SearcharxivSearch

arXiv subjects

Toshiaki Shimizu

Publications and source records attributed to Toshiaki Shimizu.

2 recordsLinked to original sources

Emergent Complexity in a Light-Driven Self-Oscillatory Crystal: A Molecular Perspective on Autonomous Behavior and Stimulus-Modulated Motion

Living organisms are molecular systems with self-sustained dynamics via energy conversion through molecular cooperation, resulting in highly complex macroscopic behaviors. Construction of such autonomous macroscopic dynamics at a molecular system level remains one of the central challenges in the field of chemistry. Looking further ahead, constructing motile systems that can receive external information and adapt their autonomous behavior represents the next frontier towards newly functional molecular devices such as microrobots. In this study, we focused on a light-driven self-oscillatory crystal that exhibits continuous flipping motion under constant light irradiation. We experimentally evaluated the oscillation frequency of the crystal under polarized light and confirmed the validity of our previously proposed mechanism, and clarified the requirements for self-oscillation. Based on this mechanism, we constructed a mathematical model that demonstrates the motion of the crystal itself. The model revealed that diverse oscillatory behaviors at the macroscopic level can emerge due to small differences in conditions even when the underlying molecular-level processes are identical. Furthermore, we found that the oscillatory behavior depended on the state generated by the previously applied light. This suggests that a memory effect contributes to the complex motion of the crystal.

physics.bio-ph

Convolutional Neural Networks for Estimation of Myelin Maturation in Infant Brain

Myelination plays an important role in the neurological development of infant brain and MRI can visualize the myelination extension as T1 high and T2 low signal intensity at white matter. We tried to construct a convolutional neural network machine learning model to estimate the myelination. Eight layers CNN architecture was constructed to estimate the subjects age with T1 and T2 weighted image at 5 levels associated with myelin maturation in 119 subjects up to 24 months. CNN model learned with all age dataset revealed a strong correlation between the estimated age and the corrected age and the coefficient of correlation, root mean square error and mean absolute error was 0. 81, 3. 40 and 2. 28. Moreover, the adaptation of ensemble learning models with two datasets 0 to 16 months and 8 to 24 months improved that to 0. 93, 2. 12 and 1. 34. Deep learning can be adaptable to myelination estimation in infant brain.

q-bio.QM