Medical Image Retrieval Based on Attention Triplet Hashing

With the wide application of X-ray, Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) methods in clinical practice, massive information retrieval and utilization of medical images has become a hot topic. Although traditional methods have shown good results in many specific medical applications, there are still many problems in large-scale medical applications. Deep hash method has been proved to be the most efficient approximate nearest neighbor search technique for large-scale image retrieval. To this end, Attention Triplet Hashing (ATH) network is proposed in this paper, which can further improve retrieval performance and ranking performance of small samples by learning low-dimensional hash codes that retain classification, ROI, and small sample information. We add channel attention to this end-to-end framework to focus on ROI information. And we add label smoothing regularization to distinguish small sample images. Finally, the validity of my framework is tested on a case-based medical dataset.