Lung cancer detection and classification using LeNet-LSTM model on computed tomography images

Lung cancer is one of the most prevalent and deadly diseases, characterized by the uncontrolled growth of cells in lung tissues. It contributes significantly to rising global mortality rates. Medical professionals believe that early diagnosis of lung cancer using computed tomography (CT) screening can considerably reduce mortality in patients with lung nodules. However, the large quantity of CT images generated during screening might complicate the diagnosis procedure, making it difficult for radiologists to effectively identify nodules due to the volume of data. To address this challenge, various handcrafted and learned techniques have been developed to assist radiologists. In this study, we propose a hybrid LeNet-LSTM model for the classification of lung cancer. The LeNet model, which included convolutional and max-pooling layers, serves as the foundation for many modern deep-learning architectures. It consists of multiple convolutional layers followed by fully connected layers and employs activation functions. On the other hand, Long Short-Term Memory (LSTM) networks are designed to process sequential data, using advanced memory cells to overcome the vanishing gradient problem typically encountered in traditional recurrent neural networks. Our proposed model is evaluated against other pre-trained models as a feature extractor and compared to the latest classification techniques. The results demonstrate that our framework achieves 99.27% accuracy, 97.77% sensitivity, and a 99.26% F1-Score. These metrics indicate that the proposed algorithm outperforms previous approaches, making it highly effective for lung cancer detection. This research holds significant potential for aiding medical professionals and improving patient outcomes by accurately distinguishing between benign, malignant, and normal tumors.