An artificial intelligence-assisted diagnostic system for the prediction of benignity and malignancy of pulmonary nodules and its practical value for patients with different clinical characteristics

Abstract

Objectives: This study aimed to explore the value of an artificial intelligence (AI)-assisted diagnostic system in the prediction of pulmonary nodules. Methods: The AI system was able to make predictions of benign or malignant nodules. 260 cases of solitary pulmonary nodules (SPNs) were divided into 173 malignant cases and 87 benign cases based on the surgical pathological diagnosis. A stratified data analysis was applied to compare the diagnostic effectiveness of the AI system to distinguish between the subgroups with different clinical characteristics. Results: The accuracy of AI system in judging benignity and malignancy of the nodules was 75.77% (p < 0.05). We created an ROC curve by calculating the true positive rate (TPR) and the false positive rate (FPR) at different threshold values, and the AUC was 0.755. Results of the stratified analysis were as follows. (1) By nodule position: the AUC was 0.677, 0.758, 0.744, 0.982, and 0.725, respectively, for the nodules in the left upper lobe, left lower lobe, right upper lobe, right middle lobe, and right lower lobe. (2) By nodule size: the AUC was 0.778, 0.771, and 0.686, respectively, for the nodules measuring 5–10, 10–20, and 20–30 mm in diameter. (3) The predictive accuracy was higher for the subsolid pulmonary nodules than for the solid ones (80.54 vs. 66.67%). Conclusion: The AI system can be applied to assist in the prediction of benign and malignant pulmonary nodules. It can provide a valuable reference, especially for the diagnosis of subsolid nodules and small nodules measuring 5–10 mm in diameter.