Source-linked AI summary
The Brain Tumor Segmentation in Pediatrics (BraTS-PEDs) Challenge: Focus on Pediatrics (CBTN-CONNECT-DIPGR-ASNR-MICCAI BraTS-PEDs)
Anahita Fathi Kazerooni, Nastaran Khalili, Xinyang Liu, Deep Gandhi, Zhifan Jiang, Syed Muhammed Anwar, Jake Albrecht, Maruf Adewole, Udunna Anazodo, Hannah Anderson, Ujjwal Baid, Timothy Bergquist, Austin J. Borja, Evan Calabrese, Verena Chung, Gian-Marco Conte, Farouk Dako, James Eddy, Ivan Ezhov, Ariana Familiar, Keyvan Farahani, Andrea Franson, Anurag Gottipati, Shuvanjan Haldar, Juan Eugenio Iglesias, Anastasia Janas, Elaine Johansen, Blaise V Jones, Neda Khalili, 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, Jeffrey D Rudie, Mariana Sanchez-Montano, Ibraheem Salman Shaikh, Nakul Sheth, Wenxin Tu, Chunhao Wang, Jeffrey B Ware, Benedikt Wiestler, Anna Zapaishchykova, Miriam Bornhorst, Michelle Deutsch, Maryam Fouladi, Margot Lazow, Leonie Mikael, Trent Hummel, Benjamin Kann, Peter de Blank, Lindsey Hoffman, Mariam Aboian, Ali Nabavizadeh, Roger Packer, Spyridon Bakas, Adam Resnick, Brian Rood, Arastoo Vossough, Marius George Linguraru
TL;DR
Pediatric brain tumors need dedicated segmentation methods, while existing automated methods are few and often target only whole tumor. BraTS-PEDs presents a multi-institutional challenge and cohort for benchmarking pediatric tumor segmentation, with expert-reviewed labels and defined evaluation procedures.
Problem
Automated pediatric tumor segmentation methods are scarce, and most existing methods target only the T2-FLAIR whole-tumor signal despite the need for dedicated tools for tumor characterization and assessment.
Source: §1 · p. 2Method
BraTS-PEDs organizes a multi-institutional challenge using pediatric high-grade glioma MRI and expert-reviewed labels for four tumor subregions.
Source: §1 · p. 2, §2.1 · p. 2, §3 · p. 3Results
The paper outlines a challenge benchmark with defined subregion labels, withheld testing, and containerized evaluation for pediatric brain tumor segmentation.
Source: §1 · p. 2, §3 · p. 3, §2.2 · p. 5Takeaways & Limitations
The authors plan to expand the cohort and future challenge data to additional institutions, tumor histologies, and postoperative or post-treatment scans.
Source: §3 · p. 5Automated segmentation may label periventricular edema or remote areas as tumor, under-segment cysts, or confuse cystic components with non-enhancing tumor.
Source: §2.1.2 · 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. Here we present the CBTN-CONNECT-DIPGR-ASNR-MICCAI BraTS-PEDs challenge, focused on pediatric brain tumors with data acquired across multiple international consortia dedicated to pediatric neuro-oncology and clinical trials. The CBTN-CONNECT-DIPGR-ASNR-MICCAI BraTS-PEDs 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 brain tumor segmentation tools are needed for tumor characterization and assessment, but automated methods explicitly developed for children are scarce and often target only the T2-FLAIR whole-tumor signal. The 2024 BraTS-PEDs challenge extends a pediatric benchmark effort using data from multiple consortia and institutions.
1 Introduction
- Pediatric brain tumors need dedicated segmentation tools for characterization and assessment of diagnosis, prognosis, and treatment response.Only a handful of automated methods have been explicitly proposed for pediatric brain tumors.Source: §1 · p. 2
- Most existing pediatric methods segment only the abnormal T2-FLAIR signal, also called whole tumor.Source: §1 · p. 2
- The 2024 BraTS-PEDs challenge continues the 2023 launch with changes to processing and tumor subregions, drawing on several pediatric consortia and institutions.ASNR collaborated on ground-truth annotations for most challenge data.Source: §1 · p. 2
2 Challenge Design
BraTS-PEDs provides a retrospective, multi-institutional cohort of pediatric high-grade glioma with four structural MRI sequences and ground-truth training labels. The challenge restricts training data to the provided cohort to support fair comparisons.
