Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/dataset.json

Name

Pediatric low-grade glioma T2-weighted brain MRI (CBTN/DFCI-BCH) used for stepwise transfer learning segmentation

Link

https://pmc.ncbi.nlm.nih.gov/articles/PMC11294948

Indexing

Keywords: pLGG, pediatric brain tumor, T2-weighted MRI, glioma, segmentation, transfer learning

Content: NR, MR, PD

RadLex: RID4071, RID12778, RID49946, RID11221, RID4412, RID10796

Organization(s)

Children’s Brain Tumor Network (CBTN)

Dana-Farber Cancer Institute/Boston Children’s Hospital (DFCI/BCH)

Mass General Brigham / Harvard Medical School (AIM Program)

Funding

Supported in part by NIH (U54CA274516, U24CA194354, U01CA190234, U01CA209414, R35CA22052, K08DE030216), NCI SPORE (2P50CA165962), European Research Council (866504), RSNA (RSCH2017), Pediatric Brain Tumor Foundation (Pediatric Low-Grade Astrocytoma Program), Botha-Chan Low Grade Glioma Consortium, William M. Wood Foundation, Brigham and Women’s Hospital, Harvard Medical School, Dana-Farber Cancer Institute, Boston Children’s Hospital, NIH Loan Repayment Program (L40CA264321).

Ethical review

Study conducted under local IRB approval with waiver of consent due to use of public datasets and retrospective design.

Comments

Retrospective multi-institutional T2-weighted brain MRI dataset of pediatric low-grade glioma (pLGG) drawn from the Children’s Brain Tumor Network (CBTN) and Dana-Farber Cancer Institute/Boston Children’s Hospital (DFCI/BCH) used to develop and evaluate stepwise transfer learning segmentation models.

References

[1] Boyd A, Ye Z, Prabhu SP, et al.. "Stepwise Transfer Learning for Expert-level Pediatric Brain Tumor MRI Segmentation in a Limited Data Scenario". Radiology: Artificial Intelligence. 2024-07-10. doi:10.1148/ryai.230254. PMID: 38984985. PMCID: PMC11294948.

[2] Baid U, Ghodasara S, Bilello M, et al.. "The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor Segmentation and Radiogenomic Classification". arXiv. 2021-07-05. Available from: https://arxiv.org/abs/2107.02314

Dataset

Motivation

To develop and externally test a deep learning autosegmentation model for pLGG using stepwise transfer learning and to evaluate clinical acceptability.

Sampling

Inclusion: age 0–25 years, histopathologically confirmed pLGG (grade I–II), preoperative brain MRI with a T2-weighted sequence; time frame May 2001–December 2015; manual QC to remove poor-quality scans.

Partitioning scheme

CBTN development set randomly split into training (n=124) and blinded internal hold-out test (n=60). External testing performed on independent DFCI/BCH set (n=100), with a subset of 60 scans having expert reference segmentations for quantitative testing; 100 scans used for clinical acceptability testing.

Missing information

No spinal cord tumors; only T2-weighted sequence used; no public release link for the pediatric MRI in this study.

Noise

Scans with substantial artifact/poor image quality were excluded during QC; failures included cases with large cystic areas, ventricular location, large section thickness (respacing), or heterogeneous lesions.

External data

Adult glioma MRI from BraTS 2021 (n=1251) used for transfer learning.

Confidentiality

Raw MRI data cannot be shared; data include pediatric patients with brain tumors.

Sensitive data

Pediatric oncologic imaging data.