• Pediatric Personalized Deep Le...

*Result*: Pediatric Personalized Deep Learning Models for Segmentation of Hepatoblastoma at CT and MRI.

Title:
Pediatric Personalized Deep Learning Models for Segmentation of Hepatoblastoma at CT and MRI.
Authors:
Modanwal G; Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Health Sciences Research Building II, 1750 Haygood Dr, Ste N647, Atlanta, GA 30322., Kumar S; Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Health Sciences Research Building II, 1750 Haygood Dr, Ste N647, Atlanta, GA 30322., Viswanathan V; Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Health Sciences Research Building II, 1750 Haygood Dr, Ste N647, Atlanta, GA 30322., Morin CE; Department of Radiology, Cincinnati Children's Hospital, Cincinnati, Ohio.; Department of Radiology, University of Cincinnati College of Medicine, Cincinnati, Ohio., Rees MA; Department of Radiology, Nationwide Children's Hospital, Columbus, Ohio., Squires JH; Department of Radiology, UPMC Children's Hospital of Pittsburgh, Pittsburgh, Pa., Tang ER; Department of Radiology, Children's Hospital of Colorado, Aurora, Colo., Katzenstein HM; Nemours Children's Hospital, Wilmington, Del., Towbin AJ; Department of Radiology, Cincinnati Children's Hospital, Cincinnati, Ohio.; Department of Radiology, University of Cincinnati College of Medicine, Cincinnati, Ohio., Madabhushi A; Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Health Sciences Research Building II, 1750 Haygood Dr, Ste N647, Atlanta, GA 30322.; Atlanta Veterans Administration Medical Center, Atlanta, Ga., Schooler GR; Department of Radiology, Cincinnati Children's Hospital, Cincinnati, Ohio.; Department of Radiology, University of Cincinnati College of Medicine, Cincinnati, Ohio.
Source:
Radiology. Imaging cancer [Radiol Imaging Cancer] 2026 Mar; Vol. 8 (2), pp. e250041.
Publication Type:
Journal Article; Multicenter Study
Language:
English
Journal Info:
Publisher: Radiological Society of North America, Inc Country of Publication: United States NLM ID: 101765309 Publication Model: Print Cited Medium: Internet ISSN: 2638-616X (Electronic) Linking ISSN: 2638616X NLM ISO Abbreviation: Radiol Imaging Cancer Subsets: MEDLINE
Imprint Name(s):
Original Publication: Oak Brook, IL : Radiological Society of North America, Inc., [2019]-
MeSH Terms:
Hepatoblastoma*/diagnostic imaging , Magnetic Resonance Imaging*/methods , Tomography, X-Ray Computed*/methods , Liver Neoplasms*/diagnostic imaging , Deep Learning*, Humans ; Male ; Female ; Child, Preschool ; Child ; Infant
Contributed Indexing:
Keywords: Abdomen/GI; Algorithm Development; Deep Learning; Liver; MR-Imaging; Pediatrics
Entry Date(s):
Date Created: 20260220 Date Completed: 20260220 Latest Revision: 20260220
Update Code:
20260221
DOI:
10.1148/rycan.250041
PMID:
41718532
Database:
MEDLINE

*Further Information*

*Purpose To evaluate the generalizability of adult-trained models for hepatoblastoma segmentation to pediatric patients and to develop two deep learning (DL) models, INLINEMATH and INLINEMATH , specifically trained on pediatric contrast-enhanced CT and T2-weighted MRI scans, respectively. Materials and Methods Imaging data from the multicenter Children's Oncology Group AHEP0731 trial (NCT00980460; May 2008-July 2018) were analyzed. DL models employing the three-dimensional U-Net architecture were trained using D<subscript>CT-Train</subscript> and D<subscript>MRI-Train</subscript>. These models were evaluated on D<subscript>CT-Val</subscript> and D<subscript>MRI-Val</subscript> using the Dice similarity coefficient (DSC), and model segmentations were compared with manual segmentations from three annotators (R<subscript>1</subscript>, R<subscript>2</subscript>, and R<subscript>3</subscript>), their consensus (R<subscript>c</subscript>), and adult-trained model ( INLINEMATH ) segmentations. Volume percentage error analysis was performed to evaluate segmentation precision. Results A total of 104 participants (mean age ± SD, 28.2 months ± 30.5; 64 male; D<subscript>CT-Train</subscript> = 56, D<subscript>CT-Val</subscript> = 48) were included in the CT dataset and 123 (31.5 months ± 38.4; 87 male; D<subscript>MRI-Train</subscript> = 50, D<subscript>MRI-Val</subscript> = 73) in the MRI dataset. INLINEMATH achieved good agreement with consensus segmentation (DSC = 0.86 [95% CI: 0.80, 0.91]) and exhibited higher agreement than INLINEMATH with R<subscript>1</subscript> (0.83 vs 0.55), R<subscript>2</subscript> (0.85 vs 0.55), R<subscript>3</subscript> (0.84 vs 0.54), and R<subscript>c</subscript> (0.86 vs 0.55) segmentations. Volume percentage error analysis revealed that INLINEMATH achieved segmentation results on par with or better than those of a novice annotator (R<subscript>3</subscript>) in high-precision scenarios. INLINEMATH also achieved a DSC of 0.86, demonstrating good agreement with R<subscript>c</subscript>. Conclusion The pediatric-trained DL-based models outperformed adult-trained models for accurate segmentation of pediatric hepatoblastoma. Keywords: Pediatrics, Deep Learning, Liver, MR-Imaging, Abdomen/GI, Algorithm Development ClinicalTrials.gov NCT00980460 Supplemental material is available for this article. © The Author(s) 2026. Published by the Radiological Society of North America under a CC BY 4.0 license.*