7月26日（金）8:30～8:50 第10会場（万代島ビル 6F 会議室）
2WD10cm1-1
認知的負荷の適応的回避行動における二次学習の負荷予測と負荷予測誤差の神経基盤
Asako Mitsuto Nagase（永瀬 光戸 麻子）1,2，Keiichi Onoda（小野田 慶一）2，Toshikazu Kawagoe（川越 敏和）2,3，Rei Akaishi（赤石 れい）4，Shuhei Yamaguchi（山口 修平）2，Kenji Morita（森田 賢治）1
1東京大院教育身体教育
2島根大医三内、出雲、日本
3立教大現代心理学部、埼玉、日本
4脳情報研究通信融合センター、大阪、日本
Earlier work demonstrated the tendency of humans to avoid mental effort. Our previous work suggested that humans would avoid mental effort adaptively with neural mechanisms to represent expected cost positively in the dorsomedial frontal cortex (dmFC)/dorsal anterior cingulate cortex (dACC) and the right anterior middle frontal gyrus (aMFG) and prediction error (PE) of mental effort positively in dmFC/dACC and other regions, irrespective of cognitive demand types. In that study, we conducted functional magnetic resonance imaging (fMRI) experiments with two variations of demand selection tasks. However, in that experimental design, the neural signals of expected cost could be confounded with the control signal. Also, the timings of the PE generation of mental effort cost were not properly identified because the predication error signal could be confounded with control signal and various signals relating task execution. Therefore, in order to reduce the effect of task preparation at the time of choice-cue and lock the timing of generating the cost PE signal of mental effort in second order learning, we added task-type cues that indicated upcoming mental demand level, either high or low, before the actual presentation of the mental effort tasks. We conducted two fMRI experiments (Exp. 1 (mental arithmetic): n = 30, Exp. 2 (spatial reasoning): n = 28), and fitted mathematical models to the choice data. The analysis of behavioral results indicated that most participants showed avoidance of high mental effort levels in both experiments (Exp. 1: n = 27, Exp. 2: n = 24). The best-fit model of about half of them (Exp. 1: 44%, Exp. 2: 54%) was a probabilistic model, in which decisions are made based solely on the experiences of immediately preceding trials. The best-fit model of most of the rest (Exp. 1: 37%, Exp. 2: 42%) was a reinforcement learning (RL) model based on mental-effort levels on multiple previous trials. Model-based fMRI analyses with the RL model in Exp. 1 revealed that activity of the dmFC, the right superior/inferior parietal lobule (S/IPL) and the left IPL was positively correlated with the expected cost for the chosen option, and activity of the right IPL was negatively correlated with PE of mental effort cost in second order learning at the time of task-type cues. These results replicated our previous findings, and further implied the possibility that the right IPL would represent cost PE of mental effort in second order learning negatively. | | 7月26日（金）8:50～9:10 第10会場（万代島ビル 6F 会議室）
2WD10cm1-2
海馬透過光画像からのCA2領域の自動検出
Shohei Morinaga（森永 祥平）1，Hiroki Ojima（尾嶋 弘貴）1，Tomoe Ishikawa（石川 智愛）2，Masato Yasui（安井 正人）2，Shuichi Akizuki（秋月 秀一）1，Mototsugu Hamada（濱田 基嗣）1，Tadahiro Kuroda（黒田 忠広）1
1慶應大理工電子
2慶應大医薬理
The Cornu Ammonis area in the hippocampus are filled with pyramidal cells, which divides into three subregions, including the CA1, CA2, and CA3. Anatomical and functional properties of each subregion have been extensively studied, however, it has been difficult to identify the anatomical area of each subregion. Molecular markers provide a powerful tool to assess anatomical area of the subregions, on the other hand, additional time-consuming techniques, such as preparation of transgenic animals or post-hoc immunostaining, are necessary. The aim of our research is the creation of an alternative method to assess anatomical area of the CA2 subregion from hippocampal transmitted-light images, by a time efficient computational machine learning approach. As an initial step towards creating such system, we utilized a modified version of a convolutional neural network architecture for medical image segmentation called U-Net, to assess the CA2 subregion from a microscopic image of the mice hippocampus. The output of our system is a confidence heatmap and a maximum confidence coordinate . This proposed system was trained by a dataset which consists of 220 transmitted-light images and manually labeled CA2 region masks created from fluorescent images. We tested the system on another dataset which consist of 12 images. All images were obtained from PFA-fixed horizontal hippocampal slices from adult (9-11 weeks) C57/BL6J mice. The proposed system achieved a detection accuracy of 100% on the test dataset. This work should provide biologists with efficient and accurate tools to assess the CA2 subregion, improving the efficiency of studies in the area. Additionally, based on the results of this work, we will extend our system to detect other subregions and anatomical areas. |
