基于光学显微成像和深度学习的无标记肺癌细胞活力检测方法OA

Lung cancer cell viability detection is widely used in anticancer drug development.How-ever,conventional detection methods can hardly meet the demand for continuous and dynamic moni-toring.In this study,a novel label-free detection method based on microscopic imaging and deep learning is developed.We take the non-small cell lung cancer A549 cell line as the object and con-struct a dataset containing bright-field images and corresponding cell viability annotations.For the al-gorithm,a dual-branch multi-scale feature fusion network Enhanced-UNet is designed,which com-bines the detail recovery capability of UNet++and the context modeling capability of DeepLabV3+.Spatial and channel squeeze-and-excitation and atrous spatial pyramid pooling modules are introduced to strengthen the extraction of detailed and global contextual features.The results show that the pro-posed model achieves a mean intersection over union(mIoU)of 0.762 3 and a mean dice coefficient(mDice)of 0.855 6 on the self-built dataset,with a prediction correlation coefficient r=0.912,out-performing five mainstream models including UNet,SegNet,FCN,PSPNet and LinkNet.Without chemical reagents,this method avoids interference to cell status and enables continuous and dynamic monitoring of A549 cell survival,providing a valuable reference for developing label-free lung cancer cell viability detection technology.