The global health problem of Tuberculosis impacts public health centers primarily in regions where resources are limited. Proficient radiologists currently play the key role in identifying Pulmonary Tuberculosis (PTB) within chest X-ray images. The proposed system implements deep learning through Convolutional Neural Networks (CNNs) to develop an automated tuberculosis detection system that delivers accurate results. The application functions as an Android program that lets users add X-ray images through an intuitive interface while it analyzes symptoms to show diagnostic assessments. For optimal patient care the system incorporates features to prevent diseases through education and it provides doctor reminder services and symptom diagnosis capabilities to support healthcare professionals. The suggested system intends to resolve time-sensitive TB diagnosis problems by providing a flexible and accessible diagnostic solution for both medical staff and patients.
Introduction
???? Overview
Pulmonary Tuberculosis (PTB) is the most common form of TB and ranks among the top 10 global causes of death.
Timely and accurate diagnosis is vital but challenged by a shortage of trained radiologists, especially in high-burden countries like India, China, Indonesia, the Philippines, and Nigeria.
AI-powered systems using chest X-rays and deep learning aim to improve early detection, especially in underserved areas.
???? Purpose of the Study
To develop a mobile application that uses Convolutional Neural Networks (CNNs) for:
Detecting PTB from chest X-rays.
Assessing risk via symptom-based questionnaires.
The system is designed for resource-limited and remote regions with minimal access to healthcare professionals.
???? System Features
Image-Based Diagnosis: Uses CNNs (e.g., EfficientNetV2) to analyze X-rays after preprocessing (contrast enhancement, normalization).
Symptom-Based Diagnosis: Utilizes structured surveys analyzed by machine learning for TB risk estimation.
Hybrid Detection Model: Combines both image and symptom analysis for improved accuracy.
Mobile & Web Application: Built with Android Studio, Django (backend), and MySQL for secure data storage.
User Tools: Includes medication reminders, doctor recommendations, and health tracking.
???? Related Work
Prior studies have used datasets (e.g., Shenzhen, Montgomery, TBX11K) with CNN architectures (VGG16, ResNet, DenseNet).
Techniques like Grad-CAM, segmentation, and contrast enhancement have improved TB lesion localization and diagnosis.
Deep learning consistently outperforms traditional methods like SVM or decision trees in TB detection.
?? Methodology
Requirement Analysis: Define healthcare gaps and user needs.
CNN Model Development: Trained on labeled chest X-ray data.
Symptom Analysis: Cough, fever, weight loss, etc., are evaluated through a structured form.
App Development: Cross-platform app using Django, HTML, CSS, JavaScript.
Education Module: Includes TB prevention strategies (e.g., ventilation, hygiene).
Testing & Validation: Includes usability and performance tests.
Deployment: Real-time cloud support and regular updates based on feedback.
???? Implementation Architecture
Three-Layer Design:
User Layer: Manages login, symptom/X-ray input, and results display.
Application Layer: Handles backend processing, data storage, and API communication.
Diagnosis Layer: Runs CNN for image-based classification and ML for symptom-based predictions.
Uses CLAHE for image enhancement, and TensorFlow Lite for mobile efficiency.
???? Results
Training Accuracy: Reached 96.50%; Validation Accuracy: Peaked at 97.02%.
Test Accuracy: 95.72%.
Confusion Matrix:
True Positives: 52
False Positives: 3
True Negatives: 1
False Negatives: 0 (indicating high sensitivity)
Demonstrates strong performance with low error rate, reliable for clinical use.
Conclusion
The System has successfully built a mobile application for early detection of pulmonary tuberculosis utilizing chest X-ray pictures. The technology produces reliable and efficient diagnostic results, decreasing the need for radiologists to analyze data manually. The results show that the suggested methodology improves early detection, treatment outcomes, and TB transmission. The application\'s user-friendly interface and extra capabilities, such as symptom-based analysis and prevention measures, make it a comprehensive tool for tuberculosis management. Future work will focus on increasing the dataset, enhancing the model\'s accuracy, and including real-time medical consultation elements to further increase the system\'s capabilities.
References
[1] Kartik K. Goswami , Rakesh Kumar , Rajesh Kumar , Akshay J. Reddy , Sanjeev K. Goswami, “Deep-learning classification of tuberculosis chest xray”,2023.
[2] Twinkle Bansal, Sheifali Gupta, Neeru Jindal, “Deep learning based comprehensive review on pulmonary tuberculosis”,2023.
[3] M. Balamurugan , R. Balamurugan, “An efficient deep neural network model for tuberculosis detection using chest xray images”,2023.
[4] TianhaoXu, Zhenming Yuan, “Convulation neural network with coordinate attention for the automatic detection of pulmonary tuberculosis”,2022.
[5] TawsifurRahman, AmithKhandakar, Muhammad Abdul Kadir, Khandaker R. Islam, Khandaker F. Islam, Rashid Mazhar, Tahir Hamid, Mohammad T. Islam, Zaid B. Mahbub, Mohamed ArseleneAyari, Muhammad E. H. Chowdhury, “Reliable Tuberculosis Detection using Chest X-Ray with Deep Learning, Segmentation and Visualization”,2020.