Breast Histopathological Image Classification Method Based on Autoencoder and Siamese Framework

The automated classification of breast cancer histopathological images is one the important tasks in computer-aided diagnosis systems (CADs). Due to characteristics small inter-class and large intra-class variances images, extracting features for difficult. To address this problem, an improved autoencoder (AE) network using a Siamese framework that can learn effective from CAD was designed. First, inputted image processed at multiple scales Gaussian pyramid obtain multi-scale features. Second, feature extraction stage, used constrain pre-trained AE so extracted have smaller variance larger variance. Experimental results show proposed method accuracy as high 97.8% on BreakHis dataset. Compared with commonly algorithms classification, has superior, faster performance.