Enhancing early lung cancer detection with MobileNet: A comprehensive transfer learning approach
Raquel Ochoa-Ornelas, Alberto Gudiño-Ochoa, Julio Alberto García-Rodríguez, Sofia Uribe-Toscano
Abstract
Lung cancer is the leading cause of cancer-related mortality globally, with over 2.2 million new cases diagnosed annually. Early detection is critical for improving patient outcomes but remains a significant challenge due to the intricate nature of histopathological image analysis, which requires specialized expertise. This study investigates the application of MobileNetV2, a state-of-the-art, lightweight convolutional neural network, for the accurate classification of lung adenocarcinoma (LAC), benign lung tissue (BLT), and lung squamous cell carcinoma (LUSC). To enhance the model’s generalization capabilities, we augmented the widely used LC25000 dataset with additional histopathological images sourced from the National Cancer Institute. The final model exhibited exceptional performance, achieving a training accuracy of 99.11%, a validation accuracy of 97.25%, and a test accuracy of 97.65%. The model also delivered high precision, recall, and F1-scores, each exceeding 98% across all classes, with perfect metrics observed for BLT. These promising results highlight the potential of MobileNetV2 as a highly effective tool for the early detection of lung cancer, offering substantial support for clinical diagnosis. By facilitating more accurate and timely diagnoses, this approach has the potential to significantly enhance patient care and improve survival rates in lung cancer patients. • High Accuracy: The model achieved test accuracy of 98.77% in detecting lung cancer. • Dataset Generalization: Augmenting the LC25000 dataset with new images from NIH. • Clinical Relevance: High precision and recall for benign lung tissue.