Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reliable Brain Tumor Segmentation Based on Spiking Neural Networks with Efficient Training

Aurora Pia Ghiardelli, Guangzhi Tang, Tao Sun

TL;DR

This work tackles the challenge of reliable 3D brain tumor segmentation under power constraints by introducing Spiking U-Seg-Net, an energy-efficient SNN-based framework trained with Forward Propagation Through Time. A multi-view ensemble across sagittal, coronal, and axial planes provides voxel-level uncertainty estimates and robustness, achieving competitive Dice scores while markedly reducing computational cost. Key contributions include the lightweight Spiking U-Seg-Net architecture, FPTT-based training for scalable temporal learning, and a multi-view ensemble that improves calibration (lower NLL) and segmentation reliability on BraTS17 and BraTS23. The results demonstrate strong performance with an 87% FLOPs reduction, highlighting the potential for edge-enabled, uncertainty-aware medical imaging on IoT and PoC platforms.

Abstract

We propose a reliable and energy-efficient framework for 3D brain tumor segmentation using spiking neural networks (SNNs). A multi-view ensemble of sagittal, coronal, and axial SNN models provides voxel-wise uncertainty estimation and enhances segmentation robustness. To address the high computational cost in training SNN models for semantic image segmentation, we employ Forward Propagation Through Time (FPTT), which maintains temporal learning efficiency with significantly reduced computational cost. Experiments on the Multimodal Brain Tumor Segmentation Challenges (BraTS 2017 and BraTS 2023) demonstrate competitive accuracy, well-calibrated uncertainty, and an 87% reduction in FLOPs, underscoring the potential of SNNs for reliable, low-power medical IoT and Point-of-Care systems.

Reliable Brain Tumor Segmentation Based on Spiking Neural Networks with Efficient Training

TL;DR

This work tackles the challenge of reliable 3D brain tumor segmentation under power constraints by introducing Spiking U-Seg-Net, an energy-efficient SNN-based framework trained with Forward Propagation Through Time. A multi-view ensemble across sagittal, coronal, and axial planes provides voxel-level uncertainty estimates and robustness, achieving competitive Dice scores while markedly reducing computational cost. Key contributions include the lightweight Spiking U-Seg-Net architecture, FPTT-based training for scalable temporal learning, and a multi-view ensemble that improves calibration (lower NLL) and segmentation reliability on BraTS17 and BraTS23. The results demonstrate strong performance with an 87% FLOPs reduction, highlighting the potential for edge-enabled, uncertainty-aware medical imaging on IoT and PoC platforms.

Abstract

We propose a reliable and energy-efficient framework for 3D brain tumor segmentation using spiking neural networks (SNNs). A multi-view ensemble of sagittal, coronal, and axial SNN models provides voxel-wise uncertainty estimation and enhances segmentation robustness. To address the high computational cost in training SNN models for semantic image segmentation, we employ Forward Propagation Through Time (FPTT), which maintains temporal learning efficiency with significantly reduced computational cost. Experiments on the Multimodal Brain Tumor Segmentation Challenges (BraTS 2017 and BraTS 2023) demonstrate competitive accuracy, well-calibrated uncertainty, and an 87% reduction in FLOPs, underscoring the potential of SNNs for reliable, low-power medical IoT and Point-of-Care systems.
Paper Structure (18 sections, 2 equations, 1 figure, 3 tables)

This paper contains 18 sections, 2 equations, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Uncertainty Estimation and Deep Ensemble
  4. Spiking Neural Networks (SNNs)
  5. Leaky Integrate-and-Fire (LIF) Neurons
  6. Training of SNNs
  7. Method
  8. Network Architecture
  9. SNN Inference and Memory-efficient FPTT Training
  10. Multi-View Ensemble
  11. Loss Function
  12. Experiments and Results
  13. BraTS Dataset
  14. Evaluation
  15. Training and Testing
  16. ...and 3 more sections

Figures (1)

  • Figure 1: (Top) Architecture of proposed Spiking U-Seg-Net; (Bottom) Tumor segmentation results across three views—sagittal, coronal, and axial—with enhancing tumor in red, necrotic/non-enhancing tumor in blue, and peritumoral edema in green.