Source-linked AI summary
The Brain Tumor Segmentation (BraTS) Challenge 2023: Focus on Pediatrics (CBTN-CONNECT-DIPGR-ASNR-MICCAI BraTS-PEDs)
Anahita Fathi Kazerooni, Nastaran Khalili, Xinyang Liu, Debanjan Haldar, Zhifan Jiang, Syed Muhammed Anwar, Jake Albrecht, Maruf Adewole, Udunna Anazodo, Hannah Anderson, Sina Bagheri, Ujjwal Baid, Timothy Bergquist, Austin J. Borja, Evan Calabrese, Verena Chung, Gian-Marco Conte, Farouk Dako, James Eddy, Ivan Ezhov, Ariana Familiar, Keyvan Farahani, Shuvanjan Haldar, Juan Eugenio Iglesias, Anastasia Janas, Elaine Johansen, Blaise V Jones, Florian Kofler, Dominic LaBella, Hollie Anne Lai, Koen Van Leemput, Hongwei Bran Li, Nazanin Maleki, Aaron S McAllister, Zeke Meier, Bjoern Menze, Ahmed W Moawad, Khanak K Nandolia, Julija Pavaine, Marie Piraud, Tina Poussaint, Sanjay P Prabhu, Zachary Reitman, Andres Rodriguez, Jeffrey D Rudie, Mariana Sanchez-Montano, Ibraheem Salman Shaikh, Lubdha M. Shah, Nakul Sheth, Russel Taki Shinohara, Wenxin Tu, Karthik Viswanathan, Chunhao Wang, Jeffrey B Ware, Benedikt Wiestler, Walter Wiggins, Anna Zapaishchykova, Mariam Aboian, Miriam Bornhorst, Peter de Blank, Michelle Deutsch, Maryam Fouladi, Lindsey Hoffman, Benjamin Kann, Margot Lazow, Leonie Mikael, Ali Nabavizadeh, Roger Packer, Adam Resnick, Brian Rood, Arastoo Vossough, Spyridon Bakas, Marius George Linguraru
TL;DR
Pediatric brain tumors differ from adult tumors, while automated pediatric subregion segmentation lacks common-data benchmarking. BraTS-PEDs 2023 establishes a multi-institutional challenge and annotated mpMRI cohort; prior testing found adult-trained methods performed poorly on pediatric tumor subregions despite relatively high whole-tumor Dice.
Problem
Automated pediatric tumor-segmentation methods are few, often target only whole tumor, and have not been compared in a controlled evaluation on common validation data.
Source: §1 · p. 3Method
BraTS-PEDs 2023 builds a multi-institutional pediatric high-grade glioma mpMRI cohort with reviewed annotations and evaluates participant algorithms on separate validation and withheld test data.
Source: §1 · p. 3, §2.1 · p. 3, §2.1.1 · p. 4Results
Prior BraTS 2022 testing found adult-trained models relatively strong on pediatric whole-tumor Dice but poor on tumor-subregion segmentation.
Source: §4 · p. 6Takeaways & Limitations
The growing annotated pediatric high-grade glioma cohort is intended to support future development of computer-aided treatment-planning tools.
Source: §4 · p. 6Challenge participants may train only on provided BraTS-PEDs data; additional-data results require separate reporting and discussion of potential differences.
Source: §2.1 · p. 4
Abstract
from arXiv · showhide
Pediatric tumors of the central nervous system are the most common cause of cancer-related death in children. The five-year survival rate for high-grade gliomas in children is less than 20\%. Due to their rarity, the diagnosis of these entities is often delayed, their treatment is mainly based on historic treatment concepts, and clinical trials require multi-institutional collaborations. The MICCAI Brain Tumor Segmentation (BraTS) Challenge is a landmark community benchmark event with a successful history of 12 years of resource creation for the segmentation and analysis of adult glioma. Here we present the CBTN-CONNECT-DIPGR-ASNR-MICCAI BraTS-PEDs 2023 challenge, which represents the first BraTS challenge focused on pediatric brain tumors with data acquired across multiple international consortia dedicated to pediatric neuro-oncology and clinical trials. The BraTS-PEDs 2023 challenge focuses on benchmarking the development of volumentric segmentation algorithms for pediatric brain glioma through standardized quantitative performance evaluation metrics utilized across the BraTS 2023 cluster of challenges. Models gaining knowledge from the BraTS-PEDs multi-parametric structural MRI (mpMRI) training data will be evaluated on separate validation and unseen test mpMRI dataof high-grade pediatric glioma. The CBTN-CONNECT-DIPGR-ASNR-MICCAI BraTS-PEDs 2023 challenge brings together clinicians and AI/imaging scientists to lead to faster development of automated segmentation techniques that could benefit clinical trials, and ultimately the care of children with brain tumors.
1 Introduction
Pediatric tumors differ from adult tumors in their imaging and clinical presentation, while existing pediatric segmentation tools and comparisons are limited. BraTS-PEDs 2023 establishes a multi-institutional benchmark to evaluate automated pediatric brain-tumor segmentation methods.
