| 河村 奈々 | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, | 松本 健一, | MD.Altaf-Ul-Amin, | 小野 直亮, | 黄 銘 |
|
title: Data augmentation of ECG waveforms for arrhythmia detection abstract: The advent of the Holter ECG has made it possible to measure ECGs for longer than 24 hours. Because Holter ECGs allow the recording of ECGs to be measured for a longer period of time while going about one's normal life, they can detect arrhythmias that are difficult to detect in a short period of time. As a result, the diagnostic burden on physicians is increasing. Therefore, the detection of arrhythmias using machine learning has been expected, and previous studies using supervised learning methods such as CNN and RNN have reported that these models are able to detect arrhythmias with high accuracy. The success of the above studies relies on an annotated dataset. However, obtaining more annotated data sets is very costly and difficult because it requires an experienced cardiologist to annotate the data. Also, many annotated datasets have few abnormal labels for data with normal labels, creating a data class imbalance problem. As a result, the amount and diversity of annotated data is still insufficient. In this study, we use various data augmentation methods to generate data and consider it. language of the presentation: Japanese | |||||||
| 髙下 大貴 | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, | 松本 健一, | MD.Altaf-Ul-Amin, | 小野 直亮, | 黄 銘 |
|
title: Smooth Embedding of Molecular Structure Using Molecular Hypergraph Variational Autoencoder with Metric Learning
abstract: Deep learning approaches are widely used to search molecular structures for a candidate drug/material. The basic approach in drug/material candidate structure discovery is to embed a relationship that holds between a molecular structure and the physical property into a low-dimensional vector space (chemical space) and search for a candidate molecular structure in that space based on a desired physical property value. Deep learning simplifies the structure search by efficiently modeling the structure of the chemical space with greater detail and lower dimensions than the original input space. In our research, we propose an effective method for molecular embedding learning that combines variational autoencoders (VAEs) and metric learning using any physical property. Our method enables molecular structures and physical properties to be embedded smoothly into VAEs’ latent space while maintaining the consistency of the relationship between the structural features and the physical properties of molecules. And we propose a new quantitative evaluation method to evaluate how well the modelled chemical space is an ideal expression for molecular structure search. We demonstrate that the embedding representation extracted by our method is effective in molecular structure design from analysis results of physical property dataset. language of the presentation: Japanese | |||||||
| 佐藤 拓馬 | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, | 松本 健一, | MD.Altaf-Ul-Amin, | 小野 直亮, | 黄 銘 |
|
title: Analysis of cell morphology by clustering of multiple stained histopathological images using deep learning
abstract: The computer aided diagnosis based on deep learning are not only helpful for classification, but also useful for feature extraction from given images, especially encoding image data into discrete representation helps us obtain new knowledge. Although most of the related studies are based on supervised learning that needs curated pathological knowledge, it is useful to extract characteristic features in the given images, using unsupervised machine learning in order to obtain new pathological findings. We applied cluster analysis using deep learning which is trained based on self-augmented training (SAT) and maximization of mutual information. It showed that it can classify those pathological images into discrete categories. Our findings can be transalated into practical use in clinical environments. language of the presentation:Japanese | |||||||