Stratified method in order to overcome the number of lop-sided cases in lung knot error detection decreasing with computer support

摘要

It is method for computer support detection (CAD) of the attention territory which is detected inside the HRCT medical graphics data and classification. The method is applied includes the post CAD machine learning technology which the degree of uniqueness of the identification of the territory/volume and in order to convert sensitivity maximally the knot or as a non knot. The territory is identified by CAD processing, is divided automatically. The feature pool, each from the territory where it is divided is identified and is extracted and, is processed by the be inherited algorithm which identifies optimum feature subset. Then, because balance of the number of cases in the class which differs is maintained data stratified method is used. The subset which is decided by GA is used although the support vectoring machine in order to classify the candidacy territory/the volume which is discovered inside the non training data is trained.