Computer-aided diagnosis system combining FCN and Bi-LSTM model for efficient breast cancer detection from histopathological images

Breast cancer (BC) is one of the most frequent types of cancer that adult females suffer from worldwide. Many BC patients face irreversible conditions and even death due to late diagnosis and treatment. Therefore, early BC diagnosis systems based on pathological breast imagery have been in demand in recent years. In this paper, we introduce an end-to-end model based on fully convolutional network (FCN) and bidirectional long short term memory (Bi-LSTM) for BC detection. FCN is used as an encoder for high-level feature extraction. Output of the FCN is turned to a one-dimensional sequence by the flatten layer and fed into the Bi-LSTM’s input. This method ensures that high-resolution images are used as direct input to the model. We conducted our experiments on the BreaKHis database, which is publicly available at http://web.inf.ufpr.br/vri/breast-cancer-database. In order to evaluate the performance of the proposed method, the accuracy metric was used by considering the five-fold cross-validation technique. Performance of the proposed method was found to be better than previously reported results.