Breast Cancer Detection Using CNN and Attention Mechanism

Breast Cancer is the most prevalent cause of death in women. Early detection of the disease can increase survival rates. To distinguish benign and malignant breast tissue, the analysis of histopathological images is needed for evaluation; unfortunately, manual assessment is subjective and time-consuming. This study introduces a hybrid deep learning model, which combines ResNet-50 with the Squeeze-and-Excitation (SE) attention feature for better recalibration of feature maps. The model was trained and tested on the BreakHis database, which ranks 7,909 breast tissue images for 40x,100x,200x,400x magnifications and has 5,429 malign and 2,480 benign samples. Performance measures included accuracy, precision, recall, and Fl-score. Results showed that the inclusion of SE mechanisms improved the model's classification performance because it is proved efficient and reliable in distinguishing between benign to malignant samples. This technique holds great potential for real-world automated histopathological diagnosis in clinical applications.