Progressive detail-content-based similarity retrieval over large lung CT image database based on WSLN model

Abstract

With the explosive growth of the medical images, the high-resolution CT image(CI) data is of great value for medical research as well as a clinical diagnosis. Finding CIs similar to the input one from the large- scale CI database can effectively assist physicians to diagnose. The state-of-the-art content-based similarity retrieval of CIs often ignores the effect of image details on retrieval accuracy, and the retrieval accuracy and efficiency are usually not satisfactory. To address this challenge, in this paper, we take lung CI as an example and propose a progressive Detail-Content-based Similarity Retrieval (DCSR) of the lung CIs based on a Weakly Supervised deep Learning Network(WSLN) model. Two enabling techniques (i.e., the WSLN model and the DIndex scheme) are proposed to facilitate the DCSR processing of the large lung CIs. The experimental dataset is from three public datasets. Extensive experiments show that the WSLN-based DCSR method is about 45% more effective than the state-of-the-art methods in terms of the mean average precision(MAP). Meanwhile, the retrieval efficiency of the DIndex scheme is about 200% higher than that of sequential retrieval.