AMPLIFAI: A Multiphase CT Dataset for Benchmarking Clinical Reasoning in LI-RADS Assessment of Liver Lesions

Pranav Kulkarni; Nikhil Shah; Amritansh Suryavanshi; Jana G. Delfino; James Tonascia; Jade Wong-You-Cheong; Barton Lane; Joseph Chirico

cs.CVAug 14, 2026

AMPLIFAI: A Multiphase CT Dataset for Benchmarking Clinical Reasoning in LI-RADS Assessment of Liver Lesions

Authors: Pranav Kulkarni, Nikhil Shah, Amritansh Suryavanshi, Jana G. Delfino, James Tonascia, Jade Wong-You-Cheong, Barton Lane, Joseph Chirico,

Organizations: University of Maryland Institute for Health Computing, North Bethesda, MD 20852 · Department of Computer Science, University of Maryland, College Park, MD 20742 · Department of Diagnostic Radiology and Nuclear Medicine, University of Maryland School of Medicine, Baltimore, MD 21201 · Department of Electrical and Computer Engineering, University of Maryland, College Park, MD 20742 ·

Abstract

Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related mortality worldwide, with early detection improving survival from <20% to >70%. The standardized Liver Imaging Reporting and Data System (LI-RADS) criteria provide an imaging-based diagnostic framework to evaluate liver lesions for HCC, serving as a foundation for automating HCC detection with artificial intelligence (AI). However, the lack of large, publicly available datasets with high-quality annotations has limited the development and evaluation of AI models for automated LI-RADS assessment. We introduce AMPLIFAI dataset, the first public dataset of 590 multiphase abdominal CT studies annotated with LI-RADS categories, lesion size, and voxel-level segmentations for three major LI-RADS features: arterial phase hyperenhancement, washout, and enhancing capsule. The dataset was curated and harmonized from four public datasets and augmented with expert annotations from five board-certified radiologists and one resident. Following the Datasheets for Datasets format, this paper details the dataset's composition, curation and harmonization process, and annotation workflow to support transparent, reproducible research in medical imaging AI.

Figures & tables

Figure 1: Dynamic contrast enhancement patterns of liver lesions in multiphase abdominal CT before contrast injection (non-contrast) and at successive time points after injection: arterial, portal venous, and delayed phases.

Figure 2: An overview of the standardized LI-RADS criteria 1 1 1 Figure inspired by Radiology Assistant: https://radiologyassistant.nl/abdomen/liver/li-rads . Categories LR-1 to LR-5 describe the increasing probability of HCC. LR-M indicates a lesion that is malignant but not specific for HCC, while LR-TIV denotes definite tumor in vein. Major LI-RADS features include lesion size (in mm), non-rim arterial phase hyperenhancement (APHE), non-peripheral washout, and enhancing capsule.

Figure 3: Representative examples from each LI-RADS category. (a) The LR-1 image displays a lesion with no enhancement. (b) The LR-2 image shows a small (10 mm) lesion with non-rim APHE but no malignant features. (c) The LR-3 image shows an intermediate (30 mm) lesion with unequivocal malignant features. (d) The LR-4 image shows a large (93 mm) lesion with non-rim APHE. (e) The LR-5 image shows a large (63 mm) lesion with both non-rim APHE and non-peripheral washout. (f) The LR-M image shows a large (30 mm) lesion with rim APHE but lacks malignant features specific for HCC. (g) the LR-TIV image shows a large (51 mm) lesion with soft tissue present in the vein.

Source	Training	Validation	Total
PLC-CECT [ 14 ]	$277$ ( $52.17\%$ )	$29$ ( $49.15\%$ )	$306$ ( $51.86\%$ )
WAW-TACE [ 12 ]	$176$ ( $33.15\%$ )	$23$ ( $38.98\%$ )	$199$ ( $33.73\%$ )
HCC-TACE-SEG [ 13 ]	$41$ ( $7.72\%$ )	$4$ ( $6.78\%$ )	$45$ ( $7.63\%$ )
TCGA-LIHC [ 11 ]	$37$ ( $6.97\%$ )	$3$ ( $5.08\%$ )	$40$ $(6.78\%$ )
Total	$531$ ( $100\%$ )	$59$ ( $100\%$ )	$590$ ( $100\%$ )

Table 1: Distribution of cases across source datasets for training and validation splits.

Phase	Non-contrast	Arterial	Portal Venous	Delayed
Training	$381$ ( $71.75\%$ )	$531$ ( $100\%$ )	$527$ ( $99.25\%$ )	$406$ ( $76.46\%$ )
Validation	$43$ ( $72.88\%$ )	$59$ ( $100\%$ )	$59$ ( $100\%$ )	$45$ ( $76.27\%$ )
Total	$424$ ( $71.86\%$ )	$590$ ( $100\%$ )	$586$ ( $99.32\%$ )	$451$ ( $76.44\%$ )

Table 2: Distribution of contrast phases across training and validation splits.

Figure 4: A multiphase abdominal CT with segmentation masks for the target lesion and three major LI-RADS features: arterial phase hyperenhancement (APHE), non-peripheral washout, and enhancing capsule.

