributed continuously, cortical spikes are grouped into discrete \npackets,  or  bursts,  with  some intervening isolated spikes.  The burst structure  can \nbe fundamentally characterized  by two  parameters:  the burst frequency  (bursts per \nsecond,  or  BPS)  and the burst duration (spikes  per  burst,  or SPB). \n\nWe  have  analyzed  cortical  spike  trains  for  these  properties  by  using  an  adaptive \nalgorithm to  define  burst  groupings;  as  a  rule  of thumb, spike  intervals of 8  msec \nor less  were  considered  to belong to bursts.  Both burst frequency  (BPS)  and struc(cid:173)\nture  (SPB)  depend  strongly  on  mean firing  rate,  but  once  firing  rate  is  corrected \nfor,  two  basic patterns emerge.  Consider two experiments,  both yielding firing  rate \nvariation  about  a  similar range.  In  one  experiment,  firing  rate  is  varied  by  vary(cid:173)\ning stimulus  contrast,  while  in  the other,  firing  rate  is  varied  by  varying stimulus \norientation.  Burst  frequency  (BPS)  depends  only on  spike  rate,  regardless  of the \ntype of experiment.  In Fig.  5A,  no systematic difference  is seen  between  the exper(cid:173)\niments in which  contrast  (filled  circles)  and orientation  (open  squares)  are  varied. \nTo quantify the difference  between  the curves,  polynomials were  fit  to each  and the \nquantity  gamma,  defined  by  the  (shaded)  area  bounded  by  the  two  polynomials, \nwas  calculated; here,  it equalled  about 0.03. \n\n\f16,----------------------------. \n\nCI) \n\n0' 14  Gamma: 0.0290 \nQ) \n~ 12 \n~ 10 \n::J \ne?.8 \n1U \n6 \na: \n1;) \n\nQ) \n\n4 \n\n_ \n\n~ \n\n:J  2 \nIII \n\nA \n\n\u2022  \u2022 \n\n\u2022 \n\nVariation of stimulus contrast \n\n--tl- Variation of stimulus orientation \n\nDual Inhibitory Mechanisms \n\n81 \n\nGamma: 0.2485 \n\nB \n\u2022 \n\u2022 \n\n\u2022 \n\u2022 \n\n\u2022 \n\u2022 \n\n\u2022 \u2022 \n\n,.-...  3.6 \nCI) \nQ)  3.4 \n~ \n'0..  3.2 \n\nCJ) -\n... \n~ \n::J m \nQ) a. \n\n~ \n\n3.0 \n\n2.8 \n\n2.6 \n\nCI) \n2.4 \nQ) \n~ \n.0..  2.2 \nCJ) \n\n2.0 \n\n60 \n\n10 \nResponse (imp/sec) \n\n0 \n\nD \n\n_ \n\nVariation of stimulus contrast \n\n-t:r- Variation of stimulus orientation \n\n60 \n\nFigure 5:  A. Comparison of burst frequency  (bursts per second) as function offiring \nrate  resulting  from  presentations  of varying  contrast  (filled  circles)  and  varying \norientation (open squares).  B.  Comparison of burst length (spikes per burst) under \nsimilar conditions.  Note  that at  a  given  firing  rate,  burst  length  is  always  shorter \nfor  experiment  parametric  on  orientation.  Shaded  area  (gamma)  is  quantitative \nindicator of difference  between  two curves. \n\nFig.  5B  shows  that  at similar firing  rates,  burst  length  (SPB)  is  markedly shorter \nwhen  firing  rate  is  controlled  by  varying  orientation  (open  squares)  rather  than \ncontrast  (filled  circles).  In this pair of curves,  the gamma (of about  0.25)  is  nearly \nten times that found in the upper curve.  This is a clear violation ofunivariance, since \nat a given spike rate (output level) J  the structure of the spike train differs depending \non  the  type  of stimulation.  Analysis  of cortical  response  merely  on  the  basis  of \noverall firing rate thus does  not give the signalling mechanisms the respect  they are \nproperly due.  This result also implies that the strength of signalling between nerve \ncells  can  dynamically  vary  independent  of firing  rate.  Because  of post-synaptic \ntemporal integration,  bursts  of spikes  with  short  interspike  intervals  will  be  much \nmore  effective  in  generating  depolarization  than  spikes  at  longer  intervals.  