Open AccessAnalysis of Shape-Based and Texture-Based Attributes in Classification of Mammographic Findings by Machine Learning Algorithms
Author(s) -
Matheus De Melo,
Andy Gajadhar,
Hugo De Oliveira,
Arnaldo De Andrade e Silva,
Leonardo Vidal Batista
Publication year - 2015
Language(s) - English
Resource type - Conference proceedings
DOI - 10.5753/sbcas.2015.10364
Subject(s) - artificial intelligence , computer science , texture (cosmology) , pattern recognition (psychology) , machine learning , algorithm , image (mathematics)
Breast cancer is the most frequent cancer type among women. We present a method of classification of nodules (malignant or benign) found in mammograms using shape-based attributes and texture-based ones. Firstly, we built a test database, then we segmented and extracted a Gray Level Cooccurrence Matrix (GLCM) from each mammographic finding and analyzed texture-based and shape-based attributes. Finally, classification was performed through machine learning algorithms. Tests reached a maximum Correct Classification Rate (CCR) of 93.75%, when performed with the Radial Basis Function Network algorithm. The largest area under the ROC curve (AUC), 0.964, was achieved with the Multilayer Perceptron algorithm.