Pulmonary Nodules Image Retrieval via Supervised Deep Hashing

Content-based image retrieval can quickly find images associated with query instance from large-scale medical image databases, which can provide effective aided diagnosis information for doctors, and is beneficial to improve the accuracy of diagnosis. Deep supervised hashing not only has the ability to extract rich features from deep neural networks but also can make full use of supervised information to improve retrieval accuracy. It is also more efficient and memory-saving due to reducing the dimension of the feature vectors by the hash mapping. Therefore, it has attracted much attention. In the paper, a novel end-to-end supervised deep hashing method is proposed, where feature extraction and binary code learning are carried out by joint optimization. The semantic similarity is maintained by label information and the neighborhood structures are preserved by graph regularization. Moreover, the discrete constrained objective function is optimized directly without relaxation. Experiments have been conducted on the pulmonary nodule image dataset and the results demonstrate the proposed method can yield better Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions retrieval performance by comparing with the state-of-the-art hashing methods.