Development of an Automated Image Segmentation Platform for Cell Migration Assays Based on Artificial Intelligence

Abstract

Biomedical image analysis has significantly benefited from advances in artificial intelligence, particularly through the application of convolutional neural networks (CNNs) for cellular structure segmentation. This study proposed the development of a web-based computer vision tool to automate the analysis of cell migration assays (wound healing), replacing the traditional manual method performed using ImageJ. A total of 716 images were collected, converted, normalized, resized, and expanded through data augmentation techniques, resulting in 4,160 images used for training. Segmentation annotations were performed using Roboflow with the Smart Polygon tool and the Segment Anything Model 2 (SAM2). The final tool was made available through an interactive interface on the Hugging Face Spaces platform, enabling users to upload and automatically analyze images with graphical overlay of the results.

The effectiveness of the tool was evaluated by comparing manual and automated analyses across datasets reviewed by three independent operators. The Roboflow fast model demonstrated superior performance compared to YOLOv11 fast (mAP@50 = 94.9; precision = 99.6%; recall = 87.7%) and improved inter-operator consistency, reducing the variability observed in the traditional manual method. These results demonstrate that the automated solution significantly reduces analysis time, eliminates subjectivity, and enhances reproducibility, establishing itself as an efficient, standardized, and accessible alternative for cell migration studies.