Breast Cancer Classification on Histopathological Images Using Inception V3 and DenseNet 201: A Comparative Study

Inception V3 and DenseNet 201 are widely regarded as excellent architectures. In any case, these two architectures always ranked high during comparisons. However, with different training approaches, it is difficult to choose which is the best architecture between the Inception V3 and the DenseNet 201 architectures. The aim of this study is to compare the performance of both architectures under the same conditions, for binary breast cancer classification on histopathological images. Two models were trained and evaluated on the BreaKHis dataset, the first model uses an Inception V3 as its backbone while the second one uses a DenseNet 201 architecture. The training approach was kept simple so that the results would be influenced mainly by the choice of the feature extractor. As for evaluation, both models were evaluated using the Recall, Accuracy, Precision and F1 score metrics. The performance results were discussed in detail, explaining the relation between each model's performance and the features extractor used. Finally, some conclusions regarding the best outcome are outlined at the end of this paper.