Abstract: Currently, lung cancer is one of the leading causes of morbidity and mortality worldwide. Accurate classification of benign and malignant pulmonary nodules will help the clinicians to accurately diagnose in stages and make the optimal treatment planning in time. To address the low accuracy rate problem for the existing classification algorithms of benign and malignant pulmonary nodules, a classification algorithm based on radiomic feature and random forests model for the classification of benign and malignant pulmonary nodules is proposed in this paper. Firstly, a novel multiscale dot enhancement filter is proposed for pulmonary nodule enhancement. Then, seeds are accurately acquired based on shape index and texture features in enhanced pulmonary nodules, and the acquired seeds are injected into the RW algorithm to accurately segment pulmonary nodules. Secondly, the intensity, texture, shape, wavelet, and clinical features are extracted in the segmented pulmonary nodules. Finally, random forests (RFs) are employed to build the predictive model for classifying benign and malignant pulmonary nodules. The LIDC database is used to train the predictive model. The experiments demonstrate the effectiveness of the proposed algorithm for the classification of benign and malignant pulmonary nodules. The accurate classification results can provide a reference for the clinical diagnosis of lung cancer and a good prognostic value.

Key words: pulmonary nodules, image classification, malignancy, random walker, random forests, radiomics