Lung Tumor Detection and Localization System Using Faster RCNN Algorithm

Lung tumors represent a significant health challenge. Although, their early detection can improve patient outcomes, traditional diagnostic methods often fall short in accuracy and efficiency. Therefore, advanced algorithms are required to facilitate accurate and earlier tumor detection and localization. This study aims to enhance detection efficacy through a systematic methodology that includes image pre-processing, segmentation, feature extraction, classification, and localization. We have invested three CNN models; Our proposed CNN, VGG-16 and RESNET50 integrated with Faster R-CNN to detect and localize lung tumor in computed tomography images from the Iraq-Oncology Teaching Hospital/National Center for Cancer Diseases (IQ-OTH/NCCD) dataset. As a result, VGG-16 and RESNET50 attained detection accuracies of 97.66% and 97.20% respectively, while Our-CNN model achieved a detection accuracy of 97.20%, with an average test precision of 97.185 and an F1-score of 0.9737. For localization, ResNet50 and Our-CNN demonstrated accuracy based on Intersection over Union (IoU) at 89.47%, while VGG-16 achieved 73.68%.