Thus, \nat  a  given  average  firing  rate,  a  cell  that  generates  longer  bursts  will  have  more \ninfluence on a  target  cell than a  cell  that distributes its spikes in shorter bursts,  all \nother factors  being equal. \n\nThis  phenomenon  was  consistent  across  a  population of 59  cells.  Gamma, which \nreflects  the  degree  of difference  between  curves  measured  by  variation of contrast \nand by variation of orientation, averaged zero for  curves  based on number of bursts \n(BPS).  For  both simple and  complex cells,  gamma for  burst  duration  (SPB)  aver(cid:173)\naged  0.15. \n\nAt  face  value,  these  results  simply mean that when  lower  spike  rates  are  achieved \nby use of non-optimal orientations, they result from shorter bursts than when lower \nspike rates result from reduction of contrast (with the spatial configuration remain(cid:173)\ning optimal). This means that non-optimal orientations and, from some preliminary \nresults,  non-optimal spatial frequencies,  result in inhibition that acts specifically to \nshorten  bursts,  whereas  contrast  manipulations for  the  most part  act  to modulate \nboth the  number and length of bursts. \n\n\f82 \n\nBonds \n\nThese results suggest strongly that there are at least two distinct forms of cortical in(cid:173)\nhibition, with unique physiological bases differentiated  by the burst organization in \ncortical spike trains.  Recent  results from our laboratory (Bonds,  Unpub.  Obs.)  con(cid:173)\nfirm that burst length modulation, which seems to reflect inhibition that depends on \nthe spatial characteristics of the stimulus, is strongly mediated by GABA. Microion(cid:173)\ntophoretic injection of GABA  shortens  burst length  and injection of bicuculline,  a \nGABA blocker, lengthens bursts.  This is wholly consistent with the hypothesis that \nGABA is  central to definition of spatial qualities of the cortical receptive field,  and \nsuggests  that one  can indirectly observe  GAB A-mediated inhibition by  spike  train \nanalysis. \n\nAcknowledgements \n\nThis  work  was  done  in  collaboration with  Ed  DeBruyn,  Lisa  Bauman  and  Brian \nDeBusk.  Supported by NIH  (ROI-EY03778-09). \n\nReferences \n\nD. G. Albrecht & D.  B.  Hamilton.  (1982) Striate cortex of monkey and cat:  contrast \nresponse  functions.  Journal  of Neurophysiology 48, 217-237. \nD.  G.  Albrecht,  S.  B.  Farrar & D.  B.  Hamilton.  (1984)  Spatial contrast  adaptation \ncharacteristics ofneurones recorded in the cat's visual cortex.  Journal of Physiology \n347,713-739. \nA.  B.  Bonds.  (1989)  The role of inhibition in  the specification of orientation selec(cid:173)\ntivity of cells of the cat striate cortex.  Visual Neuroscience  2, 41-55. \nA.  B.  Bonds.  (1991)  Temporal dynamics of contrast  gain control  in single  cells  of \nthe cat striate cortex.  Visual  Neuroscience 6, 239-255. \nL.  A.  Bauman & A.  B.  Bonds.  (1991)  Inhibitory  refinement  of spatial frequency \nselectivity in single cells of the cat striate cortex.  Vision  Research 31, 933-944. \nG.  Henry,  P.  O.  Bishop,  R.  M.  Tupper & B.  Dreher.  (1973)  Orientation specificity \nof cells  in  cat striate cortex.  Vision  Research 13,  1771-1779. \nL.  Maffei & A.  Fiorentini.  (1973)  The visual cortex as a spatial frequency  analyzer. \nVision  Research 13, 1255-1267. \n1.  Ohzawa,  G.  Sclar  & R.  D.  Freeman.  (1985)  Contrast  gain  control  in  the  cat's \nvisual system.  Journal  of Neurophysiology  54, 651-667. \nD.  Rose  & C.  B.  Blakemore.  (1974)  An  analysis  of orientation  selectivity  in  the \ncat's visual cortex.  Experimental  Brain Research 20,  1-17. \nR.  E. Soodak, R. M.  Shapley & E.  Kaplan.  (1987) Linear mechanism of orientation \ntuning in  the  retina and  lateral geniculate  of the  cat.  Journal  of Neurophysiology \n58, 267-275. \n\n\f", "award": [], "sourceid": 566, "authors": [{"given_name": "A.", "family_name": "Bonds", "institution": null}]}