本論文提出一個基於卷積神經網路架構(CNNs)的系統,並用於自動分類並辨識顯微鏡觀測人類幹細胞區域影像。我們的方法能偵測出reprogramming和reprogrammed的人類誘導性多功能幹細胞,使生物學家能進一步以試劑篩查或條件培養達成幹細胞培養。學習結果展示出我們的CNNs可以已top-1 error rates 為5.9%和top-2 error rates 為0.9%的準確度繪製機率分布圖並用以分析。實作結果顯示出這個自動化方法可以成功偵測出並定位出人類幹細胞組成,從而得到一個對幹細胞研究有幫助的潛力工具。
We present a deep learning architecture Convolutional Neural Networks (CNNs) for automatic classification and recognition of reprogramming and reprogrammed human Induced Pluripotent Stem (iPS) cell regions in microscopy images. The differentiated cells that possibly undergo reprogramming to iPS cells can be detected by this method for screening reagents or culture conditions in iPS induction. The learning results demonstrate that our CNNs can achieve the Top-1 and Top-2 error rates of 5.9% and 0.9%, respectively, to produce probability maps for the automatic analysis. The implementation results show that this automatic method can successfully detect and localize the human iPS cell formation, thereby yield a potential tool for helping iPS cell culture.