Improving lung cancer detection via MobileNetV2 and stacked-GRU with explainable AI

Accurate and timely detection of lung cancer is crucial for effective treatment planning. This study introduces MobileNetV2-SGRU, a novel transfer learning-based predictor for lung cancer classification. It utilizes MobileNetV2 to automatically extract meaningful features from lung CT images. These extracted features processed through stacked gated recurrent unit layers, enabling the model to capture the sequential and temporal information present in the images. Additionally, the model incorporates Grad-CAM, an explainable artificial intelligence technique, to enhance interpretability and transparency of the model. An evaluation of MobileNetV2-SGRU on the IQ-OTH/NCCD dataset demonstrated its outstanding performance in lung cancer classification. The model achieves impressive accuracy, precision, recall, and F1-score of 96.83%, 96.78%, 96.83%, and 96.78%, respectively, demonstrating its superior predictive capabilities compared to existing methods. Furthermore, the results indicate reduced computation time and enhanced accuracy in diagnosing lung cancer. With advanced performance metrics and the incorporation of Grad-CAM visualization for interpretability, the model offers valuable insights for clinicians in validating and understanding its predictions, ultimately contributing to more effective treatment planning.