An Effective Approach of Lung Cancer Detection Using Xception-Driven with Optimized Feature Fusion

Abstract

Early detection of lung cancer is challenging, and current assessment methods like blood tests and CT scans are both time-consuming and require substantial human involvement. To tackle these challenges, the project proposes an innovative solution named Lung-RetinaNet, specifically designed for lung tumour detection. This system utilizes a RetinaNet model, enhanced with merging of multiscale features and a context module. The Lung-RetinaNet model integrates many layers of neural networks using a multi-scale feature fusion module.   This approach enhances the model's ability to gather semantic information, which is crucial for identifying lung tumours.   Using a dilated and lightweight context module technique, Lung-RetinaNet also employs multi-scale feature fusion.   To enhance feature extraction and localise tiny tumours in lung images, this module makes advantage of contextual information with each neural network layer.   These components improve the efficiency and accuracy of the project's lung tumour detection technology.   The accuracy of lung cancer categorisation is increased to 99% with the addition of the Xception model. Additionally, the inclusion of YOLOv5 and YOLOv8 for detection purposes enhances lung cancer detection in images. This multi-faceted approach ensures a comprehensive analysis of lung cancer cases, combining accurate classification with precise object detection.