Adaptable Segmentation Pipeline for Diverse Brain Tumors with Radiomic-guided Subtyping and Lesion-Wise Model
Daniel Capellán-Martín1,2, Abhijeet Parida1,2, Zhifan Jiang1
1Sheikh Zayed Institute for Pediatric Surgical Innovation, Children's National Hospital, Washington, DC, USA.
Segmentation, Classification, and Synthesis for Brain Tumors and Traumatic Brain Injuries : MICCAI 2025 Challenges: Brats-Lighthouse 2025 and AIMS-TBI 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23, 2025
|May 8, 2026
Summary
This study introduces a flexible pipeline for robust brain tumor segmentation using multi-parametric MRI. The adaptable method achieves high performance across diverse tumor types, aiding clinical diagnosis and prognosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Brain tumor segmentation on multi-parametric MRI is challenging due to diverse tumor types.
- Existing methods struggle with generalizability across various adult and pediatric brain tumors.
Purpose of the Study:
- To develop and evaluate a flexible, modular pipeline for robust and generalizable brain tumor segmentation.
- To benchmark segmentation performance on diverse datasets including pediatric tumors, meningiomas, and brain metastases.
- To improve quantitative tumor measurement for clinical diagnosis and prognosis.
Main Methods:
- A flexible pipeline combining state-of-the-art models with tumor- and lesion-specific processing.
- Utilized radiomic features from MRI for tumor subtype detection and balanced training.
- Employed custom lesion-level performance metrics for model selection, ensemble optimization, and post-processing refinement.
Main Results:
- The proposed pipeline achieved performance comparable to top-ranked algorithms on BraTS 2025 challenge datasets.
- Demonstrated robust segmentation across diverse tumor types (PED, MEN, MEN-RT, MET).
- Showcased adaptability without reliance on specific network architectures.
Conclusions:
- Custom lesion-aware processing and model selection yield robust brain tumor segmentations.
- The adaptable pipeline shows potential for quantitative tumor measurement in clinical practice.
- Supports improved diagnosis and prognosis through accurate brain tumor segmentation.