Image classification toward breast cancer using deeply-learned quality features

Abstract

Image classification plays an important role in computer vision and its applications, such as scene categorization, image retrieval. Convolutional neural network based methods have shown competitive performance in image classification, which aims to exploit deep feature of training images. In this paper, based on CNN methods and image quality assessment (IQA) algorithms, we propose a novel method for medical application, that is breast cancer classification. First, we leverage CNN architecture to calculate the number of pixels in the lesions, where maximum pooling layers are used. Then, large density of pixel regions will be assigned with large quality scores, which reflect more texture and grayscale features. Finally, we construct a multi-SVM based image kernel using obtained quality scores to achieve breast cancer classification. Experimental results show our proposed method outperforms single recognition based image classification methods such as pixel grayscale or gradient.