The imaging data are organized following a MIDS structure ( see [https://doi.org/10.48550/arXiv.2010.00434]). Each subject has 3 sub-folders for T1-WI (dyn), T2-WI (anat) and diffusion imaging (dwi). • Anat: contains T2-WI files named “HASTE” or “blade” for fat-suppressed imaging. • Dyn: contains T1-WI files named “native”, “arterial”, “venous” and “delayed” for DCE images. Files from patients with triple arterial images were differentiated with the following naming: TTC_1, TTC_2 and TTC_3. Each patient has at least 4 T1-WI with fat suppression called “water”. In addition, some patient included in-phase and out-of-phase imaging named “inphase” and “outphase”. • Dwi: contains DWI files named “ADC” for apparent coefficient diffusion and “TRACE” for calculated images with different b-values. Each TRACE image is named with a “run” number corresponding to the b-value used to create the image in an increasing order. The b-values can be retrieved in the corresponding json file for each patient. The derivatives directory includes • Automated liver annotations: The liver masks obtained from a trained nnU-Net for liver segmentation are provided. The liver was segmented in each T1-WI phase (native, arterial, venous and delayed) with the nnU-Net. These masks were used to calculate the registration transforms. Each file contains a binary mask of the liver with the phase of reference in the file name.• Manual lesion annotations: The lesions were annotated on each phase when visible. Each file is a binary mask for a lesion with the phase of reference and the lesion index in the file name (e.g. “sub-001_acq-water_phase-arterial-TTC-3_T1w-L1_seg.nii” for the lesion index 1 of the subject 001 in the participants.tsv file).• Manual liver annotations: The liver was annotated for 16 patients in T1-WI venous phase and in fat-suppressed T2-WI. Each file contains a binary mask of the liver with the image of reference in the file name.• T1-WI registration transforms: Registration transforms for T1-WI alignment were computed and saved in the “T1_registration_transforms” derivative. Python scripts can be found in the “code” folder to apply the transforms on the images and segmentations.

Hepatocellular carcinoma (HCC) is the most common primary liver cancer and a leading cause of cancer-related mortality worldwide. While dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) plays a key role for HCC screening, diagnosis and treatment planning, the development of AI models for detection and characterization of focal liver lesions is hindered by the scarcity of publicly available annotated MRI datasets. Here, we present OpenSwissHCC, a curated multiphasic dataset of 132 multiparametric DCE-MRI scans from adult patients with chronic liver disease acquired on 1.5T and 3T MRI scanners between 2011 and 2020. The dataset includes both HCC-positive (n=63) and HCC-negative (n=69) patients, with up to 5 lesions per patient segmented across each sequence when confidently identified. Whole-liver and lesion segmentations were manually performed and expert-validated, with lesion metadata including LI-RADS scores size liver segment location major LI-RADS imaging features (non-rim APHE, washout, capsule, threshold growth) key ancillary/background findings (e.g., mosaic architecture, fat/blood products, necrosis, tumor-in-vein; cirrhosis and portal-hypertension signs) In total, 140 lesions were annotated, including 97 confirmed HCC. Additionally, we provide liver masks for all T1-weighted images (WI) sequences, generated using a pre-trained nnU-Net model, as well as pre-computed transforms to facilitate registration across T1-WI phases. By providing a comprehensive, well-annotated, and standardized collection of liver MRI examinations, OpenSwissHCC aims to assist the development and benchmarking of advanced computational methods for automated HCC detection, segmentation, and characterization, fostering progress toward precision imaging and personalized treatment planning.