Search

• Ensemble of MultiScale Networks (EMS-Net) for breast histology image classification. • Extract visual features using multiple deep convolutional networks at multiscale. • Select the optimal subset of fine-tuned deep models and combine them to EMS-Net. • Achieved an accuracy of 90.00% on the BACH Challenge classification task (Part A). Histology images analysis resulted from needle biopsy serves as the gold standard for breast cancer diagnosis. Deep learning-based classification of breast tissues in histology images, however, is less accurate, due to the lack of adequate training data and ignoring the multiscale structural and textural information. In this paper, we propose the Ensemble of MultiScale convolutional neural Networks (EMS-Net) to classify hematoxylin–eosin stained breast histopathological microscopy images into four categories, including normal tissue, benign lesion, in situ carcinoma, invasive carcinoma. We first convert each image to multiple scales, and then use the training patches cropped and augmented at each scale to fine-tune the pre-trained DenseNet-161, ResNet-152, and ResNet-101, respectively. We find that a combination of three fine-tuned models is more accurate than other combinations, and use them to form an ensemble model. We evaluated our algorithm against three recent methods on the BACH challenge dataset. It shows that the proposed EMS-Net algorithm achieved an accuracy of 91.75 ± 2.32% in the five-fold cross validation using 400 training images, which is higher than the accuracy of other three algorithms, and also achieved an accuracy of 90.00% in the online verification using 100 testing images. [ABSTRACT FROM AUTHOR]

Gradient descent is prevalent for large-scale optimization problems in machine learning; especially it nowadays plays a major role in computing and correcting the connection strength of neural networks in deep learning. However, many gradient-based optimization methods contain more sensitive hyper-parameters which require endless ways of configuring. In this paper, we present a novel adaptive mechanism called adaptive exponential decay rate (AEDR). AEDR uses an adaptive exponential decay rate rather than a fixed and preconfigured one, and it can allow us to eliminate one otherwise tuning sensitive hyper-parameters. AEDR also can be used to calculate exponential decay rate adaptively by employing the moving average of both gradients and squared gradients over time. The mechanism is then applied to Adadelta and Adam; it reduces the number of hyper-parameters of Adadelta and Adam to only a single one to be turned. We use neural network of long short-term memory and LeNet to demonstrate how learning rate adapts dynamically. We show promising results compared with other state-of-the-art methods on four data sets, the IMDB (movie reviews), SemEval-2016 (sentiment analysis in twitter) (IMDB), CIFAR-10 and Pascal VOC-2012. [ABSTRACT FROM AUTHOR]