Classification of Benign and Malignant Pulmonary Nodules Based on Mixed Features

Lung cancer is a pervasive malignancy that remains the leading cause of cancer-related mortality worldwide
^[1]
. Accurate diagnosis of benign and malignant lung nodules is crucial for the prevention and treatment of lung cancer. However, traditional methods of manually designed features face challenges in extracting and analyzing deep-level features using mathematical models. Meanwhile, methods using convolutional neural networks can extract deep-level features, but lack interpretability and are not as effective as manually designed feature methods for shallow visual features. To address these challenges, we propose a lung nodule classification method that combines shallow visual and deep learning features. We construct shallow visual and deep learning networks to extract and classify shallow visual features and deep learning features, respectively. Finally, we use a multi-model fusion strategy to achieve benign and malignant classification of lung nodules. In particular, we employ a neural network architecture search to build a deep learning network with better interpretability and performance. We conducted extensive experiments on the LIDC-IDRI dataset and compared our method with the state-of-the-art research. Our results show that our method outperforms existing methods with an accuracy and F1 score of 91.21% and 91.04, respectively. demonstrating the effectiveness and superiority of our algorithm.