拍動心筋細胞の動画を全自動撮影し、ベクトル解析によって心筋細胞を同定することで非心筋細胞との領域を判別する拍動心筋細胞動画撮影システムを開発した。また、セグメンテーションアルゴリズムにより拍動する心筋細胞を効率的に評価する画期的な深層学習法の開発を行い、改良することでより正確な領域判定を可能にした。心筋と非心筋細胞を見分けるのに有効な分類器を機械学習により構築し、加えて学習器を可視化することで分類根拠を視覚的に示すことに成功した。拍動心筋細胞動画撮影システムにより効率的に教師データを取得し、学習機によるデータの解析を行った。さらに心筋細胞への分化、成熟過程における重要な遺伝子を確認した。
We have developed a video imaging system for beating cardiomyocytes (CMs). The system was able to recognize beating CMs by vector analysis. It automatically captured their images and accurately identified regions of both CMs and non-CMs. In addition, we developed an epoch-making deep learning method that efficiently evaluated beating CMs by a segmentation algorithm, which enabled the more precise judgement of cell region. A classifier that was effective in distinguishing between CMs and non-CMs was also constructed by machine learning. Moreover, we succeeded in presenting the basis of classification by visualizing the learning device. The imaging system made it possible to efficiently acquire a large number of teaching data, which was a bottle neck in the existing analysis with artificial intelligence, and a large quantity of data was analyzed by the developed learning machine. Furthermore, we confirmed important genes in the process of differentiation into CMs and maturation of CMs.

研究種目 : 挑戦的研究 (萌芽)
研究期間 : 2019～2020
課題番号 : 19K22625
研究分野 : 循環器内科