SpikeDS: Dual Sparsity Spikformer for Perineural Invasion Prediction in 3D MRI

2026-07-13

AI Summary

This study addresses the challenge of efficiently and accurately modeling perineural invasion (PNI) prediction in 3D MRI for cholangiocarcinoma, where weak lesion boundary features and strong spatial heterogeneity hinder performance. To this end, the authors propose SpikeDS, a dual-sparse spiking Transformer architecture featuring a novel Dual-Sparse Spiking Attention (DSSA) mechanism. DSSA integrates firing-rate-based window pruning, Window-based Expert Mixture Spiking Attention (W-EMSA), and Cross-Window Asymmetric Self-Attention (CW-SSA) to jointly induce activation and spatial sparsity. This design significantly reduces computational energy consumption while preserving high diagnostic accuracy. Evaluated via five-fold cross-validation on 139 cholangiocarcinoma patients, the model achieves an AUC of 0.753 with only 14.4 mJ of energy, outperforming existing baselines in both accuracy and energy efficiency.

Abstract

Perineural invasion (PNI) is associated with poor prognosis in cholangiocarcinoma (CCA). However, its detection from 3D MRI remains challenging due to the subtle and spatially heterogeneous imaging signatures at the tumor periphery. Capturing such spatially sparse cues necessitates volumetric analysis of 3D MRI, but existing deep learning approaches incur prohibitive computational costs on volumetric medical images, limiting their clinical deployment. We propose Dual Sparsity Spikformer (SpikeDS), a spiking neural network architecture that jointly exploits activation sparsity from binary spike communication and spatial sparsity from window pruning based on firing rates. SpikeDS introduces Dual Sparsity Spiking Attention (DSSA), which combines two complementary mechanisms. The first is Window-based Expert Mixture Spiking Attention (W-EMSA), which selectively applies attention only to salient windows identified by their firing rates. The second is Cross-Window Spiking Self-Attention (CW-SSA), which enables global context exchange through an asymmetric scheme in which pruned windows still contribute as key-value sources. Evaluated on a clinical cohort of 139 CCA patients via 5-fold cross-validation, SpikeDS achieves an AUC of 0.753 while consuming only 14.4 mJ, surpassing the best baseline in both AUC and energy efficiency. These results suggest that dual sparsity provides an effective hardware-aware strategy for improving the efficiency of 3D spiking transformers without compromising diagnostic performance.

Problem

• Perineural Invasion
• 3D MRI
• Computational Efficiency
• Cholangiocarcinoma
• Deep Learning

Innovation

• spiking neural network
• dual sparsity
• 3D medical image analysis
• energy-efficient AI
• perineural invasion prediction