2 Challenge Design
- The cohort contains four structural MRI sequences from 464 pediatric high-grade glioma cases, with acquisition heterogeneity across institutions.Sequences are T1, T1CE, T2, and T2-FLAIR; the scans are treatment-naive.Source: §2.1 · p. 2
- Training data include scans and ground-truth labels, while validation scans have no shared labels and test data remain withheld for containerized evaluation.The testing submissions are evaluated on Synapse, powered by MedPerf.Source: §2.1 · p. 3
- Participants may train only on BraTS-PEDs data or models not pretrained on other datasets, preserving the challenge’s stated basis for fair comparison.Additional data may be used for publication if BraTS-PEDs-only results are also reported and differences discussed.Source: §2.1 · p. 3
3. Tumor subregion segmentation using a pediatric autosegmentation method [13] 4
The challenge defines four pediatric tumor subregions, uses automated labels as a starting point for expert-reviewed annotations, and specifies evaluation and participation procedures. It also identifies common automated segmentation errors that annotators should distinguish.
3. Tumor subregion segmentation using a pediatric autosegmentation method [13] 4
- Tumors are labeled as enhancing tumor, non-enhancing tumor, cystic component, and peritumoral edema using the stated MRI signal and location criteria.Cystic component is within the tumor, whereas edema surrounds it while preserving underlying brain structure.Source: §3 · p. 3
- Figure 1 depicts the processing pipeline and the four tumor labels over T1, T1CE, T2, and T2-FLAIR scans.Source: Figure 1 · p. 4
- Automated four-label segmentations provide preliminary annotations, which volunteers refine using four MRI sequences before attending neuroradiologists review them iteratively for release.The review process returns annotations for further refinement when needed.Source: §3 · p. 4
- Common automated errors include labeling periventricular edema or remote areas as tumor, under-segmenting cysts, and confusing cystic components with non-enhancing tumor.An enhancing rim around a cyst-like portion favors NET; very bright T2 and dark T1 signal favors cyst.Source: §2.1.2 · p. 4
2.2 Performance Evaluation
- Evaluation covers ET, NET, CC, edema, tumor core (ET, NET, and CC), and whole tumor.Source: §2.2 · p. 4
- Containerized submissions are evaluated on withheld testing data to support reproducibility and comparisons with prior BraTS instances.Source: §2.2 · p. 5
2.3 Participation Timeline
- The challenge proceeds through training with labels, validation without shared labels, and final ranking on unseen testing data.Validation allows multiple submissions; final testing submissions are containerized and evaluated on data not made available to participants.Source: §2.3 · p. 5
3 Discussion
The paper outlines BraTS-PEDs as a benchmark for pediatric brain tumor segmentation. The authors plan to expand the cohort and broaden future challenge data to additional institutions, histologies, and scan settings.
3 Discussion
- BraTS-PEDs is designed to benchmark methods for pediatric brain tumor segmentation.Source: §3 · p. 5
- The authors are working to increase cohort size to provide a larger dataset of these rare tumors and facilitate future development of computer-aided treatment-planning tools.Source: §3 · p. 5
- Future challenges are planned to add institutions and tumor histologies and include postoperative or post-treatment scans.Source: §3 · p. 5
Funding
The publication was partly supported by NIH awards and grant funding from two pediatric brain tumor organizations; its content is the authors’ responsibility, not the NIH’s official view.
Funding
- The research was partly supported by NIH awards NCI/ITCR:U01CA242871 and NCI:UH3CA236536.Source: Funding · p. 5
- Grant funding also came from the Pediatric Brain Tumor Foundation and DIPG/DMG Research Funding Alliance (DDRFA).Source: Funding · p. 5
- The publication states that its content is solely the authors’ responsibility and does not represent official NIH views.Source: Funding · p. 5