Background and Aim It is necessary to establish universal methods for endoscopic diagnosis of Helicobacter pylori (HP) infection, such as computer‐aided diagnosis. In the present study, we propose a multistage diagnosis algorithm for HP infection. Methods The aims of this study are to: (i) to construct an interpretable automatic diagnostic system using a support vector machine for HP infection; and (ii) to compare the diagnosis capability of our artificial intelligence (AI) system with that of endoscopists. Presence of an HP infection determined through linked color imaging (LCI) was learned through machine learning. Trained classifiers automatically diagnosed HP‐positive and ‐negative patients examined using LCI. We retrospectively analyzed the new images from 105 consecutive patients; 42 were HP positive, 46 were post‐eradication, and 17 were uninfected. Five endoscopic images per case taken from different areas were read into the AI system, and used in the HP diagnosis. Results Accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of the diagnosis of HP infection using the AI system were 87.6%, 90.4%, 85.7%, 80.9%, and 93.1%, respectively. Accuracy of the AI system was higher than that of an inexperienced doctor, but there was no significant difference between the diagnosis of experienced physicians and the AI system. Conclusions The AI system can diagnose an HP infection with significant accuracy. There remains room for improvement, particularly for the diagnosis of post‐eradication patients. By learning more images and considering a diagnosis algorithm for post‐eradication patients, our new AI system will provide diagnostic support, particularly to inexperienced physicians.

Functional near-infrared spectroscopy (fNIRS) is suitable for noninvasive mapping of relative changes in regional cortical activity but is limited for quantitative comparisons among cortical sites, subjects, and populations. We have developed a convolutional neural network (CNN) analysis method that learns feature vectors for accurate identification of group differences in fNIRS responses. In this study, subject gender was classified using CNN analysis of fNIRS data. fNIRS data were acquired from male and female subjects during a visual number memory task performed in a white noise environment because previous studies had revealed that the pattern of cortical blood flow during the task differed between males and females. A learned classifier accurately distinguished males from females based on distinct fNIRS signals from regions of interest (ROI) including the inferior frontal gyrus and premotor areas that were identified by the learning algorithm. These cortical regions are associated with memory storage, attention, and task motor response. The accuracy of the classifier suggests stable gender-based differences in cerebral blood flow during this task. The proposed CNN analysis method can objectively identify ROIs using fNIRS time series data for machine learning to distinguish features between groups.

INF-β has been widely used to treat patients with multiple sclerosis (MS) in relapse. Accurate prediction of treatment response is important for effective personalization of treatment. Microarray data have been frequently used to discover new genes and to predict treatment responses. However, conventional analytical methods suffer from three difficulties: high-dimensionality of datasets; high degree of multi-collinearity; and achieving gene identification in time-course data. The use of Elastic net, a sparse modelling method, would decrease the first two issues; however, Elastic net is currently unable to solve these three issues simultaneously. Here, we improved Elastic net to accommodate time-course data analyses. Numerical experiments were conducted using two time-course microarray datasets derived from peripheral blood mononuclear cells collected from patients with MS. The proposed methods successfully identified genes showing a high predictive ability for INF-β treatment response. Bootstrap sampling resulted in an 81% and 78% accuracy for each dataset, which was significantly higher than the 71% and 73% accuracy obtained using conventional methods. Our methods selected genes showing consistent differentiation throughout all time-courses. These genes are expected to provide new predictive biomarkers that can influence INF-β treatment for MS patients.

This study examines the effects of focused-attention meditation on functional brain states in novice meditators. There are a number of feature metrics for functional brain states, such as functional connectivity, graph theoretical metrics, and amplitude of low frequency fluctuation (ALFF). It is necessary to choose appropriate metrics and also to specify the region of interests (ROIs) from a number of brain regions. Here, we use a Tucker3 clustering method, which simultaneously selects the feature vectors (graph theoretical metrics and fractional ALFF) and the ROIs that can discriminate between resting and meditative states based on the characteristics of the given data. In this study, breath-counting meditation, one of the most popular forms of focused-attention meditation, was used and brain activities during resting and meditation states were measured by functional magnetic resonance imaging. The results indicated that the clustering coefficients of the eight brain regions,

No abstract

<p>Deep generative models are used to generate arbitrary molecular structures with the desired chemical properties. MolGAN is a renowned molecular generation models that uses generative adversarial networks (GANs) and reinforcement learning to generate molecular graphs in one shot. MolGAN can effectively generate a small molecular graph with nine or fewer heavy atoms. However, the graphs tend to become disconnected as the molecular size increase. This poses a challenge to drug discovery and material design, where large molecules are potentially inclusive. This study develops an improved MolGAN for large molecule generation (L-MolGAN). In this model, the connectivity of molecular graphs is evaluated by a depth-first search during the model training process. When a disconnected molecular graph is generated, L-MolGAN rewards the graph a zero score. This procedure decreases the number of disconnected graphs, and consequently increases the number of connected molecular graphs. The effectiveness of L-MolGAN is experimentally evaluated. The size and connectivity of the molecular graphs generated with data from the ZINC-250k molecular dataset are confirmed using MolGAN as the baseline model. The model is then optimized for a quantitative estimate of drug-likeness (QED) to generate drug-like molecules. The experimental results indicate that the connectivity measure of generated molecular graphs improved by 1.96 compared with the baseline model at a larger maximum molecular size of 20 atoms. The molecules generated by L-MolGAN are evaluated in terms of multiple chemical properties, QED, synthetic accessibility, and log octanol–water partition coefficient, which are important in drug design. This result confirms that L-MolGAN can generate various drug-like molecules despite being optimized for a single property, i.e., QED. This method will contribute to the efficient discovery of new molecules of larger sizes than those being generated with the existing method.<br></p>

Abstract-It is reported that simulated annealing (SA), which changes one design variable at a time, is effective when applied to high-dimensional continuous optimization problems. However, if a correlation exists among the design variables, it is not efficient to search each dimension. In this paper, we propose SA with a mechanism to determine an appropriate search direction according to the landscape of the given problems. Its effectiveness is verified for high-dimensional problems with correlation among design variables.

Supplemental Digital Content is Available in the Text.