Nodules Detection in Lungs CT Images Using Improved YOLOV5 and Classification of Types of Nodules by CNN-SVM

Abstract

Lung nodule detection in computed tomography (CT) imaging is crucial for the early diagnosis and treatment of lung disorders. This study explores the application of deep learning algori thms, specifically YOLOv5, Faster R-CNN, and a proposed improved YOLOv5 (iYOLOv5) for detecting lung nodules. Using the LIDC-IDRI dataset, models are trained to recognize lung nodules accurately. The proposed iYOLOv5 model consistently outperformed the baseline models, achieving an impressive mean Average Precision (mAP) of 0.853 at a learning rate of 0.01 and a batch size of 32, compared to YOLOv5’s mAP of 0.793. Additionally, iYOLOv5 achieved a mAP of 0.864 with a learning rate of 0.001 and a batch size of 32, surpassing YOLOv5’s 0.858. Proposed iYOLOv5 also demonstrated better convergence during training, with lower loss values across various configurations, such as 0.0177 at a learning rate of 0.1 and a batch size of 32, compared to YOLOv5’s 0.0217. Additionally, lung nodules were classified into well-circumscribed, juxta-vascular, and juxta-pleural categories using CNN and CNN-SVM, with CNN-SVM showing promising results with an accuracy of 98.17%. These findings underscore the potential of the iYOLOv5 model in enhancing lung nodule detection and classification, contributing to improved diagnostic accuracy in medical imaging.

Keywords

• CT images
• faster RCNN
• lung nodules
• object detection
• YOLOV5
• iYOLOV5