Home / Current Issue / Paper 1716815
Automated Detection of Pulmonary Tuberculosis from Chest X-Rays Using Fine-Tuned Convolutional Neural Networks
Subject area: Science,Engineering and Technology · Area of research: Health Care
DOI: https://doi.org/10.64388/IREV9I10-1716815
Abstract
Tuberculosis (TB) remains one of the most severe infectious diseases globally, claiming over one million lives annually. Early and accurate diagnosis is critical for effective treatment and containment. Conventional diagnostic methods, including sputum smear microscopy and culture tests, are time-consuming and demand significant clinical expertise. This paper presents an automated deep learning system for binary classification of pulmonary tuberculosis from chest X-ray images using a fine-tuned VGG16 Convolutional Neural Network (CNN). The proposed system leverages transfer learning from ImageNet-pretrained weights and incorporates a two-stage training strategy: initial training of custom classification layers followed by selective fine-tuning of the final convolutional block. The model is trained and evaluated on publicly available chest X-ray datasets from the ChinaSet (662 images) and Montgomery County (138 images), totalling 800 annotated radiographs. Extensive data augmentation techniques are applied to address the limited dataset size. The system achieves a validation accuracy of approximately 95%, demonstrating strong diagnostic capability. A critical data pipeline issue related to class-sorted ordering was identified and resolved, which dramatically improved TB recall from 0% to clinically meaningful levels. Evaluation metrics include a confusion matrix, classification report with precision, recall and F1-score, and ROC-AUC analysis. The proposed pipeline is modular, reproducible, and supports both training and real-time single-image inference modes.
Keywords
Tuberculosis Detection, Transfer Learning, VGG16, Convolutional Neural Network, Chest X-Ray, Medical Image Classification, Deep Learning, Binary Classification, Fine-Tuning, Data Augmentation, TensorFlow, Keras
How to cite this paper
@article{1716815,
author = {Prashant Saraswat, Ananya Rai, Devesh Thakur, Dr. Danish, Prof. (Dr.) Sanjay Pachauri},
title = {Automated Detection of Pulmonary Tuberculosis from Chest X-Rays Using Fine-Tuned Convolutional Neural Networks},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {9},
number = {10},
pages = {2806-2814},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1716815.pdf},
abstract = {Tuberculosis (TB) remains one of the most severe infectious diseases globally, claiming over one million lives annually. Early and accurate diagnosis is critical for effective treatment and containment. Conventional diagnostic methods, including sputum smear microscopy and culture tests, are time-consuming and demand significant clinical expertise. This paper presents an automated deep learning system for binary classification of pulmonary tuberculosis from chest X-ray images using a fine-tuned VGG16 Convolutional Neural Network (CNN). The proposed system leverages transfer learning from ImageNet-pretrained weights and incorporates a two-stage training strategy: initial training of custom classification layers followed by selective fine-tuning of the final convolutional block. The model is trained and evaluated on publicly available chest X-ray datasets from the ChinaSet (662 images) and Montgomery County (138 images), totalling 800 annotated radiographs. Extensive data augmentation techniques are applied to address the limited dataset size. The system achieves a validation accuracy of approximately 95%, demonstrating strong diagnostic capability. A critical data pipeline issue related to class-sorted ordering was identified and resolved, which dramatically improved TB recall from 0% to clinically meaningful levels. Evaluation metrics include a confusion matrix, classification report with precision, recall and F1-score, and ROC-AUC analysis. The proposed pipeline is modular, reproducible, and supports both training and real-time single-image inference modes.},
keywords = {Tuberculosis Detection, Transfer Learning, VGG16, Convolutional Neural Network, Chest X-Ray, Medical Image Classification, Deep Learning, Binary Classification, Fine-Tuning, Data Augmentation, TensorFlow, Keras},
month = {April},
doi = {https://doi.org/10.64388/IREV9I10-1716815}
}