Deep Mining of Subtle Differences in Cell Morphology via Deep Learning. Issue 2 (10th December 2020)

• Record Type: Journal Article
• Authors: Xue, Yunfan; Wang, Jing; Ren, Kefeng; Ji, Jian
• Is Part Of: Advanced theory and simulations. Volume 4:Issue 2(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-12-10
• Subjects: cell morphology -- convolutional neural networks -- deep learning -- substrate stiffness; Science -- Simulation methods -- Periodicals; Science -- Methodology -- Periodicals; Engineering -- Simulation methods -- Periodicals; Engineering -- Methodology -- Periodicals; 507.21
• DOI: 10.1002/adts.202000172
• Languages: English
• ISSNs: 2513-0390

Abstract

Abstract: Cell morphology analysis is crucial in life science. Accurate determination of differences in cell morphology is of great significance in understanding cell states under different conditions. However, conventional approaches for morphology analysis are constrained in efficiency or accuracy under many circumstances. Thus, an efficient and reliable method to profile morphological differences is needed. In this study, deep learning is used in cell image analysis to demonstrate its ability to find non‐apparent cell morphology differences. The convolutional neural network can accurately classify cell images from substrates of different stiffness that are undistinguishable to the human eye or conventional statistical methods. Moreover, with analysis of feature maps, and the assistance of a fully connected neural network and a random forest classifier, morphological information in images is systematically proved as the main basis of classification for the deep learning model. The above results indicate that deep learning is valuable for the in‐depth analysis of morphology to better understand subtle changes in cells, which can provide people with deeper insights into cell biology. Abstract : Conventional methods for morphology analysis are constrained in efficiency or accuracy under many circumstances. Deep learning, an emerging tool for biomedical image processing, can accurately classify images of cells grown on different substrates that are undistinguishable to the human eye or conventional statistical methods. In this study, deep learning is proved efficient for the in‐depth analysis of cell morphology.