Yoshiro Suzuki

Optical coherence tomography (OCT) scanning is useful in detecting various retinal diseases. However, there are not enough ophthalmologists who can diagnose retinal OCT images in much of the world. To provide OCT screening inexpensively and extensively, an automated diagnosis system is indispensable. Although many machine learning techniques have been presented for assisting ophthalmologists in diagnosing retinal OCT images, there is no technique that can diagnose independently without relying on an ophthalmologist, i.e., there is no technique that does not overlook any anomaly, including unlearned diseases. As long as there is a risk of overlooking a disease with a technique, ophthalmologists must double-check even those images that the technique classifies as normal. Here, we show that our deep-learning-based binary classifier (normal or abnormal) achieved a perfect classification on 108,308 two-dimensional retinal OCT images, i.e., true positive rate = 1.000000 and true negative rate = 1.000000; hence, the area under the ROC curve = 1.0000000. Although the test set included three types of diseases, two of these were not used for training. However, all test images were correctly classified. Furthermore, we demonstrated that our scheme was able to cope with differences in patient race. No conventional approach has achieved the above performances. Our work has a sufficient possibility of raising automated diagnosis techniques for retinal OCT images from "assistant for ophthalmologists" to "independent diagnosis system without ophthalmologists".

We built a machine learning model (ML model) which input the number of daily infection cases and the other information related to COVID-19 over the past 24 days in each of 17 provinces in South Korea, and output the total increase in the number of infection cases in each of 17 provinces over the coming 24 days. We employ a combination of XGBoost and MultiOutputRegressor as machine learning model (ML model). For each province, we conduct a binary classification whether our ML model can classify provinces where total infection cases over the coming 24 days is more than 100. The result is Sensitivity = 3/3 = 100%, Specificity = 11/14 = 78.6%, False Positive Rate = 3/11 = 21.4%, Accuracy = 14/17 = 82.4%. Sensitivity = 100% means that we did not overlook the three provinces where the number of COVID-19 infection cases increased by more than100. In addition, as for the provinces where the actual number of new COVID-19 infection cases is less than 100, the ratio (Specificity) that our ML model can correctly estimate was 78.6%, which is relatively high. From the above all, it is demonstrated that there is a sufficient possibility that our ML model can support the following four points. (1) Promotion of behavior modification of residents in dangerous areas, (2) Assistance for decision to resume economic activities in each province, (3) Assistance in determining infectious disease control measures in each province, (4) Search for factors that are highly correlated with the future increase in the number of COVID-19 infection cases.

Presently, topology optimization requires multiple iterations to create an optimized structure for given conditions. Among the conditions for topology optimization,the design area is one of the most important for structural design. In this study, we propose a new deep learning model to generate an optimized structure for a given design domain and other boundary conditions without iteration. For this purpose, we used open-source topology optimization MATLAB code to generate a pair of optimized structures under various design conditions. The resolution of the optimized structure is 32 * 32 pixels, and the design conditions are design area, volume fraction, distribution of external forces, and load value. Our deep learning model is primarily composed of a convolutional neural network (CNN)-based encoder and decoder, trained with datasets generated with MATLAB code. In the encoder, we use batch normalization (BN) to increase the stability of the CNN model. In the decoder, we use SPADE (spatially adaptive denormalization) to reinforce the design area information. Comparing the performance of our proposed model with a CNN model that does not use BN and SPADE, values for mean absolute error (MAE), mean compliance error, and volume error with the optimized topology structure generated in MAT-LAB code were smaller, and the proposed model was able to represent the design area more precisely. The proposed method generates near-optimal structures reflecting the design area in less computational time, compared with the open-source topology optimization MATLAB code.

A multiscale numerical solver called the seamless-domain method (SDM) is used in linear heat conduction analysis of nonperiodic simulated fields. The practical feasibility of the SDM has been verified for use with periodic fields but has not previously been verified for use with nonperiodic fields. In this paper, we illustrate the mathematical framework of the SDM and the associated error factors in detail. We then analyze a homogeneous temperature field using the SDM, the standard finite difference method, and the conventional domain decomposition method (DDM) to compare the convergence properties of these methods. In addition, to compare their computational accuracies and time requirements, we also simulated a nonperiodic temperature field with a nonuniform thermal conductivity distribution using the three methods. The accuracy of the SDM is very high and is approximately equivalent to that of the DDM. The mean temperature error is less than 0.02% of the maximum temperature in the simulated field. The total central processing unit (CPU) times required for the analyses using the most efficient SDM model and the DDM model represent 13% and 17% of that of the finite difference model, respectively.

