Twelve hospitals from three continents are retrospectively collecting routine MRI scans of pathologically confirmed solid-appearing liver lesions. With no restrictions on age, sex, or underlying liver diseases, the project aims to gather anonymised clinical and imaging data from over 3000 patients. The dataset will include common and rare types of malignant and benign lesions. With multiple MRI sequences per patient, this will be the first and largest publicly available MRI dataset focused on solid-appearing liver lesions. This dataset will be used to develop AI models for comprehensive MRI-based liver lesion phenotyping through robust automated machine learning (AutoML). A Grand Challenge will be organized to compare different methods in a fair and unbiased way, inviting research teams worldwide to participate by submitting their algorithms. Subsequently, we will rigorously validate the most promising AI models in increasingly realistic clinical settings.