National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering, Health Science Center, Shenzhen University, Shenzhen, China.
Department of Medical Cell Biology and Genetics, Guangdong Key Laboratory of Genomic Stability and Disease Prevention, Shenzhen Key Laboratory of Anti-Aging and Regenerative Medicine, and Shenzhen Engineering Laboratory of Regenerative Technologies for Orthopaedic Diseases, Health Science Center, Shenzhen University, Shenzhen, 518060, China.
Comput Methods Programs Biomed. 2021 Sep;208:106235. doi: 10.1016/j.cmpb.2021.106235. Epub 2021 Jun 22.
Induced pluripotent stem cells (iPSCs) have great potential as the basis of regenerative medicine. In this paper, we propose an automatic quality evaluation model based on multi-source feature ensemble learning to divide the iPSC colonies into three categories: good, medium and bad.
First, we obtained iPSCs samples using a Sendai virus reprogramming method. Second, we collected the bright field-images of iPSC colonies and processed them with adaptive gamma transform and data enhancement. The evaluation for the iPSC colony quality was further verified with living cell fluorescent staining, currently accepted as the optimal biological method. Third, multi-source features were extracted using three deep convolutional neural networks (DCNNs) and four traditional feature descriptors. Finally, we utilized a support vector machine (SVM) to perform classification. Before feeding into the SVM, the features were processed by principal component analysis algorithm to save computational cost and training time.
Experimental results on the collected iPSC dataset (46,500 images) show that the proposed method could obtain 95.55% classification accuracy.
Our study could provide a method to efficiently and quickly judge the biological quality of a single iPSC colony or populations and facilitate the large-scale iPSC manufacturing.
诱导多能干细胞（iPSCs）作为再生医学的基础具有巨大的潜力。本文提出了一种基于多源特征集成学习的自动质量评估模型，将 iPSC 集落分为三类：良好、中等和差。
首先，我们使用仙台病毒重编程方法获得 iPSCs 样本。其次，我们采集 iPSC 集落的明场图像，并对其进行自适应伽马变换和数据增强处理。最后，使用支持向量机（SVM）进行分类。在将特征输入 SVM 之前，我们使用主成分分析算法对特征进行处理，以节省计算成本和训练时间。
在收集的 iPSC 数据集（46500 张图像）上的实验结果表明，该方法可以获得 95.55%的分类准确率。
我们的研究可以提供一种高效、快速判断单个 iPSC 集落或群体的生物学质量的方法，为大规模 iPSC 制造提供便利。
