Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DRBD-Mamba for Robust and Efficient Brain Tumor Segmentation with Analytical Insights

Danish Ali, Ajmal Mian, Naveed Akhtar, Ghulam Mubashar Hassan

TL;DR

The paper introduces DRBD-Mamba, a dual-resolution, locality-preserving 3D brain tumor segmentation model that uses a Morton space-filling curve to map 3D features into 1D sequences. It combines bidirectional Mamba sequence modeling with per-channel gated fusion and vector quantization to capture multi-scale long-range dependencies while maintaining linear-time complexity. Extensive BraTS2023 experiments with both random and systematic five-fold folds show Dice improvements across tumor subregions and a substantial (~15x) reduction in computational cost compared with state-of-the-art Mamba-based methods, indicating robust generalization under heterogeneous data. This approach offers a practical, efficient, and robust solution for automated brain tumor segmentation with potential for clinical translation.

Abstract

Accurate brain tumor segmentation is significant for clinical diagnosis and treatment but remains challenging due to tumor heterogeneity. Mamba-based State Space Models have demonstrated promising performance. However, despite their computational efficiency over other neural architectures, they incur considerable overhead for this task due to their sequential feature computation across multiple spatial axes. Moreover, their robustness across diverse BraTS data partitions remains largely unexplored, leaving a critical gap in reliable evaluation. To address this, we first propose a dual-resolution bi-directional Mamba (DRBD-Mamba), an efficient 3D segmentation model that captures multi-scale long-range dependencies with minimal computational overhead. We leverage a space-filling curve to preserve spatial locality during 3D-to-1D feature mapping, thereby reducing reliance on computationally expensive multi-axial feature scans. To enrich feature representation, we propose a gated fusion module that adaptively integrates forward and reverse contexts, along with a quantization block that improves robustness. We further propose five systematic folds on BraTS2023 for rigorous evaluation of segmentation techniques under diverse conditions and present analysis of common failure scenarios. On the 20% test set used by recent methods, our model achieves Dice improvements of 0.10% for whole tumor, 1.75% for tumor core, and 0.93% for enhancing tumor. Evaluations on the proposed systematic folds demonstrate that our model maintains competitive whole tumor accuracy while achieving clear average Dice gains of 1.16% for tumor core and 1.68% for enhancing tumor over existing state-of-the-art. Furthermore, our model achieves a 15x efficiency improvement while maintaining high segmentation accuracy, highlighting its robustness and computational advantage over existing methods.

DRBD-Mamba for Robust and Efficient Brain Tumor Segmentation with Analytical Insights

TL;DR

The paper introduces DRBD-Mamba, a dual-resolution, locality-preserving 3D brain tumor segmentation model that uses a Morton space-filling curve to map 3D features into 1D sequences. It combines bidirectional Mamba sequence modeling with per-channel gated fusion and vector quantization to capture multi-scale long-range dependencies while maintaining linear-time complexity. Extensive BraTS2023 experiments with both random and systematic five-fold folds show Dice improvements across tumor subregions and a substantial (~15x) reduction in computational cost compared with state-of-the-art Mamba-based methods, indicating robust generalization under heterogeneous data. This approach offers a practical, efficient, and robust solution for automated brain tumor segmentation with potential for clinical translation.

Abstract

Accurate brain tumor segmentation is significant for clinical diagnosis and treatment but remains challenging due to tumor heterogeneity. Mamba-based State Space Models have demonstrated promising performance. However, despite their computational efficiency over other neural architectures, they incur considerable overhead for this task due to their sequential feature computation across multiple spatial axes. Moreover, their robustness across diverse BraTS data partitions remains largely unexplored, leaving a critical gap in reliable evaluation. To address this, we first propose a dual-resolution bi-directional Mamba (DRBD-Mamba), an efficient 3D segmentation model that captures multi-scale long-range dependencies with minimal computational overhead. We leverage a space-filling curve to preserve spatial locality during 3D-to-1D feature mapping, thereby reducing reliance on computationally expensive multi-axial feature scans. To enrich feature representation, we propose a gated fusion module that adaptively integrates forward and reverse contexts, along with a quantization block that improves robustness. We further propose five systematic folds on BraTS2023 for rigorous evaluation of segmentation techniques under diverse conditions and present analysis of common failure scenarios. On the 20% test set used by recent methods, our model achieves Dice improvements of 0.10% for whole tumor, 1.75% for tumor core, and 0.93% for enhancing tumor. Evaluations on the proposed systematic folds demonstrate that our model maintains competitive whole tumor accuracy while achieving clear average Dice gains of 1.16% for tumor core and 1.68% for enhancing tumor over existing state-of-the-art. Furthermore, our model achieves a 15x efficiency improvement while maintaining high segmentation accuracy, highlighting its robustness and computational advantage over existing methods.
Paper Structure (22 sections, 9 equations, 4 figures, 4 tables)

This paper contains 22 sections, 9 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. CNN-based Methods
  4. Transformer-based Hybrid Architectures
  5. Mamba-based Methods
  6. The Proposed Methodology
  7. Overview
  8. Mamba: Input-Dependent State Space Models
  9. Network Encoder
  10. Locality-Preserving Bi-Directional Quantized Mamba
  11. 3D Locality Preserving Sequencer
  12. Bi-directional Mamba
  13. Gated Fusion Module
  14. Quantizer
  15. Experiments
  16. ...and 7 more sections

Figures (4)

  • Figure 1: The overall architecture of the proposed dual-resolution bi-directional Mamba (DRBD-Mamba) with Mamba blocks placed in the bottleneck and skip connection to capture multi-scale context.
  • Figure 2: Comprehensive analysis of tumor heterogeneity and its impact on segmentation performance for the proposed systematic folds. (a) Distribution of average tumor subregion volumes (ED, NC, ET) across Dice score bins, (b) Dice score variation across ET volume quintiles (Bin 1 only).
  • Figure 3: Qualitative performance comparison of SegMamba, SegMambaV2, and the proposed method. Mid-slices of BraTS test cases are shown, with ground-truth masks and models' predictions. Tumor subregions are indicated as edema (green), necrotic core (red), and enhancing tumor (yellow).
  • Figure 4: Comparison of computational complexity of the proposed model and state-of-the-art models with respect to the increase in spatial resolution in BraTS2023.