A multiscale numerical solver called the seamless-domain method (SDM) is composed of macroscopic global analysis and microscopic local analysis. Previous work presented a nonlinear solver using the SDM technique that does not couple these two analyses interactively. In addition, the practicality of this solver was verified only for use with periodic fields. In this work, we present another nonlinear SDM solver that couples the multiple scales completely interactively. We solve an example problem of a nonlinear heat conduction analysis of a nonperiodic field using the presented SDM, the standard finite difference method, and the conventional domain decomposition method (DDM). The target temperature field has thermal conductivity distribution that is nonuniform, nonperiodic, and has temperature dependency. This problem thus has material nonlinearity. The accuracy of the SDM solution is very high, and the root mean squared error in temperature is less than 0.044% of the maximum temperature in the field. In contrast, the error of the DDM is 0.10–0.18%, which is larger than twice the error of the SDM. The finite difference method requires 7–36 times the computation time of the SDM to generate a solution as accurate as that of the SDM.

This paper uses classical laminate theory (CLT) and experimental methods to predict the longitudinal specific modulus of braided high modulus polyethylene (HMPE) rope without a matrix. When applying conventional CLT, the modulus, braided angle of strand, and packing factor (PF), i.e., the cross-sectional area ratio of the strand to the rope, are required. Because the void (space between strands) and PF of braided rope without a matrix readily change during the application of load, and given the difficulty measuring PF experimentally, it is difficult to predict the modulus by conventional CLT. This paper proposes the use of the unit of N/tex in place of conventional MPa for CLT. This study demonstrates that changes in PF due to void changes can be neglected when using the N/tex unit. The predicted longitudinal specific modulus of the rope using N/tex unit was found to be in qualitatively agreement with the longitudinal modulus measured experimentally.

In recent years, synthetic fiber ropes have attracted much attention because of their potential to increase the load capacity, reduce the size, and lighten the weight of tendon-driven mechanisms. However, the mechanical characteristics of synthetic fiber ropes in a dynamic loading situation remain an open problem because of their visco-elasto-plasticity. This letter focuses on modeling synthetic fiber ropes, and the frequency response of a long-distance cable-pulley system for a tendon-driven robot. We show that a synthetic fiber rope can be modeled by Flory's model, and that it can be reduced to a conventional four-element model with sufficient preloading. After empirically acquiring the parameters of the four-element model for four different synthetic fiber ropes, each frequency response of a long-distance cable-pulley servo system was measured and compared to the analytical results based on the model, as well as the results of a stainless wire rope. Consequently, we show that synthetic fiber ropes achieved comparable bandwidth to that of the stainless wire rope. The damping of synthetic fiber ropes was found to suppress the servo system gain. This is useful for joint control of a tendon-driven robot with a relatively large amount of inertia, such as a long-reach robot arm.

We present an unsteady analysis using the Seamless-Domain Method (SDM), which is a multiscale modeling technique. The SDM has previously been applied to steady-state problems to demonstrate that complicated behavior in a heterogeneous structure can be represented with relatively few points. In this paper, an unsteady analysis is carried out using SDM in the space-time domain. This domain is assumed to be composed of repeating units called space-time unit cells," which are discretized by coarse-grained points (CPs). The first step is a local analysis of the space{time domain consisting of multiple unit cells, which derives the space{time interpolation functions. The next step is a global analysis to obtain the variable distribution in the entire global domain using the interpolation functions. This two-scale analysis with respect to both space and time is computationally efficient, resulting in highly accurate solutions at low computational cost. A method that improves the computational accuracy by search in the optimum set of "reference CPs" given in the interpolation is also presented. We consider an example problem of two-dimensional thermal diffusion in a heterogeneous structure, and compute the solution using unsteady SDM and a conventional finite-difference method. The solutions are compared in terms of computational accuracy and time.

In this paper, a collision monitoring method utilizing electrostatic induction is proposed, designed, tested, and validated. When a foreign object, which is regarded as a charged body because any object is charged to a greater or lesser extent, approaches and collides with a grounded electric conductive material, an electric current caused by electrostatic induction flows to ground. The electric current is used for collision monitoring in the present study. The principle of the proposed collision monitoring method can be used for any conductive plate, which is in the air or a vacuum. However, the method is particularly aimed at collision monitoring on CFRP (carbon fiber reinforced polymer) structures for reducing the inspection cost by identifying the area where detailed inspection, such as ultrasonic inspection, needs to be conducted. The sensors are composed of the monitored object of a large conductive plate, such as a CFRP laminate, and strip conductors, such as copper tape arrayed in grid. When a foreign object collides with the monitored object, electric currents flowing through the strip conductors are measured and analyzed, and the collision point, angle, and speed of the foreign object are identified using an optimization method. As a result of several experiments, the collision point, angle, and speed are identified with high accuracy (within 5% error except for a specific case).