Study on calcific signs in pulmonary nodules based on pattern classification / 实用放射学杂志

ABSTRACT

Objective To study feasibility in extraction of calcific sign within pulmonary nodules with pattern classification.Meth-ods 49 cases with pulmonary nodules (benign in 16 and malignant in 33)confirmed by pathology or clinical follow-up were included in this study and all cases underwent chest CT examinations.CT images were interpreted double-blind by two associate chief radiolo-gists to draw a conclusion that there were any calcification within pulmonary nodules.Meanwhile,the calcifications in the regions of interest(ROI)on CT images were estimated with extraction of the sign of gray value,geometric and lung markings in ROI,and based on pattern classification algorithm at supporting vector machine(SVM).Results According to the results assessed by senior radiologists for classification within pulmonary nodules,the area under ROC curve was 0.95 which was extracted by automatic pat-tern classification algorithm,the extraction performance was stable(k=1),and was goodness fit with visual observation by doctors (k=0.939).Conclusion The ability of automatic pattern classification in detecting calcification within pulmonary nodules is about the same as that of visual assessment by senior doctors.