BSLungXNet: An Imbalance-Aware Deep Learning Model for Lung Cancer Prediction Using CT Images

Abstract

Lung cancer is one of the most difficult diseases to deal with, and early detection significantly improves patient survival rate. However, the lung cancer diagnosis process requires expertise, time and resources. Though the experts are good in cancer diagnosis, human error is inevitable. Those errors could cost a human life which may lead to a crisis in a family. A reliable AI model has to be designed to avoid such scenarios. To overcome these challenges, this study proposed BSLungXNet, a novel Convolutional Neural Network (CNN) based model to predict lung cancer from CT scan images. Unlike conventional CNN-based approaches, the proposed model specifically addresses class imbalance by generating informative synthetic samples near decision boundaries. It helped the model to learn more effectively and improved the prediction accuracy. The proposed BSLungXNet model predicted the classes with 92.41% of accuracy, 92.36% of precision, 92.41% of recall, and 92.38% of F1-Score. To understand the logic behind the prediction of the model, Grad-CAM was applied. It ensured the reliability and explainability of the proposed model. Hence, the BSLungXNet would be a reliable tool for the healthcare professional in the process of lung cancer diagnosis.