Abstract
Automated Liver Imaging Reporting and Data System (LI-RADS) classification from multi-phase computed tomography (CT) requires integrating 3D lesion appearance with dynamic enhancement patterns across contrast phases. This task is particularly challenging because of the strong class imbalance and the frequent absence of one or more imaging phases. We present a supervised seven-class LI-RADS classification framework developed for the AMPLIFAI challenge. Each available CT phase and the challenge-provided lesion mask are processed by a shared 3D encoder, while phase embeddings and a Transformer model inter-phase relationships and explicitly handle missing acquisitions. Training combines inverse-frequency weighted cross-entropy, ordinal regularization, manifold mixup, and auxiliary prediction of radiological attributes including arterial phase hyperenhancement, washout, and capsule appearance. These attributes are used exclusively as auxiliary supervision and are not required at inference. We evaluate ResNet18 and ConvNeXt3D as 3D encoders. ResNet18 achieved the highest validation score (0.531 vs. 0.521), while ConvNeXt3D performed best on the test set (0.230 vs. 0.210). These results underscore the difficulty of LI-RADS classification in the presence of class imbalance and missing CT phases.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMPLIFAI_001.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=NaujPzdDzd
BibTex
@InProceedings{ConPie_Ordinalaware_MICCAISAT2026,
author = { Conze, Pierre-Henri AND Puybareau, Élodie AND Badic, Bogdan AND Jaouen, Vincent AND Visvikis, Dimitris},
title = { { Ordinal-aware Multi-Phase Classification with Manifold Mixup for LI-RADS Assessment in CT } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17271},
month = {pending},
page = {pending}
}