Cancer cell smear image segmentation model based on EfficientNet-B0 collaborating with Gaussian pyramid

Cancer represents a severe hazard to human well-being, and the precise segmentation of cancer cells plays a pivotal role in cancer diagnosis and therapy. In order to enhance the accuracy and efficiency of cancer cell segmentation, this study proposes the EB0-SEGP (EfficientNet-B0 SE Block Gaussian Pyramid) model constructed based on EfficientNet-B0 and the Gaussian pyramid. The model combines the channel attention mechanism (SE Block), up-sampling layers, and NIN blocks, and is optimized on the foundation of the traditional U-Net framework. During the research process, cancer cell smear images were collected from the open-source website CSDN, and after pre-processing and data augmentation, Dataset A was constructed. Accuracy, loss value, and Dice coefficient were used as indicators to evaluate the performance of the model. The research results show that the EB0-SEGP model achieves an accuracy of 88%, a Dice coefficient of 0.88, and a loss value of 0.06 on the test set, demonstrating a notable improvement compared to U-Net and ERU-Net. Meanwhile, the standard deviation of the training curve is significantly lower than that of the comparison models, indicating a lighter degree of overfitting. This study provides a new technical means for cancer treatment and has important theoretical significance and clinical application value in the field of cancer diagnosis and treatment.