Abstract
Automated Liver Imaging Reporting and Data System (LI-RADS) assessment from multiphase CT is not naturally a standard seven-class classification problem. LR-1–LR-5 form an ordered spectrum of increasing hepatocellular-carcinoma likelihood, whereas LR-M and LR-TIV represent clinically distinct outcomes that leave this ordinal continuum. We present PHORA, a physiology-guided hierarchical radiomics framework that models these two aspects of the task separately while retaining clinically recognizable intermediate evidence. PHORA combines conventional radiomics with spatial and temporal enhancement descriptors, distills training-only radiologist annotations into inference-time estimates of major LI-RADS concepts, and integrates special-category recognition with ordinal severity estimation through a clinically constrained router. On the 59-case public validation set, PHORA achieved a composite score of 0.718, adjusted quadratic weighted kappa of 0.738, and special-category recognition of 0.605. Among reference ordinal cases retained on the ordinal pathway, 30 of 31 predictions were within one LI-RADS category of the reference, indicating that residual staging errors were usually local rather than severe. The concept models recovered clinically important features such as non-rim arterial phase hyperenhancement and delayed washout with AUROCs of 0.915 and 0.905, respectively, while prediction accuracy decreased from 0.800 in the lowest-uncertainty quartile to 0.467 in the highest. These results show that PHORA provides a clinically structured and auditable formulation of LI-RADS prediction in which ordinal severity, special-category evidence, latent imaging concepts, routing logic, and model uncertainty can all be examined rather than collapsing the task into a single nominal class decision.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMPLIFAI_007.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=2TwGxdPBvf
BibTex
@InProceedings{RagDha_PHORA_MICCAISAT2026,
author = { Raghavan, Dharini AND Madabhushi, Advait AND Singh, Amritpal AND Lebowitz, Mendel AND Modanwal, Gourav AND Madabhushi, Anant},
title = { { PHORA: Physiology-Guided Hierarchical Ordinal Radiomics for LI-RADS Classification in Multiphase 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}
}