Establishing Assistant Diagnosis Models of Solitary Pulmonary Nodules Based on Intelligent Algorithms

• Published in: Cellular physiology and biochemistry Vol. 35; no. 6; pp. 2463 - 2471
• Main Authors: Zhao, Zhijun, Chen, Jingtao, Yin, Xiaoxiang, Song, Huayong, Wang, Xinchun, Wang, Jing
• Format: Journal Article
• Language: English
• Published: Basel, Switzerland Cell Physiol Biochem Press GmbH & Co KG 01.01.2015
• Subjects: Adult; Aged; Algorithms; Biomarkers - metabolism; Cell Differentiation - physiology; CT imaging features; Diagnosis model; Female; Humans; Inflammation - diagnosis; Inflammation - metabolism; Inflammation - pathology; Intelligent algorithm; Lung Neoplasms - diagnosis; Lung Neoplasms - metabolism; Lung Neoplasms - pathology; Male; Middle Aged; Models, Theoretical; Original Paper; Sensitivity and Specificity; Solitary pulmonary nodule; Solitary Pulmonary Nodule - diagnosis; Solitary Pulmonary Nodule - metabolism; Solitary Pulmonary Nodule - pathology; Tomography, X-Ray Computed - methods; Tuberculosis - diagnosis; Tuberculosis - metabolism; Tuberculosis - pathology; Tumor markers; Young Adult; Intelligent algorithm; Diagnosis model; Tumor markers; Solitary pulmonary nodule; CT imaging features
• ISSN: 1015-8987; 1421-9778; 1421-9778
• DOI: 10.1159/000374046

Aims: This study aimed to establish an auxiliary diagnosis model for solitary pulmonary nodules (SPNs) and evaluate its test efficacy. Methods: Three hundred thirty-two pathologically diagnosed SPN patients (186 malignant, 146 benign) were collected as subjects. The serum levels of 8 types of markers and 9 computed tomography (CT) imaging features of each patient were treated as independent variables. The corresponding pathological classification results (fungal inflammation, tuberculosis and tuberculoma, inflammatory pseudotumor, tumor middle differentiation, cancer) of quantized patients were treated as dependent variables. A 17-to-1 mathematical auxiliary SPN diagnosis model was established using a back propagation (BP) algorithm and a support vector machine (SVM) algorithm. A 40-case test set was used to estimate the effect. Results: Two different auxiliary SPN diagnosis models were successfully established. The diagnostic accuracy, sensitivity and specificity of the BP algorithm diagnosis model were 60%, 68% and 46.7%, respectively, and those of the SVM algorithm model were 80%, 85.7% and 66.7%, respectively. Conclusion: The accuracy, sensitivity and specificity of the SVM diagnostic model were relatively high, indicating that the model has important reference value for determining the degree of SPN differentiation and is suitable for the auxiliary diagnosis of benign and malignant SPN.