Deep Learning Method for Content-Based Retrieval of Focal Liver Lesions Using Multiphase Contrast-Enhanced Computer Tomography Images

Abstract

A content-based image retrieval (CBIR) system can support radiologists in making clinical diagnosis through image analysis. Multiphase contrast-enhanced computer tomography (CT) images are more effective than single contrast-enhanced CT images in detecting and characterizing focal liver lesions (FLLs). This study proposes a deep learning method for the CBIR of FLLs using multiphase contrast-enhanced CT images. We use deep convolutional neural networks (DCNNs) to extract the temporal– spatial features from multiphase CT images. Compared with the conventional low- and mid-level features, the high-level features extracted by the DCNN can significantly improve the retrieval accuracy. The effectiveness of the proposed method was demonstrated through experiments with our multiphase FLL CT dataset, which is called as MPCT-FLL dataset. The mean average precision (mAP) was improved from 0.76 to 0.84.