StemQuant: an automated method for quantifying mesenchymal stem cell migration in scratch assays

Mesenchymal stem cells (MSCs) have attracted significant attention in wound healing fields due to their remarkable potential in tissue repair and regeneration. However, analyzing MSC migration in scratch assays remains challenging due to low image contrast, irregular wound shapes, and complex cell morphologies. To address these problems, this study proposes a fully automated MSC migration analysis system that integrates deep learning-based segmentation with end-to-end quantitative analysis. We first constructed the MSC-Scratch-123 time-series dataset, which includes scratch assay images captured under diverse wound orientations and lighting conditions. Subsequently, to improve the segmentation performance, we designed the HiLo-Net architecture, integrating the High-Low Frequency Fusion Block (HLFFB) within the decoder. HLFFB consistently improves segmentation performance across different backbones, achieving the best results of 93.5% mIoU and 96.6% DSC. This improvement is supported by visualization and gradient analysis, which demonstrate enhanced edges and structural details. We further developed an automated migration rate calculation system based on morphological analysis that outputs cell migration rates directly without manual parameter tuning. The method demonstrates strong concordance with manual annotations, with an absolute error below 0.04 and a Pearson correlation coefficient of 0.999. In addition, compared to ImageJ software, this system generates smoother and more complete scratch boundaries, showing significant advantages in both qualitative and quantitative evaluations. In conclusion, the HiLo-Net-based system provides an accurate, efficient, and standardized solution for MSC scratch assays, indicating great potential for applications in stem cell research and wound healing.