[15a-B414-1~9] FS.1 Focused Session "AI Electronics"

△ [15a-B414-5] Synchrotron-Radiation Nanotomography of Neuronal Networks of Schizophrenia Cases and its Application to the Neural Network Design

〇(BC)Yo Maruta¹, Rino Saiga¹, Ryuta Mizutani¹, Yoshiro Yamamoto², Susumu Takekoshi³, Chie Inomoto³, Naoya Nakamura³, Makoto Arai⁴, Mitsuhiro Miyashita⁴, Kenichi Oshima⁴, Masanari Itokawa⁴, Youta Torii⁵, Itaru Kushima⁵, Norio Ozaki⁵, Shuji Iritani⁶, Yuichi Yamashita⁷, Masayuki Uesugi⁸, Akihisa Takeuchi⁸, Kentaro Uesugi⁸, Yasuko Terada⁸, Yoshio Suzuki⁹, Viktor Nikitin¹⁰, Francesco De Carlo¹⁰ (1.Sch of Engineering, Tokai Univ, 2.Sch of Science, Tokai Univ, 3.Sch of Medicine, Tokai Univ, 4.Tokyo Metropolitan Institute of Medical Sci, 5.Grad Sch of Medicine, Nagoya Univ, 6.Okehazama Hospital Fujita Kokoro Care Center, 7.National Center of Neurology and Psychiatry, 8.JASRI/SPring-8, 9.KEK, 10.Argonne National Lab)

Keywords:synchrotron radiation, neural network, brain

Synchrotron radiation nano-CT analysis of brain tissues of schizophrenia cases revealed that (1) neurites are thin and tortuous in schizophrenia, and that (2) the auditory hallucination score correlates with the neurite curvature (r = -0.80, p = 0.00018). We integrated the structural findings into fully-connected layers and found that the schizophrenia-mimicking network outperforms conventional networks in MNIST and CIFAR-10 classification tasks. Number of parameters can be reduced a priori by 50% in a schizophrenia-mimicking VGG network. These results indicate that the neuronal alterations in schizophrenia can enhance certain brain abilities. It has been reported that creativity and schizophrenia share the same genetic roots. We suggest that our results should represent the creativity advantage associated with schizophrenia.