1 Introduction
- Pediatric tumors differ from adult tumors in location, age at diagnosis, and imaging findings, supporting the need for dedicated characterization tools.For example, DMGs often occur in the pons in children aged 5–10, and enhancing or necrotic regions are less common or clear at initial diagnosis than in GBM.Source: §1 · p. 2
- Manual segmentation is time-consuming and variable, while 2D measurements inadequately capture irregular tumor shape and volume.Volumetric measurements have been reported to predict clinical endpoints better than 2D methods in adult tumors.Source: §1 · pp. 2–3
- Automated pediatric methods are few, often target only T2-FLAIR whole tumor, and lack controlled evaluation on common validation data.The section identifies a benchmarking gap for pediatric tumor-segmentation tools.Source: §1 · p. 3
- BraTS-PEDs 2023 is a multi-institutional, multi-consortium initiative to benchmark pediatric brain-tumor segmentation using unseen validation and test data.The cohort draws on CBTN, the DIPG Registry, CONNECT, and participating pediatric institutions, with ASNR contributing ground-truth annotation.Source: §1 · p. 3
2 Materials & Methods
BraTS-PEDs 2023 assembles and annotates pediatric high-grade glioma mpMRI data, then evaluates participant algorithms through staged validation and testing. Participants receive three challenge labels derived from four manually reviewed tumor-subregion annotations.
2.1 Data
- The cohort comprises 228 pediatric high-grade gliomas imaged with four structural mpMRI sequences, split into training (99), validation (45), and testing datasets.Training labels are shared; validation ground truth and testing data are withheld from participants.Source: §2.1 · p. 3
- Scans are standardized through DICOM-to-NIfTI conversion, SRI24 co-registration, and resampling to 1 mm^3 isotropic resolution.The pipeline also removes header PHI and skull-strips images to prevent potential facial reconstruction or recognition.Source: §2.1.1 · p. 4
- Four tumor subregions are initialized by two pediatric deep-learning models, manually refined by volunteer neuroradiologists, and reviewed iteratively by three attending neuroradiologists.The original annotations distinguish enhancing tumor (ET), nonenhancing tumor (NET), cystic component (CC), and peritumoral edema (ED).Source: §2.1.1 · p. 4
- Participants receive ET, nonenhancing component (NC), and ED labels, with NC combining nonenhancing tumor, cystic component, and necrosis.The four underlying regions are consolidated into three labels to align with other BraTS 2023 challenges and allow use of adult-glioma challenge data for training.Source: §2.1.1 · p. 4
- Automated segmentation errors include peri-ventricular edema or remote areas labeled as tumor, cyst under-segmentation, and confusion between cyst and nonenhancing tumor.The automated masks are preliminary and undergo manual revision before release.Source: §2.1.2 · p. 4, §2.1.1 · p. 4
2.2 Leaderboard Evaluation
- Evaluation regions are ET, tumor core (ET, cystic component, and necrosis), and whole tumor (WT), with submissions scored through the evaluation platform on withheld testing data.Participants select a method after validation; organizers evaluate its containerized submission on the hidden test set.Source: §2.2 · pp. 4–5
- The figure contrasts four original annotated subregions with the three labels supplied to participants.The participant labels are ET, NC, and ED; NC combines nonenhancing tumor, cystic component, and necrosis.Source: §2.1.1 · p. 4, Figure 1 · p. 5
2.3 Participation Timeline
- The challenge proceeds through training with labels, validation without labels, and final containerized testing and ranking on unseen data.Validation permits repeated submissions to online platforms, while the testing set is not made available to participants.Source: §2.3 · p. 5
3 Results
Nine short-paper submissions met the criteria for inclusion in the testing-phase ranking. The results section describes a methods-and-performance table, but its team names and reported ranking and Dice values are represented as placeholders.
3 Results
- Nine short-paper submissions met the criteria for inclusion in the testing-phase ranking.The total number of teams requesting dataset access and participating in validation is represented as XXX.Source: §3 · p. 6
3.1 Performance Evaluation
- The results table is described as summarizing participant methods, rankings, and performance metrics, but the reported details are represented as XXX.The passage identifies a best-ranked team and top-three Dice scores without providing their names or values.Source: §3.1 · p. 6
4 Discussion
The discussion presents BraTS-PEDs 2023 as the first pediatric brain-tumor segmentation challenge and argues that pediatric-specific training is needed for subregion segmentation. Its annotated cohort is intended to grow and support future tool development.
4 Discussion
- Adult-trained models achieved relatively high whole-tumor Dice on pediatric mpMRI but performed poorly on tumor subregions in BraTS 2022.The authors cite those findings as support for training pediatric methods at least partly on curated, annotated pediatric data.Source: §4 · p. 6
- BraTS-PEDs provides the largest annotated cohort of pediatric high-grade glioma, and the authors are working to increase its size.Future challenges are planned to include more institutions and post-operative or post-treatment scans.Source: §4 · p. 6
- The cohort is intended to facilitate development of tools for computer-aided treatment planning.The discussion also identifies future BraTS-PEDs challenges as an avenue for expanding the cohort and scan scope.Source: §4 · p. 6
Funding
The publication received partial support from NIH awards and grants from pediatric brain tumor research organizations.
Funding
- Partial support came from NIH awards and grants from the Pediatric Brain Tumor Foundation and DIPG/DMG Research Funding Alliance (DDRFA).NIH awards: NCI/ITCR:U01CA242871 and NCI:UH3CA236536.Source: Funding · p. 6