Deep Semantic Segmentation of Angiogenesis Images

Area of Science:

• Biomedical Engineering
• Cell Biology
• Medical Imaging

Background:

• Angiogenesis, the formation of new blood vessels, is vital for human health.
• In vitro endothelial cell culture on Matrigel is a common method to study angiogenesis.
• Manual analysis of microphotographs in these studies is time-consuming and labor-intensive.

Purpose of the Study:

• To develop an automated technique for annotating images of capillary-like structures in vitro.
• To address the limitations of manual image analysis in angiogenesis research.
• To apply deep learning for semantic segmentation of angiogenesis images.

Main Methods:

• Development of a tool utilizing a convolutional Unet++ encoder-decoder architecture.
• Semantic segmentation of in vitro angiogenesis simulation images.
• Postprocessing of segmentation masks for expert data analysis.
• Creation and public release of the first annotated dataset in this field, AngioCells.

Main Results:

• Demonstration of the first deep learning-based tool for semantic segmentation of in vitro angiogenesis images.
• Introduction of the AngioCells dataset, the first publicly available annotated dataset for angiogenesis image analysis.
• Establishment of a method to automate the analysis of capillary-like structures.

Conclusions:

• The developed tool and dataset offer a significant advancement in the automation of angiogenesis research.
• Deep learning, specifically the Unet++ architecture, is effective for semantic segmentation of angiogenesis images.
• Automated analysis will accelerate research and improve the efficiency of studying this crucial biological process.