Deep Learning Method for Estimating Germ-layer Regions of Early Differentiated Human Induced Pluripotent Stem Cells on Micropattern Using Bright-field Microscopy Image

• Published in: 2024 46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) Vol. 2024; pp. 1 - 4
• Main Authors: Chu, Slo-Li, Yokota, Hideo, Wang, Pai-Ting, Abe, Kuniya, Hayashi, Yohei, Cho, Dooseon, Tsai, Ming- Dar
• Format: Conference Proceeding Journal Article
• Language: English
• Published: United States IEEE 01.07.2024
• Subjects: Accuracy; Cell Differentiation; CNN; Convolution; Deep Learning; Engineering in medicine and biology; Estimation; Germ Layers - cytology; human induced pluripotent stem cell differentiations; Humans; Image Processing, Computer-Assisted - methods; Image segmentation; Immunofluorescence; Induced Pluripotent Stem Cells - cytology; micropattern microscopy images; Microscopy - methods; Safety; Stem cells; Visualization
• ISSN: 2694-0604
• DOI: 10.1109/EMBC53108.2024.10782655

Live cell staining is expensive and may bring potential safety issues in downstream clinical applications, bright-field images rather than staining images should be more suitable to reveal time-series changes of differentiating hiPSCs (human induced pluripotent stem cells) and three-germ layers differentiated from the hiPSCs. This study proposed a deep learning method for estimating immunofluorescence regions on a bright-field microscopy images. The networks trained by multiple types of fluorescence images can estimate the types of fluorescence images from a bright-field image. The estimated pseudo Hoechst image is used to segment hiPSCs, and the others classify the segmented hiPSCs as respective germ-layer cells. The experimental results show over 75% correct rates for the segmentation and classification were achieved, indicating the proposed method can be useful tool in evaluating pluripotency of hiPSC and delineating the germ layer formation process without cell staining.