This study focuses on the task of cell image classification and automatic labeling, and proposes a deep learning model that integrates attention mechanism. The model is based on Dual-Attention Co-Learning Network (DACN) with dual-branch architecture, which includes multi-scale feature extraction and dual-path attention mechanism. Through the synergistic effect of space and channel attention, the feature expression of key areas is effectively strengthened, while noise interference is suppressed. In addition, a collaborative learning framework of classification and labeling is constructed, and high-quality pseudo-labeling is generated by weak supervised learning, which improves the efficiency and accuracy of labeling. Experimental results on public and clinical data sets show that the accuracy of the proposed model in classification tasks can reach 98.2%, and the F1-Score is 0.968. Dice coefficient reaches 0.927, Hausdorff distance is as low as 6.7 pixels, and it shows good diagnostic ability on cell images with different pathological grades, which provides an efficient and accurate solution for the field of cell image analysis.
Ju et al. (Sun,) studied this question.