A new deep model based on part pooling for medical image classification

This paper proposes an effort to improve the discriminative ability of deep learning model for medical image classification. We formulate this problem as a fine-grained visual categorization task and introduce a deep neural network with part-level features which are trained by independent loss functions. The experiment is conduct on two open-source benchmark dataset. The accuracy and stability of the present model in classification prediction are tested via various metrics, such as accuracy, precision, recall, and Fl-score. Moreover, the learned feature is also visualized via dimensionality reduction technique. It is shown that the proposed network architecture is effective for improving model's performance for medical image classification, and the part-level feature efficiently enriches the granularity of feature increasing the discriminative ability.