2026

LXNet: A lightweight CNN for lung disease classification from Chest X-ray with XAI-based interpretability

Juiria Humayan, Md Najmus Sakib Nahid, Amir Sohel, Md Alamgir Kabir, Md Shakhawat Hossain, Zahid Ullah, Mona Jamjoom

PloS one

View Publication DOI: https://doi.org/10.1371/journal.pone.0351762

LXNet: A lightweight CNN for lung disease classification from Chest X-ray with XAI-based interpretability

Abstract

Lung diseases such as pneumonia and tuberculosis remain major global health challenges, particularly in resource-limited settings. This study presents LXNet, a lightweight and explainable convolutional neural network (CNN) for nine-class chest X-ray (CXR) classification. Evaluated on 6,743 CXR images from a private imaging center in Brazil, LXNet contains only 0.35 million parameters and preserves subtle diagnostic features through a no-pooling final block. Image enhancement using adaptive CLAHE, grayscale normalization, and stratified class balancing further improved robustness. Compared with DenseNet201, ResNet50V2, and InceptionV3 under identical training settings, LXNet achieved 96.1%five-fold cross-validation accuracy, outperforming the baselines by 1–8%, while requiring only 308 s for training. Explainability was provided through Grad-CAM, Score-CAM, and LIME, highlighting clinically relevant regions, and statistical significance was confirmed using the Wilcoxon signed-rank test (p = 0.03125). Although external validation showed reduced performance, LXNet offers an efficient and interpretable framework for multiclass lung disease classification.

Citation

Juiria Humayan, Md Najmus Sakib Nahid, Amir Sohel, Md Alamgir Kabir, Md Shakhawat Hossain, Zahid Ullah, Mona Jamjoom. "LXNet: A lightweight CNN for lung disease classification from Chest X-ray with XAI-based interpretability." PloS one (2026).

BibTeX

@article{pub53_2026,
  title={LXNet: A lightweight CNN for lung disease classification from Chest X-ray with XAI-based interpretability},
  author={Juiria Humayan, Md Najmus Sakib Nahid, Amir Sohel, Md Alamgir Kabir, Md Shakhawat Hossain, Zahid Ullah, Mona Jamjoom},
  journal={PloS one},
  year={2026},
  doi={https://doi.org/10.1371/journal.pone.0351762}
}