Skip to main navigation Skip to search Skip to main content

Parallel dual-path multiscale feature fusion network for Alzheimer's disease classification

  • Jikai Wang
  • , Mingfeng Jiang
  • , Wei Zhang
  • , Yang Li
  • , Tao Tan
  • , Yaming Wang
  • , Tie qiang Li
  • Zhejiang Sci-Tech University
  • Lishui University
  • Karolinska Institutet

Research output: Contribution to journal › Article › peer-review

Abstract

Multiscale feature fusion is an important method for the accurate early classification of Alzheimer's disease (AD). However, conventional multiscale feature fusion methods often face the problems of scale information loss and misalignment between different scales, which degrade the effective fusion of multiscale features. To address these challenges, a novel classification network based on parallel dual-path multiscale learning and hybrid attention mechanism is proposed to improve AD classification performance by effectively blending multiscale information. Specifically, a parallel dual-path multiscale feature fusion deep learning network is designed to implement multiscale information fusion through global, local, and hybrid scale feature extraction and enhancement. The network has three core components: a global feature extraction module(PST), a local feature extraction module(RLL), and a parallel dual-path scale fusion mechanism module(PDP). The global feature extraction module uses a pooling sparse transformer to learn global relationships between multiscale information, and the local feature extraction module can capture local features through convolution, residual joining, and nonlinear mapping. Moreover, the parallel dual-path scale fusion mechanism is implemented by using spatial attention, channel attention, and nonlinear mapping relationships to enhance the interaction and fusion between multiscale features effectively. The ADNI datasets are used to validate the proposed method, and the experimental results show that the proposed method outperforms other SOTA classification methods with the best AD classification accuracy.

Original languageEnglish
Article number111095
JournalBiomedical Signal Processing and Control
Volume127
DOIs
Publication statusPublished - 1 Nov 2026

Keywords

  • 3D multiscale fusion network
  • Alzheimer's disease
  • Attention mechanism
  • Fine-grained features
  • Parallel dual-path

Fingerprint

Dive into the research topics of 'Parallel dual-path multiscale feature fusion network for Alzheimer's disease classification'. Together they form a unique fingerprint.

Cite this