A hybrid classifier for mammogram mass classification using various new geometric shape and margin features. (17th June 2011)

• Record Type: Journal Article
• Authors: Surendiran, B.; Vadivel, A.
• Is Part Of: International journal of rapid manufacturing. Volume 2:Number 1/2(2011)
• Page Start: 56
• Page End: 75
• Publication Date: 2011-06-17
• Subjects: discriminant analysis -- digital mammograms -- geometric shape -- margin features -- hybrid classifiers -- ANOVA -- analysis of variance -- neural networks -- RBFN -- radial basis function networks -- benign masses -- malignant masses -- breast imaging -- mammogram mass classification -- breast screening; Manufacturing processes -- Periodicals; 670.4205
• Languages: English
• ISSNs: 1757-8817

Abstract

In this paper, a hybrid classifier system is proposed, which combines statistical classifier analysis of variance (ANOVA) discriminant analysis (ADA) and neural network (NN)/radial basis function network (RBFN). According to breast imaging reporting and data system, the benign and malignant masses can be characterised by its shape, size and margins. Benign masses tend to have round, oval, lobular shapes and malignant masses are lobular or irregular in shape. Various 17 geometrical shape and margin features were introduced to characterise the morphology of masses. The mammogram mass is classified as either benign or malignant. The proposed features do not rely on grey values of the masses, which make it effective compared to statistical measures. The ADA classifier produces discriminant function score, which is used as extra feature combined with existing 17 features and classified using NN and RBFN. The proposed hybrid ADA?NN and ADA?RBFN classifier system shows improvement in classification accuracy over conventional single classifiers. The highest classification rate achieved is 91.56%. The results indicate that the hybrid classifier approach contributes greatly for the classification of mammogram mass into benign and malignant.