Enhancing representation learning with frequency-augmented feature mixture for robust tuberculosis screening

Abudouresuli Tuersun

doi:10.1007/s00371-026-04405-1

Enhancing representation learning with frequency-augmented feature mixture for robust tuberculosis screening

2026-04-22

0

PRE

AI

Saimaitikari Abudoubari

DOI：10.1007/s00371-026-04405-1

Original

Original request for help

Save

Abstract

En 中文

Tuberculosis (TB) remains a global health challenge, where chest X-rays (CXRs) are a primary tool for screening. Automated TB screening is complicated by the disease’s diverse and subtle visual manifestations. To address this, we propose FMoF, a novel deep learning framework that integrates frequency-domain information with spatial feature learning for robust CXR classification. Our core contributions include a frequency-enhanced attention module, which selectively enhances discriminative spectral components to enrich spatial feature maps, and a mixture-of-features module that dynamically aggregates multi-scale contextual representations. Through extensive benchmark evaluations on public datasets including TCX and TBX11K, our method achieves state-of-the-art performance, demonstrating superior accuracy and generalization. The proposed approach demonstrates the significant potential of hybrid domain reasoning for improving medical image analysis.

Keywords:

Image classification

Fourier transform

Multi-scale fusion

Tuberculosis diagnosis

Enhancing representation learning with frequency-augmented feature mixture for robust tuberculosis screening

Abstract

Journal

Organization