Breast ultrasound images classification using morphometric parameters ordered by mutual information

Abstract

This work aims to assess the potentiality of Mutual Information (MI) in ordering morphometric parameters, according to its relevance, to classify breast ultrasound images. Seven parameters were calculated over normalised radial length and convex polygons from 246 segmented tumour images. MI was calculated between each one and the outcome, and between each other. MI indicated that two parameters had negligible relevance. A feedforward neural network was implemented with the seven parameters as input to classify the images. The best performance (accuracy=88%) was obtained with the five first parameters, thus confirming the pour relevance of the last two ones.