Faster R-CNN for IPSC-Derived Mesenchymal Stromal Cells Senescent Detection from Bright-Field Microscopy

Abstract

iPSC-derived mesenchymal stromal cells (iMSCs) play an important role in cell therapy and regenerative medicine, but the differentiation and proliferation ability of senescent iMSCs decline greatly, which will also bring heterogeneity and potential side effects. The whole senescent degree of iMSCs can only be obtained by vital stain. However, this process will cost a lot of manpower, money and time. To solve this problem, we apply deep learning for automated iMSCs senescent recognition, which can quickly and accurately get the senescent situation of single-cell without staining. The adopted Faster R-CNN uses ResNet as the backbone network with an FPN module. Experiments on the collected dataset show that our method has achieved a detection accuracy of 0.768 in the mixed test set of each generation of cells and the independent test set of each generation of cells.