Computer-aided Diagnosis (CAD) Of Solitary Pulmonary Nodule (SPN) In High Definition CT (HDCT)

Aims: This paper attempts to design a computer-aided diagnosis system(CADS) for solitary pulmonary nodule(SPN) based on high definition CT(HDCT) image to increase the detection rate of malignant SPN, making the diagnosis more objective and scientific.Methods: 120 cases of SPN confirmed by clinical pathology were collected(including malignant tumors, benign tumors, tuberculosis and inflammatory pseudotumor). 60 cases were randomly selected as experimental set, and other 60 cases as validation set. After image preprocessing, segmentation of region of interest(ROI) by tag-based watershed algorithm and texture feature parameter extraction of ROI for HDCT image of experimental set, statistical processing of 5 texture feature parameters that obtained was performed. The statistical results are applied to the system in order for the analysis of benign and malignant SPN and offer the prompting message. After entering HDCT image of validation set to the system, the system reliability was evaluated by comparing the predicting results of system and by chief & resident physician.Results: P values of t-test of contrast, correlation, entropy, homogeneity and second moment were 0.000, 0.002, 0.914, 0.295 and 0.002, respectively. Benign ranges of contrast, correlation and second moment were [903, 2003], [2.76, 3.48] and [0.01, 1.54], respectively. Malignant ranges of contrast, correlation and second moment were [502,898], [3.49, 3.71] and [1.79, 29.86], respectively. Sensitivity of the system, chief & resident physicians were 83.3%, 93.3% and 76.7%, respectively. The false positive rates were 13.3%, 16.7% and 26.7% respectively. Accuracy rate were 85%, 88.3% and 75%, respectively.Conclusions: Nodules with chest wall and mediastinal adhesions, and ground-glass lesions could be better splited and extracted by tag-based watershed algorithm. There was statistical significance for contrast, correlation and second moment. On the predicting of benign and malignant SPN, the system has high sensitivity and accuracy, and lowest false positive. CADS has some clinical value in the diagnosis of benign and malignant SPN. This system can assist the diagnosis of benign and malignant SPN for clinicians.