Deep Learning-Based Classification of Lung Cancer in CT Images Using Transfer Learning and Random Forest

Timothy Olukunle Olaniyi

Abstract: Lung cancer is the leading cause of cancer-related deaths globally, with late diagnoses contributing to poor survival rates. Early diagnosis and accurate detection of the lung cancer stage can save the patients? lives. However, as early detection is vital, it is often hindered by non-specific symptoms and limitations due to traditional diagnostic methods. Several image processing, biomarker based, and machine automation approaches are used to identify lung cancer, but accuracy and early diagnosis are challenging for medical practitioners. This study presents a compelling approach to early lung cancer detection using CT scan images, leveraging deep learning for feature extraction and machine learning for classification. By implementing pretrained models like VGG16 and DenseNet121 alongside a Random Forest classifier, the research aims to improve diagnostic accuracy while minimizing computational cost. It is evident that the VGG16 + RF combination outperformed DenseNet121 in both 70/30 and 80/20 training-test splits, achieving up to 95.9% accuracy. This suggests that hybrid architecture can play a vital role in resource-constrained settings, such as rural healthcare. However, limitations due to dataset homogeneity must be acknowledged, with future work proposed around diverse datasets and explainable AI to enhance generalizability and clinical trust.

Keywords: Lung Cancer Detection, CT Scan Classification, Transfer Learning, Random Forest Classifier, Deep Learning