SMC-LUD: Liver Ultrasound Dataset (HCC vs Hemangioma)

The dataset consists of 5,466 anonymized 2D B-mode ultrasound images categorized into two classes. The distribution of images for training and testing is detailed in [Table 1].The full dataset, including class labels and metadata, is publicly available in a data repository under a CC-BY license. It is organized into two main directories according to class: hepatocellular carcinoma (HCC), hemangioma. Within each class directory, images are further grouped by patients, with each patient assigned a unique subdirectory containing all associated ultrasound images.The dataset generated and analyzed during the current study is publicly available for non-commercial research use only, in consideration of patient privacy. This study was approved by the Data Review Board of Samsung Medical Center (SMC) (DRB No. 2025-0111).Table 1. Dataset ClassesDiagnosisSubgroupNo. of PatientsNo. of ImagesTrainValidationTestHemangiomaClean3231,267886253128Caliper3741,402Subtotal4222,669HCCClean4211,389972277178Caliper5571,327Subtotal6002,716Clean (Trained)7442,6561,858530306Total1,0215,3853,7691,076540A metadata file in CSV format is provided, containing the following columns: patient_id, image_filename, class_label, and additional clinical or acquisition metadata as available. Each row in the CSV file corresponds to a single image and provides the relevant annotation and metadata.Table 2. Patient and Tumor CharacteristicsCharacteristicHCC (patients=600)Hemangioma (patients=421)DemographicsAge, years66.65 ± 11.02 (20–90)55.50 ± 12.76 (30–90)Sex, male491 (81.8%)174 (41.3%)Sex, female109 (18.2%)247 (58.7%)Height, cm163.57 ± 16.39163.72 ± 8.45Weight, kg67.44 ± 11.24 (42–104)63.85 ± 12.95 (34–118)Tumor characteristicsMax diameter, cm2.90/2.10/4.50(Median/Q1/Q3)–UICC stage (2000)pT158 (9.7%)–pT2120 (20.0%)–pT3353 (58.8%)–pT469 (11.5%)–AJCC stage (2010)pT1164 (27.3%)–pT2393 (65.5%)–pT322 (3.7%)–pT421 (3.5%)–