An innovative algorithm combining attention mechanism and feature fusion for circulating tumor cells detection

In this work, we propose an automated detection system for circulating tumor cells (CTCs) identification in spiked samples based on the bright-field microscope images. Specifically, the precise identification of CTCs relied on the modified single-shot multibox detector (SSD)–based neural network. Moreover, we choose attention mechanism and feature fusion for the performance improvement. With this method, the detection performance was considerably boosted with the mean average precision (mAP) value of 91.54% with respect to 85.00% in case of general SSD. It turns out that our model has stronger generalization ability and higher small target detection ability, equipped with a cell counting function, which can assist pathologists in qualitative and quantitative analysis of CTCs in blood visually and accurately.