Source-linked AI summary
The Brain Tumor Segmentation (BraTS) Challenge 2023: Glioma Segmentation in Sub-Saharan Africa Patient Population (BraTS-Africa)
Maruf Adewole, Jeffrey D. Rudie, Anu Gbadamosi, Oluyemisi Toyobo, Confidence Raymond, Dong Zhang, Olubukola Omidiji, Rachel Akinola, Mohammad Abba Suwaid, Adaobi Emegoakor, Nancy Ojo, Kenneth Aguh, Chinasa Kalaiwo, Gabriel Babatunde, Afolabi Ogunleye, Yewande Gbadamosi, Kator Iorpagher, Evan Calabrese, Mariam Aboian, Marius Linguraru, Jake Albrecht, Benedikt Wiestler, Florian Kofler, Anastasia Janas, Dominic LaBella, Anahita Fathi Kzerooni, Hongwei Bran Li, Juan Eugenio Iglesias, Keyvan Farahani, James Eddy, Timothy Bergquist, Verena Chung, Russell Takeshi Shinohara, Walter Wiggins, Zachary Reitman, Chunhao Wang, Xinyang Liu, Zhifan Jiang, Ariana Familiar, Koen Van Leemput, Christina Bukas, Maire Piraud, Gian-Marco Conte, Elaine Johansson, Zeke Meier, Bjoern H Menze, Ujjwal Baid, Spyridon Bakas, Farouk Dako, Abiodun Fatade, Udunna C Anazodo
TL;DR
Evidence is limited on whether state-of-the-art glioma machine-learning methods can be implemented in SSA, where MRI quality and disease presentation differ from settings represented in prior challenges. BraTS-Africa provides annotated African MRI data and a segmentation challenge; its dataset includes adult cases from African imaging centres, while the image-defined tumor regions are not strict biological entities.
Problem
It is unclear whether state-of-the-art glioma machine-learning methods can be widely implemented in SSA, where lower-quality MRI and disease presentation differ from settings represented in prior challenges.
Source: §1 · pp. 2–3Method
BraTS-Africa challenges participants to automatically segment intracranial gliomas into three classes using routine-care multiparametric MRI and expert-reviewed annotations.
Source: §2.1 · p. 4Results
The challenge provides annotated training, validation, and testing datasets from adult populations collected through African imaging centres.
Source: §2.1 · p. 3Takeaways & Limitations
The heterogeneous routine-care scans provide real-world cases for developing robust machine-learning methods.
Source: §2.1 · p. 4The BraTS tumor sub-regions are image-based visual features and do not represent strict biological entities.
Source: §2.1.2 · p. 6
Abstract
from arXiv · showhide
Gliomas are the most common type of primary brain tumors. Although gliomas are relatively rare, they are among the deadliest types of cancer, with a survival rate of less than 2 years after diagnosis. Gliomas are challenging to diagnose, hard to treat and inherently resistant to conventional therapy. Years of extensive research to improve diagnosis and treatment of gliomas have decreased mortality rates across the Global North, while chances of survival among individuals in low- and middle-income countries (LMICs) remain unchanged and are significantly worse in Sub-Saharan Africa (SSA) populations. Long-term survival with glioma is associated with the identification of appropriate pathological features on brain MRI and confirmation by histopathology. Since 2012, the Brain Tumor Segmentation (BraTS) Challenge have evaluated state-of-the-art machine learning methods to detect, characterize, and classify gliomas. However, it is unclear if the state-of-the-art methods can be widely implemented in SSA given the extensive use of lower-quality MRI technology, which produces poor image contrast and resolution and more importantly, the propensity for late presentation of disease at advanced stages as well as the unique characteristics of gliomas in SSA (i.e., suspected higher rates of gliomatosis cerebri). Thus, the BraTS-Africa Challenge provides a unique opportunity to include brain MRI glioma cases from SSA in global efforts through the BraTS Challenge to develop and evaluate computer-aided-diagnostic (CAD) methods for the detection and characterization of glioma in resource-limited settings, where the potential for CAD tools to transform healthcare are more likely.
1 Introduction
Glioma survival has improved in the Global North but not in LMICs, especially SSA, where death rates have risen. The paper motivates curated SSA MRI data to evaluate whether existing machine-learning methods can be applied in the region.
1 Introduction
- SSA glioma death rates rose by about 25% from 1990 to 2016, while rates in the Global North fell by 10–30%.The passage also notes that U.S. glioma mortality decreased by 7% over the past 30 years.Source: §1 · p. 2
- BraTS has benchmarked machine-learning methods for glioma diagnosis since 2012, but their clinical applicability in SSA is unclear given lower-quality MRI and regional disease presentation.SSA MRI may have poor contrast and resolution, and patients may present at advanced stages; the passage also mentions suspected higher rates of gliomatosis cerebri.Source: §1 · pp. 2–3
- The paper identifies a need for curated, annotated SSA MRI that represents routine-care imaging for developing and evaluating glioma machine-learning methods.Source: §1 · p. 3
2 Materials & Methods
BraTS-Africa establishes an annotated, multi-institutional MRI dataset and segmentation challenge using routine-care scans from African centres. Its procedures standardize imaging and expert review while defining visual tumor regions, known segmentation errors, evaluation, and participation.
2.1 Data
- The challenge provides adult training, validation, and testing data from African imaging centres, with routine-care scans from multiple institutions and scanners.The heterogeneous image quality is intended to provide real-world cases for robust machine-learning development.Source: §2.1 · pp. 3–4
2.1.1 Imaging Data Description
- Scans undergo BraTS-standard preprocessing, including NIfTI conversion, registration, 1 mm^3 resampling, and skull stripping; annotations are generated by STAPLE fusion of prior algorithms.The workflow also removes DICOM-header protected health information and mitigates facial reconstruction risk through skull stripping.Source: §2.1.1 · pp. 4–5
2.1.2 Tumor Annotation Protocol
- Expert-reviewed annotations use three visual regions—enhancing tumor, non-enhancing tumor core, and surrounding FLAIR hyperintensity—that are not strict biological entities.Volunteers refine automated segmentations, which senior board-certified radiologists review iteratively for approval.Source: §2.1.2 · pp. 5–6
2.1.3 Common errors of automated segmentations
- Common automated segmentation errors include ET extension into vessels or choroid plexus, SNFH under- or over-segmentation, and hemorrhage labeled as ET.Source: §2.1.3 · p. 6
2.1.4 Performance Evaluation
- Submissions are containerized for site-based evaluation and ranked using combined ranks across segmentation metrics, with permutation analyses assessing ranking uncertainty and significance.Metrics include Dice, 95% Hausdorff distance, sensitivity, specificity, and precision; missing results receive worst-case values.Source: §2.1.4 · pp. 6–7
Participation
- Participants progress from labeled training data to unlabeled validation and withheld testing, with African teams eligible for equity prizes subject to funding and workshop participation.Source: Participation · p. 7
Funding
The authors report partial support from the NIH and Lacuna Fund.
Funding
- The research was partly supported by NIH/NCI/ITCR award U01CA242871 and the Lacuna Fund for Machine Learning Datasets for Better Healthcare Outcomes.The authors state that the publication's content is solely their responsibility and does not represent official NIH views.Source: Funding · p. 8