一般口演
視覚 1
Vision 1
座長：石川 理子（慶應義塾大学） |
|---|
| 2022年7月2日 10:00～10:15 沖縄コンベンションセンター 会議場A2 第7会場
3O07m2-01
超高密度電極を用いたマカクV4野における機能構築の解明
High-density recording reveals functional domains of shapes and textures across laminae in macaque V4
*波間 智行(1)、Erin Kempkes(1)、Polina Zamarashkina(1)、Natalia Owen(1)、Anitha Pasupathy(1)
1. ワシントン大学
*Tomoyuki Namima(1), Erin Kempkes(1), Polina Zamarashkina(1), Natalia Owen(1), Anitha Pasupathy(1)
1. University of Washington
Keyword: Neuropixels, Object recognition, Visual area, Non human primate
Macaque monkey area V4, an intermediate stage of visual processing, includes neurons that exhibit exquisite selectivity for stimulus form and surface texture. Recent studies using 2 photon calcium imaging and intrinsic optical imaging have reported functional domains for object form in the superficial layers of macaque V4 (Jiang et al., 2021; Hu et al., 2021). However, the functional organization across laminae is not known. In present study, using high-density probes in awake monkeys, we identified clusters of neurons that exhibit similar tuning for shape, texture or both, and then asked where those neuronal cluster located along V4 laminae. We also studied how the tuning consistency of responses for shape and texture varies along the laminae. We performed neuronal recordings in two fixating macaque monkeys as visual stimuli were flashed (200 ms) separated by blank periods (200 ms) in random order, in the near-periphery. We used a set of 2D shapes at two luminance contrasts (dark/bright) or a set of textures with two luminance gradients (original/luminance-reversed). Each day we studied activities from tens of well-isolated, nearby neurons using Neuropixels probe. Across penetrations in both animals, we saw clear clusters of 5 or more neurons with high similarity in shape/texture preference. Neuronal clusters preferentially occupy the superficial layers for shape stimuli but less so for texture. Within clusters of similar tuning, we found that neurons exhibited greater consistency in tuning across contrast reversals and this trend was also more evident for shape as compared to texture. Overall, our results provide the first documentation of specialized functional domains for contrast invariant encoding of shape and texture that could play a critical role in the processing of form and surface characteristics. | | 2022年7月2日 10:15～10:30 沖縄コンベンションセンター 会議場A2 第7会場
3O07m2-02
サルV4神経集団の応答とヒト図地知覚との相関
Population responses of monkey V4 neurons correlate with human figure-ground perception
*宍倉 基文(1)、田村 弘(2)、酒井 宏(1)
1. 筑波大学、2. 大阪大学大学院
*Motofumi Shishikura(1), Hiroshi Tamura(2), Ko Sakai(1)
1. Dept Computer Science, University of Tsukuba, Ibaraki, Japan, 2. Graduate School of Frontier Biosciences, Osaka University, Osaka, Japan
Keyword: VISION, V4, PSYCHOPHYSICS, ELECTROPHYSIOLOGY
Segregation of natural scenes into objects (figure) and background (ground) is one of the most important steps leading to object recognition. Our recent study reported that neurons in monkey V4 exhibited preferences to figure and ground (FG neurons) in local natural images. However, the response consistencies of individual neurons across stimuli were no more than slightly above a chance rate, and thus whether these neural activities underlie the perception of FG segregation was skeptical. Here, we investigated the contribution of the population responses of FG neurons to FG perception. Specifically, we estimated perceptual consistencies (PC) in FG judgement for the natural patches across trials and participants and compared them with population neural consistencies (PNC) across trials that were estimated from the integration of the responses of a few tens of neurons. A positive correlation between the two would support the link between the population responses and perception. First, we performed psychophysical experiments to estimate the PCs for each of 210 local natural images. We classified the stimuli into two groups based on the PC: the EASY stimuli with the top 100 PC values (perceptually easy to judge FG) and the HARD stimuli with the lowest 100 PC values (difficult to judge FG). The mean PCs for EASY and HARD stimuli were 80% and 66%, respectively, and were significantly different. Next, we examined whether the PCs were correlated with the PNCs. We used a support vector machine to estimate FG from the responses of multiple FG neurons, and PNCs were computed from the machine responses. With the integration of 50 neurons, the PNCs for the EASY and HARD stimuli were 74% and 63%, respectively, and were significantly different. This result indicates that the population response of FG neurons correlates with the consistency of FG perception. In addition, we evaluated neural modulation latencies for EASY and HARD stimuli and compared them with perceptual reaction times. A positive correlation was expected since perceptually hard stimuli needed more reaction times for FG judgement than easy stimuli. The mean modulation latencies of individual FG neurons for EASY and HARD stimuli were 68 and 115 ms, respectively, indicating a positive correspondence between the modulation latencies and the reaction times. These results support that a population of FG neurons plays a crucial role in FG perception. | | 2022年7月2日 10:30～10:45 沖縄コンベンションセンター 会議場A2 第7会場
3O07m2-03
サルにおける視覚探索と触感探索を連合させた素材感識別
Material perception associating haptic and visual inspections in non-human primate subjects
*伊藤 南(1)、林田 梨央(1)、小山 佳奈子(1)、菅原 理沙(1)
1. 東京医科歯科大学
*Minami Ito(1), Rio Hayashida(1), Kanako Koyama(1), Risa Sugawara(1)
1. Tokyo Medical and Dental University (TMDU)
