PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
Synapse
⌘+K
Synapse
March 27, 2026Coronaviruses

Exploring Segmentation and Adaptive Attention in Deep Learning Models for COVID-19 CXR Diagnosis

View Full Paper
Ask AI
Bookmark
Share

Authors

FSFatin Nabilah ShaariANAimi Salihah Abdul NasirWMWan Azani Mustafa

Discussion

Loading...

Member takes

Overview

Demonstrates improvements in CXR classification using adaptive attention and hybrid CNN-LSTM models, implying better diagnostic strategies for COVID-19.

Key Points

  • The study investigates how the Self-Adaptive Convolutional Block Attention Module (SA-CBAM) enhances CXR classification performance.
  • Used a U-Net model for lung region segmentation from the COVID-QU-Ex dataset.
  • Compared various models, including CNN, attention mechanisms, and SA-CBAM.
  • Evaluated hybrid architectures combining CNN with LSTM networks.
  • Assessed models using metrics like accuracy, recall, specificity, F1-score, and MCC on segmented and unsegmented CXR images.
  • Unsegmented CXR models outperformed segmented ones consistently.
  • CNN-SA-CBAM improved accuracy from 88.42% to 90.08% and MCC from 82.82% to 85.36% on unsegmented data.
  • CNN-SA-CBAM-LSTM achieved 99.90% accuracy and 99.85% MCC on unsegmented CXR.
  • Segmentation produced clearer attention maps but resulted in lower classification performance.

Cite This Study

Shaari et al. (2026) studied this question.

synapsesocial.com/papers/69c61fd715a0a509bde18458https://doi.org/10.2174/0126667975436309260114074717
View Full Paper
Ask AI
Bookmark
Share