Breast cancer is a major disease that poses a serious threat to the lives and health of women. A new framework was proposed to address the common challenges of high dimensional and data imbalances in image classification. This framework integrates particle swarm optimization (PSO) and transfer learning into a convolutional neural network model based on the ResNet34 architecture. The respective strengths complement each other to enhance the performance and efficiency of the classification model. Through parameter optimization and functional selection of PSO, the global search of the model has been improved. Transfer learning lets the model use large pre-trained datasets to learn more quickly on small sample datasets, which is especially helpful in areas where there are a lot of images that don’t have labels. Experimental findings reveal that our framework attains a 97.83% accuracy rate on the dataset and notably shortens the training cycle, demonstrating its effectiveness in improving breast cancer diagnosis performance with small sample sizes.
Read full abstract