MyoV: a deep learning-based tool for the automated quantification of muscle fibers
Shuang Gu,Chaoliang Wen,Zhen Xiao,Qiang Huang,Zheyi Jiang,Honghong Liu,Jia Gao,Junying Li,Congjiao Sun,Ning Yang
DOI: https://doi.org/10.1093/bib/bbad528
IF: 9.5
2024-01-27
Briefings in Bioinformatics
Abstract:Accurate approaches for quantifying muscle fibers are essential in biomedical research and meat production. In this study, we address the limitations of existing approaches for hematoxylin and eosin-stained muscle fibers by manually and semiautomatically labeling over 660 000 muscle fibers to create a large dataset. Subsequently, an automated image segmentation and quantification tool named MyoV is designed using mask regions with convolutional neural networks and a residual network and feature pyramid network as the backbone network. This design enables the tool to allow muscle fiber processing with different sizes and ages. MyoV, which achieves impressive detection rates of 0.93–0.96 and precision levels of 0.91–0.97, exhibits a superior performance in quantification, surpassing both manual methods and commonly employed algorithms and software, particularly for whole slide images (WSIs). Moreover, MyoV is proven as a powerful and suitable tool for various species with different muscle development, including mice, which are a crucial model for muscle disease diagnosis, and agricultural animals, which are a significant meat source for humans. Finally, we integrate this tool into visualization software with functions, such as segmentation, area determination and automatic labeling, allowing seamless processing for over 400 000 muscle fibers within a WSI, eliminating the model adjustment and providing researchers with an easy-to-use visual interface to browse functional options and realize muscle fiber quantification from WSIs.
biochemical research methods,mathematical & computational biology