Abstract
Liver tumors exhibit significant heterogeneity in etiology, pathology, and treatment response, making accurate differential diagnosis critical for diagnosis and management. While multi-phase contrast-enhanced computed tomography (CT) provides valuable imaging patterns for differentiation, visual assessment alone is often limited by overlapping features. To address this, we present MCT-LTDiag, a comprehensive Multi-phase CT dataset for Liver Tumor Diagnosis, comprising 517 cases with four-phase contrast-enhanced CT scans (non-contrast, arterial, portal venous, and delayed phases) and five tumor subtypes: hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma (ICC), colorectal liver metastasis (CRLM), breast cancer liver metastasis (BCLM), and hepatic hemangioma (HH). The dataset features standardized preprocessing, rigorous quality control, and expert-annotated tumor masks. Baseline experiments using radiomics-based machine learning and deep learning models demonstrate the dataset's utility, with multi-phase integration significantly improving diagnostic performance. MCT-LTDiag serves as a benchmark for advancing automated liver tumor subtype classification and is publicly available to support future research.