Adult Glioma Segmentation in Sub-Saharan Africa using Transfer Learning on Stratified Finetuning Data

TL;DR

This study addresses the challenge of glioma segmentation in Sub-Saharan Africa where MRI data are scarce and of limited quality. It leverages transfer learning with two pretrained models, nnU-Net and MedNeXt, applying a radiomics-driven stratified fine-tuning strategy on BraTS2023-Adult-Glioma and BraTS-Africa to produce a robust SSA segmentation pipeline. A weighted ensemble and adaptive post-processing further enhance lesion-wise segmentation accuracy, achieving a top performance on the BraTS-Africa 2024 SSA task with lesion-wise Dice scores of $0.870$, $0.865$, and $0.927$ for ET, TC, and WT respectively. The approach demonstrates how local data integration and stratified refinement can bridge imaging capabilities between resource-limited regions and developed healthcare systems, with a dockerized implementation available for practical deployment.

Abstract

Gliomas, a kind of brain tumor characterized by high mortality, present substantial diagnostic challenges in low- and middle-income countries, particularly in Sub-Saharan Africa. This paper introduces a novel approach to glioma segmentation using transfer learning to address challenges in resource-limited regions with minimal and low-quality MRI data. We leverage pre-trained deep learning models, nnU-Net and MedNeXt, and apply a stratified fine-tuning strategy using the BraTS2023-Adult-Glioma and BraTS-Africa datasets. Our method exploits radiomic analysis to create stratified training folds, model training on a large brain tumor dataset, and transfer learning to the Sub-Saharan context. A weighted model ensembling strategy and adaptive post-processing are employed to enhance segmentation accuracy. The evaluation of our proposed method on unseen validation cases on the BraTS-Africa 2024 task resulted in lesion-wise mean Dice scores of 0.870, 0.865, and 0.926, for enhancing tumor, tumor core, and whole tumor regions and was ranked first for the challenge. Our approach highlights the ability of integrated machine-learning techniques to bridge the gap between the medical imaging capabilities of resource-limited countries and established developed regions. By tailoring our methods to a target population's specific needs and constraints, we aim to enhance diagnostic capabilities in isolated environments. Our findings underscore the importance of approaches like local data integration and stratification refinement to address healthcare disparities, ensure practical applicability, and enhance impact. A dockerized version of the BraTS-Africa 2024 winning algorithm is available at https://hub.docker.com/r/aparida12/brats-ssa-2024 .

Figures (3)

• Figure 1: Training examples showing the differences between the BraTS-Adult-Glioma and the BraTS2023-Africa datasets.
• Figure 2: Proposed method: Unsupervised stratified fold split, fine-tuning of pre-trained models on BraTS2023-GLI, model ensembling and adaptive post-processing. Outputs are obtained from two state-of-the-art deep learning models. These outputs are subjected to nonlinear activation functions and ensembling strategies. Finally, the ensembled predictions are subjected to a specifically tailored adaptive post-processing step.
• Figure 3: Qualitative results of models after post processing on the validation sample of BraTS-SSA-00227-000. The top row of the figure shows T1CE and the performance of model in segmenting different regions(NCR-red, ED-green, ET-blue).