Abstract: Feature selection and classification of region of interest(ROI) is investigated in this paper.Firstly,several features of ROI are extracted according to medical symptoms,and in order to improve the classification accuracy,these features are selected by their separability of probability distribution.Secondly,the selected features are described quantita-tively by SVM.Quantitative feature parameters are used to improve Mahalanobis distance with weight,and the improved Mahalanobis distance can enlarge the difference between the two ROIs.Finally,ROIs are classified into nodule and non-nodule by the improved weighted Mahalanobis distance.The presented ROI feature selection and classification algorithm is used in lung nodule detection experiment.The experiment results indicate that the detection sensitivity is 94.6% and the omission diagnostic rate is 5.4%,and the presented algorithm can provide doctors with helpful information for early stage diagnosis of lung cancer.