Sequential ResNet Framework for Lung Cancer Classification: A Comparative Analysis of Bottom-Up and Top-Down Approaches

Lung cancer detection in histopathological images is pivotal in timely and accurate diagnosis, directly influencing treatment outcomes and patient survival rates. This research paper presents an automated detection and classification system utilizing ResNet architectures. Initially, individual evaluations of ResNet-50, ResNet-101, and ResNet-152 are conducted. Subsequently, a sequential ResNet framework is developed to assess the combined performance of these architectures. In this framework, both top-down (ResNet-152 ➔ ResNet-101 ➔ ResNet-50) and bottom-up (ResNet-50 ➔ ResNet-101 ➔ ResNet152) approaches are employed for classification tasks. The effectiveness of the framework is measured using standard assessment metrics, including Accuracy, G-Mean, F1-score, Precision, Recall, and ROC curves for each architecture. The results signifies that the bottom-up approach achieves the outstanding performance in accuracy of 93.28% with an AUC value of 0.9800, indicating excellent classification performance and minimal false positives.