近年來,人類誘導性多能幹細胞培養在再生醫學領域的重要性日益增加。微模式方法常用於培養多能幹細胞,測試其分化能力。本研究基於深度學習技術提出了一種新的MP-UNet架構,用於分析在微模式螢光顯微影像,對人類誘導性多能幹細胞的早期分化細胞,進行細胞分割與分析。本研究所提出的MP-UNet架構,可適應不同影像大小,提取足夠的影像特徵,以辨識緻密細胞影像並進行螢光顯微影像之細胞分割。結合CNN影像辨識機制,判斷特定細胞分布範圍,以進行更詳細的細胞分類與統計。本研究提出的方法適用於不同大小的微模式人類誘導性多能幹細胞的顯微影像,並且提供細胞數量密度等重要分析數據。
In recent years, human induced pluripotent stem cells (hiPSC) is becoming important in the field of regenerative medicine. Micro-pattern chips are used to culture pluripotent stem cells and evaluate the differentiation of cells. This study proposes a new MP-UNet architecture based on deep learning technology, which is used to analyze the micro-pattern fluorescence microscopy images, and performs cell segmentation and analysis on the early differentiated cells of hiPSC. The MP-UNet architecture proposed in this study can analyze different image sizes and extract sufficient image features to identify compact cells and perform cell segmentation for fluorescence microscopy images. Integrated with the proposed CNN model, the mechanism can determine the range of specific cells for cell classification and statistics. The methods proposed in this study are feasible for various dimensions of the micro-pattern hiPSC microscopy images, and provide important analysis results such as cell number and density.