Abstract
Hepatocellular carcinoma can be diagnosed directly from multiphasic CT via the Liver Imaging Reporting and Data System (LI-RADS). However, evaluating subtle phase-dependent enhancement patterns causes substantial inter-reader variability. Conventional deep learning models typically regress LI-RADS categories directly, discarding the guidelines’ rule-based structure and proving highly vulnerable to the severe class imbalances inherent in clinical cohorts. To overcome this, we propose a transparent pipeline that explicitly decouples imaging feature prediction from deterministic clinical decoding. At its core, an efficient, context-aware multiphasic vision transformer confines attention to clinically relevant anatomy. A domain-grounded routing mechanism then pools these spatiotemporal embeddings into specific tissue regions to predict individual LI-RADS features. Finally, a probabilistic decoding strategy marginalizes these predictions over the official LI-RADS table. By embedding explicit clinical knowledge into the architecture, our method ensures interpretability and robustness. Our experiments revealed that our proposed multiphasic encoder equipped with the clinically-grounded decoder outperforms alternative methods by over 12 points in overall composite score. Ultimately, our approach achieved a composite score of 0.60 on the external test set, securing first place on the official AM-PLIFAI challenge leaderboard at the time of submission. To support reproducibility, our code and models are publicly available5.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMPLIFAI_008.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=ceMKFpqEAT
BibTex
@InProceedings{AlbMuh_Clinicallygrounded_MICCAISAT2026,
author = { Alberb, Muhammad AND Cheung, Helen M. C. AND Martel, Anne L.},
title = { { Clinically-grounded Multiphasic Modeling for Interpretable LI-RADS Categorization } },
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}
}