Variables	Training	Validation	Total
Lesion size (in mm)	$57.83\pm 43.31$	$56.81\pm 36.98$	$57.72\pm 42.79$
APHE
Non-rim APHE	$258$ ( $48.59\%$ )	$33$ ( $55.93\%$ )	$291$ ( $49.32\%$ )
Rim APHE	$62$ ( $11.68\%$ )	$7$ ( $11.86\%$ )	$69$ ( $11.69\%$ )
Absent	$211$ ( $39.73\%$ )	$19$ ( $32.20\%$ )	$230$ ( $38.98\%$ )
Washout

Table 3: Distribution of annotations across training and validation splits.

Figure 5: Overview of the annotation workflow. (a) We curated and harmonized 590 cases from four public datasets, using available liver lesion segmentations as initial seeds. (b) Each case was independently annotated by a radiology resident and one of three attending radiologists, including LI-RADS category, lesion size, and segmentation of three major LI-RADS features. An iteratively trained nnU-Net was used to update segmentation seeds throughout annotation process. (c) Cases with two independent annotations were reviewed by one of two independent validators, who selected or combined the annotations. Cases with disagreements were returned to the annotators for correction before final validation.

Figure 6: Overview of the custom 3D Slicer module for the radiologist annotation workflow.

Medical Image Benchmarks

Medical image benchmarks compare AI methods on curated clinical images, tasks, and evaluation protocols to measure accuracy, robustness, and clinical usefulness.

See all 337 papers

An interpretable deep learning method for early hepatocellular carcinoma diagnosis on multiphase CT
Oct 21, 2026Emma Sarfati, Alexandre Bône, Pietro Gori +8Medical Diagnosis Medical Image Classification
Guerbet Research, Villepinte, France · LTCI, Télécom Paris, Institut Polytechnique de Paris, Saclay, France · MAP5 (UMR 8145), Université Paris Cité, 45 rue des Saints-Pères, Paris, France +7
Semi-Supervised Learning-Based Genetic Biomarkers Dataset for Multiple-Stage Hepatocellular Carcinoma Prediction
Sep 15, 2026Ahmed Ammar Kubba, Manar Abu Talib, Jibran Sualeh Muhammad +4Cancer Genomics Genomics
Department of Computer Science, University of Sharjah, Sharjah, UAE · Department of Biomedical Sciences, University of Birmingham, Birmingham, United Kingdom · Department of Computer Engineering, University of Sharjah, Sharjah, UAE +1
CT-Bench: A Benchmark for Multimodal Lesion Understanding in Computed Tomography
Feb 16, 2026Qingqing Zhu, Qiao Jin, Tejas S. Mathai +10Medical Image Benchmarks Computed Tomography

Oct 21, 2026J/K move · Enter open · S save

An interpretable deep learning method for early hepatocellular carcinoma diagnosis on multiphase CT

Emma Sarfati, Alexandre Bône, Pietro Gori, Sisi Yang+7

Guerbet Research, Villepinte, France · LTCI, Télécom Paris, Institut Polytechnique de Paris, Saclay, France · MAP5 (UMR 8145), Université Paris Cité, 45 rue des Saints-Pères, Paris, France+7

Medical Diagnosis Medical Image Classification Explainable Medical Image Analysis Computed Tomography

Open paper AskPDF arXiv

Early diagnosis of hepatocellular carcinoma is critical to preserve access to curative treatment options. The Liver Imaging Reporting and Data System (LI-RADS) provides a standardized framework for non-invasive diagnosis on multiphase computed tomography, based on major visual imaging features. However, these features may be difficult to assess consistently, particularly in small liver lesions. We developed dLI-RADS, an interpretable artificial intelligence method designed to estimate the probability of hepatocellular carcinoma from multiphase computed tomography images of manually segmented liver lesions, while remaining aligned with LI-RADS-related imaging concepts. dLI-RADS combines neural network-derived predictions with handcrafted quantitative imaging measurements reflecting LI-RADS major features and integrates these inputs using logistic regression. In this multicenter retrospective study, patients with multiphase computed tomography and at least one liver lesion were included, while lesions larger than 40 mm were excluded. Histology served as the reference standard for hepatocellular carcinoma, and a composite reference standard was used for non-hepatocellular carcinoma lesions. The lesion-level accuracy of dLI-RADS and LI-RADS were compared on the internal evaluation set using 10,000-repetition bootstrap non-inferiority tests, with 5% margins and significance levels. On the external test set, sensitivities were similarly compared. dLI-RADS was non-inferior to LI-RADS for lesion-level accuracy in the internal evaluation set, achieving 80% accuracy compared with 74% for LI-RADS (p = 0.04)...

Abstract

Medical Image Benchmarks

Explore similar work

An interpretable deep learning method for early hepatocellular carcinoma diagnosis on multiphase CT

Semi-Supervised Learning-Based Genetic Biomarkers Dataset for Multiple-Stage Hepatocellular Carcinoma Prediction

CT-Bench: A Benchmark for Multimodal Lesion Understanding in Computed Tomography