Enhancing lung cancer diagnosis in Uganda using convolutional neural networks on CT images from Uganda and Iraq

Abstract

Lung cancer remains the leading cause of cancer-related mortality worldwide, with over 2.2 million new cases and 1.8 million deaths annually. Early detection significantly improves survival rates; however, in low- and middle-income countries (LMICs) such as Uganda, diagnosis often occurs at advanced stages, leading to poor treatment outcomes and high mortality. Computed Tomography (CT) scans are the gold standard for early detection, but their impact is constrained by a severe shortage of radiologists and pulmonologists, Uganda has fewer than 50 radiologists and under ten pulmonologists for over 45 million people. Despite government investments in CT infrastructure, including regional scanner installation and a National Tele-Radiology Centre, specialist shortages continue to delay diagnosis and underutilize imaging capacity.

This study developed and validated a lightweight, 10-layer Convolutional Neural Network (CNN) for automated classification of lung CT scans into cancerous and non-cancerous categories. Using a combined dataset from the public Iraq-Oncology Teaching Hospital/National Center for Cancer Diseases (IQ-OTH/NCCD) repository and de-identified CT scans from Mulago National Referral Hospital. The final model achieved a test-set accuracy of 98%, 97.5% sensitivity, 98.0% specificity, and an area under curve (AUC) of 1.00 on the test set. To address the gap between algorithm development and clinical use, the model was integrated into a web-based diagnostic interface built with Flask backend and React.js frontend, enabling real-time predictions from uploaded scans. Manual testing of 20 unseen images via the interface yielded 95% accuracy, confirming operational stability outside controlled environments.

By reducing reliance on scarce specialist interpretation, optimizing existing CT infrastructure, and enabling faster, more accurate lung cancer detection, this system has potential to improve early diagnosis rates and reduce mortality in Uganda. Its Central Processing Unit (CPU), only deployment capability ensures suitability for resource-limited settings, supporting scalable integration into national cancer control programs and public health screening initiatives.

Key Words: Lung cancer, Convolutional Neural Networks, early detection, CT scans, deep learning, Uganda, artificial intelligence, medical imaging, web-based application