In this research, we introduce a novel deep-learning-based approach for segmenting sparse-view CT images using multi-energy projections. Unlike traditional methods that depend on pre-reconstructed CT images, our technique integrates CT reconstruction and segmentation into a single network, enhancing its applicability to sparse-view CT. This approach is crucial for reducing radiation dose and acquisition time. Our method extracts pixel fragments from multi-energy projections, corresponding to specific CT image voxels. The network, composed of 'filtering,' 'back-projection,' and 'segmentation' sub-networks, directly segments CT images from projections, bypassing the need for prior reconstruction. This segmentation technique is particularly memory-efficient, as it operates on a bar-by-bar basis, avoiding memory-intensive 3D convolutions. The result is a high-quality segmentation that effectively overcomes issues of sparse-view artifacts and the memory constraints common in previous methods. Our integrated approach thus offers a significant advancement in CT imaging, combining efficiency with improved segmentation quality.

Deterministic lateral displacement (DLD) is a microfluidic technique for separating microparticles based on their sizes through periodically arrayed micropillars. The size-tunable DLD device can adjust the critical diameter (D_c) which determines the migration modes of particles, to perform separation in a wide range of predetermined particle sizes. Previous research reported a size-tunable DLD consisting of thermo-response hydrogel pillars on glass for particle separation in pure water. However, the poor thermal conductivity of glass could potentially introduce delays in heat transfer, affecting D_c control. Therefore, it is necessary to construct an efficient thermal response device and evaluate the heat transfer and response times. In this study, we used silicon with extreme thermal conductivity as the substrate and fabricated the thermo-responsive hydrogel poly(N-isopropyl acrylamide)(PNIPAM) pillars on it, and sealed with a poly(dimethylsiloxane) microchannel. The geometric changes of the pillars with temperature in pure water were measured. Further, we evaluated the ability to switch migration trajectories (bump or zigzag mode) of particles (12-μm microspheres) by varying the temperature.

Currently, underground excavation machines, such as shield machines, have been developed to excavate tunnels as small as several meters in diameter. In this study, we will investigate a mechanism to reduce the size of this excavation mechanism to a few centimeters in diameter.

We are developing a feeding mechanism for colonoscopy. In developing a device for an automated mechanism that does not depend on the skill of the physician, there is an urgent need to reduce the size of the mechanical system. This study reports on our investigation of the miniaturization of the colonoscopy delivery system.

Sketches as line drawings are useful for designing mechanical objects. If the sketches can be automatically converted into 3D models, the design process of mechanical objects will be efficient. In this study, a conversion method that can handle curved surfaces is proposed although they have not handled by previous researches. The method is developed using the characteristics of mechanical objects, and aims at practicability.

In order to further improve the expressive power of log-aesthetic curves,Tsuchie and Yoshida investigated the derivation process of log-aesthetic curves and they proposed a α-curve with variable shape parameter α. The derivation method extends the differential equation regarding curvature and curve length (arc length) that defines the log-aesthetic curve, It is a relatively complicated process deriving a new curve formula, and its intuitive meaning is not clear.In this report, we will investigate the log-aesthetic curve, its extension σ-curve and their similar curve τ-curve and propose a method to improve the ability to express curves by setting the shape parameter α to a polynomial of degree m with respect to the arc length parameter.

In this research, we divided the curve generated based on Kenjiro Miura's FTC curve formula into three segments, adjusted the parameters and created a dataset.We learned with GAN and created ChatGPT3 from a design vocabulary. We have developed an aesthetic CAD system that performs shape design using ChatGPT3.5Turbo.