Magnetic Resonance Imaging Feature-Based Subtyping and Model Ensemble for Enhanced Brain Tumor Segmentation

TL;DR

This work tackles automated brain tumor segmentation in multi-parametric MRI across pediatric and adult subtypes (PED, MEN-RT, MET) within the BraTS 2024 framework. It introduces a radiomics-guided, unsupervised subtype clustering approach coupled with a robust ensemble of three state-of-the-art segmentation models (nnU-Net, MedNeXt, SwinUNETR) and an adaptive post-processing pipeline. Radiomic subtypes inform both pre-processing and post-processing, while a weighted ensemble and threshold-based refinement improve robustness across heterogeneous tumors. The method achieves high lesion-wise Dice on whole-tumor segmentation (PED: 0.926, MEN-RT: 0.801, MET: 0.688) and is released with open-source code and a web application to facilitate adoption and further validation.

Abstract

Accurate and automatic segmentation of brain tumors in multi-parametric magnetic resonance imaging (mpMRI) is essential for quantitative measurements, which play an increasingly important role in clinical diagnosis and prognosis. The International Brain Tumor Segmentation (BraTS) Challenge 2024 offers a unique benchmarking opportunity, including various types of brain tumors in both adult and pediatric populations, such as pediatric brain tumors (PED), meningiomas (MEN-RT) and brain metastases (MET), among others. Compared to previous editions, BraTS 2024 has implemented changes to substantially increase clinical relevance, such as refined tumor regions for evaluation. We propose a deep learning-based ensemble approach that integrates state-of-the-art segmentation models. Additionally, we introduce innovative, adaptive pre- and post-processing techniques that employ MRI-based radiomic analyses to differentiate tumor subtypes. Given the heterogeneous nature of the tumors present in the BraTS datasets, this approach enhances the precision and generalizability of segmentation models. On the final testing sets, our method achieved mean lesion-wise Dice similarity coefficients of 0.926, 0.801, and 0.688 for the whole tumor in PED, MEN-RT, and MET, respectively. These results demonstrate the effectiveness of our approach in improving segmentation performance and generalizability for various brain tumor types. The source code of our implementation is available at https://github.com/Precision-Medical-Imaging-Group/HOPE-Segmenter-Kids. Additionally, an open-source web-application is accessible at https://segmenter.hope4kids.io/ which uses the docker container aparida12/brats-peds-2024:v20240913 .

Figures (4)

• Figure 1: Proposed Method: Unsupervised stratified fold splitting, model training, model ensemble, and adaptive post-processing. Outputs are generated using three state-of-the-art deep learning models, which are processed through ensemble strategies. The ensembled predictions are then refined using a specifically tailored adaptive post-processing step.
• Figure 2: Qualitative Results: The model ensemble, after post-processing, was evaluated on a PED validation case (BraTS-PED-00300-000). Numbers represent the segmentation performance in lesion-wise (LW) Dice for different tumor regions: enhancing tumor (ET) in red, non-enhancing tumor (NET) in green, cystic components (CC) in blue, edema (ED) in yellow, combined tumor core (TC), and whole tumor (WT).
• Figure 3: Qualitative Results: The model ensemble, after post-processing, was evaluated on an MEN-RT validation case (BraTS-MEN-RT-0698-1). The number represents the segmentation performance in lesion-wise (LW) Dice for the gross tumor volume (GTV) in red.
• Figure 4: Qualitative results: The model ensemble, after post-processing, was evaluated on an MET validation case (BraTS-MET-00908-000). Numbers represent the segmentation performance in lesion-wise (LW) Dice for different tumor regions: enhancing tumor (ET) in blue, combined tumor core (TC), and whole tumor (WT). Labels represent non-enhancing tumor (NET) in red and surrounding non-enhancing FLAIR hyperintensity (SNFH) in green.