Advanced Deep Learning Techniques for Lung Cancer Diagnosis: The VDSFusion-Net Approach

Abstract

This paper introduces an innovative approach for identifying and categorizing lung cancer using the VDSFusion-Net model, which merges VGG16 and DenseNet121 with an SVM classifier. The suggested method entails a meticulous preprocessing pipeline, which encompasses edge-directed interpolation, Wiener filtering, and CLAHE. This is then followed by segmentation utilizing Kapur's thresholding and morphological processing. The utilization of RANSAC in circle detection enhances the precision of tumor identification. The VDSFusion-Net model utilizes sophisticated methods for extracting features and classifying data, resulting in exceptional performance metrics: an accuracy of 99.32%, sensitivity of 99.45%, and specificity of 99.7%. The VDSFusion-Net model demonstrates its usefulness and uniqueness in accurately classifying lung cancer, providing substantial enhancements in diagnostic accuracy and reliability.