Benchmarking Deep Neural Networks for Lung Nodule Classification in LUNA25

Abstract

Early and accurate classification of pulmonary nodules as benign or malignant is critical for improving lung cancer survival rates, particularly when tumors are detected at an asymptomatic stage. In this study, we present a comprehensive benchmark of 2D and 3D deep learning models for malignancy risk estimation on the LUNA25 dataset. We explore a wide range of architectures, from classical CNNs like ResNet to modern models such as ConvNeXt. Furthermore, we assess the impact of Focal Loss versus Binary Cross-Entropy and propose a custom 3D ResNet model that outperforms other models. Our findings highlight that simple 3D architectures, when carefully optimized, offer significant gains in performance.