下側頭回ニューロンによる図形特徴の符号化

 サルの下側頭回前半部は、視覚対象物を知覚・再認する脳過程の、最終段階に関与している。この領域の神経細胞は、第一次視覚野、視覚前野または下側頭回後半部の細胞に比べ、格段に複雑な刺激に選択的に応答する。しかし、複雑といっても、「顔細胞」・「手細胞」を除き、有効視覚刺激は物体を特定するほど複雑ではない。刺激選択性も相対的なものである。にもかかわらず、サルが、様々な物体を見分けることができるのはなぜか。また、これらの刺激選択性はどうやってできあがってくるのだろうか。これらの問いに答えることは、形態知覚・対象認知の脳機構を理解する上での重要なステップである。我々は、最近、下側頭回前半部が、微小な機能的モジュールから成っており、各モジュールに含まれる細胞は、同じカテゴリーに属する図形に共通に反応するが、細かい特徴の違いに対しては、互いに異なった選択性を有することを見いだした(Fujita, Cheng & Tanaka, 1990)。この結果から、我々は「下側頭回における情報符号化の単位は、個々の細胞ではなく、モジュールに含まれる細胞集団ではないか。」という仮説を持つに至り、ここを突破口として上記の問題にアプローチしていきたいと考えている。

VISUAL ANALYSIS OF SHAPE BY INFEROTEMPORAL CORTEX NEURONS

 Neurons in the anterior two-thirds of the monkey inferotemporal cortex (AIT) selectively respond to particular visual images.  Except for a class of neurons selective to biologically significant stimuli such as faces and hands, the stimulus features essential to activate AIT neurons are "moderately complex"-not as simple as slits, edges or spots, but not as complex as particular objects.  Examples of such stimulus features include star-shapes, T-shapes, and polygons filled with stripes.  How is the stimulus selectivity of AIT neurons generated?  How can the AIT neurons code objects more complex than their selectivities? 
  An essential step towards answering these questions is to learn how cells with similar or dissimilar selectivities are spatially arranged in the AIT.  Kang Cheng, Keiji Tanaka and I have recently obtained the evidence listed below which suggests that the AIT is organized in a modular fashion: (1) When we recorded simultaneously from adjacent neurons in the AIT, they responded to similar or related stimuli.  The optimal feature or the degree of tuning was, however, often slightly different among adjacent neurons.  (2) Crosscorrelation analysis showed that adjacent neurons were functionally connected.  (3) When we directed electrodes vertically to the AIT surface, we obtained neurons with similar or related selectivity over a distance of more than 1mm, suggesting that neurons with related selectivity cluster across cortical layers.  (4) When electrodes were advanced parallel to the AIT surface in a few preliminary experiments, neurons with related selectivity were found to be distributed in 2 or 3 clusters.  A single cluster measured 0.2-0.6mm and was separated from its neighbor by 0.4 to 1mm.
  The clustering of neurons with preference to related stimulus features suggests that interaction among adjacent cells contributes to the generation of stimulus selectivity. The slight difference in stimulus selectivity among adjacent neurons suggests another possible function of the modular organization.  All neurons within a module may respond to shapes and patterns which belong to the same category, but they are differentially activated by a particular stimulus because of the difference in tuning to fine parameters.  Activity patterns across a neuronal population within a module may thus be capable of representing complex stimulus features which single neurons cannot specify.


Back