Abstract
Automatic Liver Imaging Reporting and Data System (LI-RADS) classification for multi-phase CT remains a challenging task due to the complexity of the data and the diversity of the tumors. The AM-PLIFAI Challenge proposes to classify individual lesions using a complex dataset composed of multi-phase CT scans with limited sample size, class imbalance, and missing phases. For this challenge, we propose a method combining image-based features with morphological ones to classify the lesions. Firstly, arterial phase (ART) and portal venous phase (VEN) CT are used as dual-channel inputs to train a 3D segmentation network with lesion-mask supervision, whose encoder extracts lesion-related visual features. Subsequently, a 3D CNN extracts lesion morphological features, while PyRadiomics [1] provides shape, intensity, and texture features; four additional DEL–VEN intensity-change features are also incorporated. Finally, the three feature types are fused for seven-class LI-RADS prediction using a category-weighted loss. The experimental results show that the fusion of the three feature sources achieved the highest final score on the official validation set, and the CT encoder trained with lesion segmentation supervision achieved a higher validation final score than the evaluated general medical image pre-trained encoders. As of 00:00 CEST on August 17, 2026, the proposed method achieved a score on the official validation set that exceeded all results reported on the validation leaderboard and ranked second on the provisional official test leaderboard.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMPLIFAI_006.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=rVaOrayUMo
BibTex
@InProceedings{YanYi_MultiSource_MICCAISAT2026,
author = { Yang, Yi AND Tochon, Guillaume AND Conze, Pierre-Henri AND Géraud, Thierry AND Puybareau, Élodie},
title = { { Multi-Source Feature Fusion for LI-RADS Classification in Multi-Phase 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}
}