Breast Cancer is a disease that widely affects millions. A widely recognized approach for diagnosing breast cancer is through the analysis of histopathological images. In order to achieve notable results with these images require building complex ensemble models that are augmented through several techniques, making the process cumbersome. In this paper, the authors investigate famous thresholding techniques: Otsu Threshold, Adaptive Mean Threshold and Adaptive Gaussian Threshold, that are trained on basic transfer learning models: ResNet50, DenseNet201 and Inception v3 to improve diagnosis without creating complex architectures. The Adaptive Gaussian Threshold and Otsu Threshold showed promising results when trained on ResNet50 and Inception v3.