Official information site for the MICCAI 2026 AMPLIFAI competition. The challenge has officially concluded.
View Winners Open Codabench Download DataThe AMPLIFAI Challenge has officially concluded. Thank you to all participating teams for their contributions to advancing AI research in multi-phase liver imaging.
| Rank | Team | Team Members | Institution(s) |
|---|---|---|---|
| 🥇 1st | Team Shin | Yuanye Liu | Fudan University |
| 🥈 2nd | epita | Elodie Puybareau, Guillaume Tochon, Yi Yang | EPITA |
| 🥉 3rd | Team Amplifire | Advait Madabhushi, Dharini Raghavan, Amritpal Singh, Gourav Modanwal, Anant Madabhushi | Emory University and Georgia Institute of Technology |
Join us in person at MICCAI 2026 in Strasbourg on Thursday, October 1, 2026, 8:00–10:00 in the Amsterdam room for the AMPLIFAI challenge session, including keynotes, winner talks, proceedings presentations, and awards.
Welcome to the Annotated Multi-Phase Liver Imaging for Artificial Intelligence (AMPLIFAI) Challenge, a Medical Image Computing and Computer Assisted Intervention (MICCAI) 2026 benchmark for clinically meaningful artificial intelligence (AI) in liver lesion assessment on multi-phase computed tomography (CT). AMPLIFAI is built around the Liver Imaging Reporting and Data System (LI-RADS)® framework to support trustworthy and interpretable models for real-world hepatocellular carcinoma (HCC) diagnosis.
The AMPLIFAI task is to classify the LI-RADS score of one pre-defined target lesion per case from multi-phase CT. Public training data is harmonized and preprocessed from four source datasets, with radiologist-provided lesion segmentations and annotations. Final ranking is based solely on classification performance.
The AMPLIFAI challenge invites the medical imaging community to develop AI models capable of classifying lesions across the full LI-RADS spectrum. Participants will be provided with training data containing multi-phase CT volumes, lesion-level LI-RADS category labels, and voxel-level segmentations of key imaging features (e.g., non-rim APHE, non-peripheral washout, and enhancing capsule) of the relevant liver lesion per case. This unique dataset enables both end-to-end and feature-guided approaches to LI-RADS classification; however, models will be evaluated solely on their predicted LI-RADS category, leaving participants free to leverage the provided feature annotations however they see fit.
