Raqamli Transformatsiya va Sun’iy Intellekt
LIGHTWEIGHT CNN ARCHITECTURES FOR THREE-CLASS TUBERCULOSIS X-RAY CLASSIFICATION
1 Tashkent University of Information Technologies named after Muhammad ibn Musa al-Khwarizmi
2 National Research University “Tashkent Institute of Irrigation and Agricultural Mechanization Engineers”
Zenodo DOI: 10.5281/zenodo.19872829
Keywords
Abstract
Tuberculosis (TB) remains one of the leading infectious diseases worldwide, causing approximately 1.3 million deaths annually. Automated computer-aided diagnosis (CAD) systems based on chest X-ray (CXR) imaging offer a promising solution for early TB screening, particularly in resource-limited settings. We present a systematic comparison of three pretrained CNN backbones - DenseNet-121, EfficientNet-B3, and MobileNet-V3 - for three-class CXR classification (Healthy, Sick-but-non-TB, and TB) using the TBX11K benchmark dataset under identical training conditions. MobileNet-V3 achieves the highest macro AUC of 99.98% with a TB sensitivity of 99.37%, while EfficientNet-B3 achieves perfect TB sensitivity of 100.00%. Grad-CAM visualizations confirm clinically meaningful activation patterns, and external validation on a confirmed TB case demonstrates real-world applicability.
References
- Ravi, V., Acharya, V., Alazab, M. (2023). A multichannel EfficientNet deep learning-based stacking ensemble for lung disease detection. Cluster Computing, 26, 1181–1203.
- Lin, T.Y., et al. (2017). Focal loss for dense object detection. ICCV, pp. 2980–2988.
- World Health Organization. (2023). Global Tuberculosis Report 2023. WHO, Geneva.
- Liu, Y., Wu, Y.H., Ban, Y., Wang, H., Cheng, M.M. (2020). Rethinking Computer-Aided Tuberculosis Diagnosis. CVPR, pp. 2735–2744.
- Häggström, M. (2014). Chest radiograph of miliary tuberculosis [Radiograph]. Wikimedia Commons.
- Kotei, E., Thirunavukarasu, R. (2024). A comprehensive review on deep learning techniques for automatic detection of tuberculosis from chest X-ray images. Archives of Computational Methods in Engineering, 31, 455–474.
- Tan, M., Le, Q.V. (2019). EfficientNet: Rethinking model scaling for convolutional neural networks. ICML, pp. 6105–6114.
- Huang, G., Liu, Z., van der Maaten, L., Weinberger, K.Q. (2017). Densely connected convolutional networks. CVPR, pp. 4700–4708.
- Selvaraju, R.R., et al. (2017). Grad-CAM: Visual explanations from deep networks via gradient-based localization. ICCV, pp. 618–626.
- Howard, A., et al. (2019). Searching for MobileNetV3. ICCV, pp. 1314–1324.
- Das, P.K., et al. (2025). OSLTBDNet: Orthogonal softmax layer-based tuberculosis detection network with small dataset. Biomedical Signal Processing and Control. https://doi.org/10.1016/j.bspc.2025.000953
- Mirugwe, A., Tamale, L., Nyirenda, J. (2025). Improving Tuberculosis Detection Through Transfer Learning and Deep Learning. JMIRx Med, 6, e66029.
- Rajpurkar, P., et al. (2017). CheXNet: Radiologist-level pneumonia detection on chest X-rays with deep learning. arXiv:1711.05225.