By Frank Kamuntu
In a major advance for global health and medical artificial intelligence, a groundbreaking study published in JMIRx Med has revealed that VGG16, a lightweight convolutional neural network (CNN), significantly outperforms more complex architectures such as ResNet152 and Inception-ResNet-V2 in detecting tuberculosis (TB) from chest X-ray (CXR) images. This landmark research, spearheaded by Victoria University’s Dr. Lillian Tamale, underscores the university’s unwavering commitment to fostering cutting-edge innovation and addressing critical global health challenges.
The study was a collaborative effort between Dr. Lillian Tamale, Head of the Department of Computing and Information Science in the Faculty of Science and Technology at Victoria University, Uganda; Mr. Alex Mirugwe of Makerere University; and Dr. Juwa Nyirenda from the University of Cape Town. Dr. Tamale’s exceptional leadership and Victoria University’s robust institutional support were instrumental in driving this research forward, affirming the university’s position as a regional hub for applied AI and digital health innovation. Victoria University’s dedication to providing an environment that nurtures such impactful research is truly commendable, setting a benchmark for other institutions.
Among the study’s most striking results, VGG16 achieved an impressive 99.4% classification accuracy, remarkably outperforming more computationally demanding CNN models. Notably, the research also challenges conventional assumptions in AI-based medical imaging by demonstrating that data augmentation techniques such as rotations and flips offered no performance gains, suggesting that the base dataset was sufficiently robust on its own. This finding is particularly significant for low-resource settings, where computational power and extensive data augmentation might be limitations.
This discovery arrives at a critical time. According to the World Health Organization (2023), TB remains one of the world’s leading causes of death from infectious disease, with over 10 million new cases and 1.5 million deaths annually. Yet, current diagnostic tools remain expensive, slow, and often inaccessible in rural and under-resourced settings. Victoria University, through this research, is directly contributing to a solution that could revolutionize TB diagnostics globally.
The research underscores the transformative potential of simpler, more resource-efficient AI models like VGG16 in addressing these diagnostic challenges. “This is a game-changer for AI in public health,” said Dr. Tamale. “With VGG16, even remote or underserved clinics can adopt AI-powered diagnostics without the need for advanced computing infrastructure. It’s a major step toward equitable healthcare access.” This statement perfectly encapsulates Victoria University’s vision of impactful research that benefits society at large.
To conduct the study, the research team evaluated six CNN architectures – VGG16, VGG19, ResNet50, ResNet101, ResNet152, and Inception-ResNet-V2 – using a dataset of 4,200 chest X-ray images (700 TB-positive and 3,500 normal cases). The models were assessed using standard evaluation metrics: accuracy, precision, recall, F1-score, AUC-ROC, as well as training time and computational efficiency. VGG16 emerged as the top performer across multiple metrics, showcasing its superior efficiency and effectiveness.
Perhaps one of the most surprising findings was that data augmentation – a commonly used method in deep learning to artificially expand datasets – had no measurable impact on model performance. This revelation may prompt a reevaluation of best practices in medical image classification, especially in settings where computational resources are limited, further highlighting the practical relevance of Victoria University’s research.
The broader implications are clear: Victoria University’s involvement, through the stewardship of Dr. Tamale, highlights the critical role of African universities in advancing cutting-edge solutions to global health challenges. The efficient, high-performing VGG16 model could accelerate the rollout of cost-effective, AI-driven TB screening tools in high-burden regions such as sub-Saharan Africa and South Asia, fulfilling Victoria University’s mission of making a tangible difference in the world.
The full study, titled “Improving Tuberculosis Detection in Chest X-Ray Images Through Transfer Learning and Deep Learning: Comparative Study of Convolutional Neural Network Architectures”, is available in the July 2025 edition of JMIRx Med. A preprint version was previously shared on medRxiv (2024), and the dataset used in the research is openly available on Kaggle, encouraging further research and practical deployment.
Victoria University’s commitment to open science and collaboration is evident in making this valuable resource accessible to the wider scientific community.