Enhanced Lung Cancer Detection Using Custom Convolutional Neural Network with Comparative Analysis of Deep Learning Models

Abstract

Lung cancer dominates cancer-related deaths across the globe, and its early diagnosis is the only way for effective treatment. The diagnosis of lung cancer has improved using CT scans along with deep learning techniques. The custom CNN architecture is proposed to classify chest Computed Tomography (CT) scan images into three categories namely normal, benign and malignant lung tissue. For the evaluation of the proposed CNN architecture, various classification metrics have been used, considering their performance with VGG16 and ResNet50, the other two advanced deep learning models. It was concluded that the proposed custom CNN outperformed VGG16 and ResNet50 with an accuracy of 97.6%, precision rate of 98%, recall rate of 98%, F1 score of 98% and proved to be effective as a model for accurate classification even in cases presenting nodules of smaller size. These results showed the potential of deep learning approach of custom CNNs for detecting lung cancer with high accuracy and automation, thus offering a promising tool to support radiologists in further early diagnosis and treatment planning.