A 3D prediction model for benign or malignant of pulmonary nodules based on neural architecture search

Lung cancer poses a huge threat to human life and health. Early diagnosis and early treatment are the most effective means to improve patient survival and reduce mortality. Aiming at the low correlation among dimensional features of lung CT 3D image data and low accuracy of single manually designed convolutional neural network, this paper proposes a 3D prediction model for benign or malignant of pulmonary nodule based on neural architecture search. The main contribution is to design a cross-dimensional interactive quadruple attention module to increase the feature extraction and feature representation capabilities of the 3D classification network for pulmonary nodules. Moreover, a multi-model prediction fusion method based on multi-ensemble learning algorithms is designed to improve the reliability of the result. The experimental results on LUNA16 data set achieve 90.85% Accuracy, 93.02% AUC and 88.89% F1 value, thus showing that the proposed approach has high accuracy and reliability.