Research Article
Enhancing Tuberculosis Detection with Deep Learning: A CNN-Based Approach with Data Augmentation and Regularization
@INPROCEEDINGS{10.4108/eai.28-4-2025.2357819, author={D. J. Ashpinpabi and R. Nidhya and D. Renuka and S. Salva and K. SaiSushma and A. Rajesh}, title={Enhancing Tuberculosis Detection with Deep Learning: A CNN-Based Approach with Data Augmentation and Regularization}, proceedings={Proceedings of the 4th International Conference on Information Technology, Civil Innovation, Science, and Management, ICITSM 2025, 28-29 April 2025, Tiruchengode, Tamil Nadu, India, Part I}, publisher={EAI}, proceedings_a={ICITSM PART I}, year={2025}, month={10}, keywords={deep learning convolutional neural networks (cnns) support vector machine (svm) data augmentation feature extraction computer-aided detection radiological imaging}, doi={10.4108/eai.28-4-2025.2357819} }- D. J. Ashpinpabi
R. Nidhya
D. Renuka
S. Salva
K. SaiSushma
A. Rajesh
Year: 2025
Enhancing Tuberculosis Detection with Deep Learning: A CNN-Based Approach with Data Augmentation and Regularization
ICITSM PART I
EAI
DOI: 10.4108/eai.28-4-2025.2357819
Abstract
Tuberculosis (TB) remains a major global health concern, requiring early and accurate diagnosis for effective treatment. Traditional radiological assessments face challenges in distinguishing TB from other pulmonary diseases. Recent studies, such as the Multiscale Eigendomain Gradient Boosting (MEGB) approach, have attempted to automate TB detection using handcrafted features from chest X-rays, achieving 96.42% accuracy. However, feature extraction-based methods may suffer from generalization issues. In this study, we propose a deep learning-based Convolutional Neural Network (CNN) model for automated TB detection, incorporating data augmentation and dropout layers to enhance generalization and prevent overfitting. Our best performing CNN model, with three dropout layers, achieves 99.32% accuracy, significantly outperforming previous methods. Additionally, we compare CNN with a Support Vector Machine (SVM) classifier, achieving 93% accuracy. Our results demonstrate that deep learning models can effectively learn spatial features, providing superior diagnostic accuracy and robustness compared to feature extraction-based approaches.