Keyword: MATERIAL PERCEPTION, MONKEY, CATEGORIZATION, BEHAVIORAL ANALYSIS
Materials perception provides us a powerful clue to understand object recognition. This might be given by an interaction between visual texture information and secondary haptic information. However, the haptic perception might be dependent on animal's sensation and experiences, and the property is not well understood in primates used for physiological studies. In the previous study, we have trained animals by the material categorization task due to the haptic inspections (Task1). Here, we introduced the second material categorization task (Task2), to make animal associate haptic and visual information. Three Japanese monkeys (macaca fuscata, female, 6.7, 7.0 & 6.4kg) have been trained to touch a cue lever and then one of 5 response levers (metal, wood, fabric, gel-sheet, and fur), over 60, 54 & 42 months. Material-samples (5x5cm) are attached on the cue levers, which were presented in a random order. To get water reward, animals have to choose a collect response lever of the cued material. In Task1, order of the response lever were fixed and the task was conducted under haptic alone condition. In Task 2, materials on the response lever were shifted for each session, so that animals examined material category by haptic inspections and then looked for a corresponding response lever by visual inspections. In Task2, the animals showed severe confusions even though visual information is available during the task, and took longer time to improve their performance above 90%. Then, 96 new material samples (5x5cm) were presented once or twice in a daily session and measured 10 times. Distribution of the responses represents animals’ material categorization into 5 categories, which were compared with those of human subjects as we described in the previous study. In a two-dimensional material space given by human studies, animals’ categorization was consistent. However, the border of category was rather unique for each animal (kappa coefficient, chi-square goodness-of-fit test (p<0.05 ). In conclusion, the non-human primate subjects did have a kind of material perception, but their properties were not identical to that of human subjects. Apparently, making association between the haptic information and visual information in Task2 demands more for primate subjects, resulting in more errors. Thus, it is necessary to see properties of the material perception in order to interpret physiological data in each animal. | | 2022年7月2日 10:45～11:00 沖縄コンベンションセンター 会議場A2 第7会場
3O07m2-04
サル側頭葉TE野における顔応答性ニューロンの受容野とコードする顔画像情報との相関関係
Correlations between receptive-fields of face-responsive neurons in monkey area TE and their coded facial information
*菅生-宮本 康子(1)、塩谷 佳介(2)、林 和子(1,3)、松田 圭司(1)、片上 舜(2)、松本 有央(1)、赤穗 昭太郎(1)、岡田 真人(2)、河野 憲二(1)
1. 産業技術総合研究所、2. 東京大学、3. 日本学術振興会
*Yasuko Sugase-Miyamoto(1), Keisuke Shioya(2), Kazuko Hayashi(1,3), Keiji Matsuda(1), Shun Katakami(2), Narihisa Matsumoto(1), Shotaro Akaho(1), Masato Okada(2), Kenji Kawano(1)
1. AIST, Japan, 2. The University of Tokyo, Japan, 3. JSPS, Japan
Keyword: temporal cortex, face, representation, neurophysiology
Our previous study showed that face inversion decreased the amount of information about facial identities/expressions but did not affect the information about a global category, i.e., monkey vs. human vs. shapes, suggesting that different neuronal members may contribute to the representation of the facial identities/expressions and that of the global category (Sugase-Miyamoto et al., 2014). Because receptive field (RF) size of area TE neurons is known to vary (Op de Beeck and Vogels, 2000), we investigated the correlation between RF characteristics of face-responsive neurons and the information represented in their firing rates. Neuronal activities were recorded in area TE of two rhesus monkeys (Macaca mulatta) performing two fixation tasks, one for the coded facial information and the other for their RFs. Test stimuli were 20 colored pictures consisting of nine monkey faces, nine human faces (three models with three expressions each) and two shapes (size: within 6 deg x 6 deg), and each stimulus was presented for 400 ms at the center of the CRT-screen. To identify the RF of each neuron, one of the stimuli that evoked the strongest response was presented at one of the 15 positions in a 3 x 5 grid (rows x columns; size of each cell: 6 deg x 6 deg) centered on the screen while the monkey fixated on the central fixation point. The responses to the stimulus at each position were averaged in a window 50-350 ms after the stimulus onset, the baseline activity (the average response during 200 ms before the stimulus onset) was set to zero, and the responses were normalized. Linear interpolation was used to obtain a contour map of the RF of each neuron, and the RF size was determined as the area in the contour map that was at or above 50% of the maximum response. Eighty-one face-responsive neurons were recorded, and the RF was identified in seventy-two. Principal component (PC) analysis applied on the normalized responses across the seventy-two revealed that the first PC was correlated with the RF sizes. To determine the information represented in their firing rates, we applied a mutual information analysis to their responses similar to our previous studies. The RF sizes positively correlated with the amount of information about the global category, whereas the RF sizes negatively correlated with the amount of information about the monkey expression, showing that neurons with relatively small RFs represented information about monkey expressions. | |
|
|
