Experimental Investigation of Classification Algorithms for Predicting Lesion Type on Breast DCE-MR Images

Abstract

Timely revealing of breast cancer is one of the most important issues in determining prognosis for women with malignant tumors. Dynamic contrast-enhanced (DCE) MRI is being increasingly used in the clinical setting to help detect and characterise tissue, suspicious for malignancy and has been shown to be the most sensitive modality for screening highrisk women. Computer-assisted evaluation (CAE) systems have the potential to assist radiologists in the early detection of cancer. A crucial module of the development of such a CAE system will be the selection of an appropriate classification function responsible for separating malignant and benign lesions. The motivation of this paper is to provide qualitative evaluation of three advanced classifiers like artificial neural network, support vector machine and artificial bee colony optimization algorithm trained neural network are being developed for classification of the suspicious lesions in breast MRI. A comparative study of these techniques for lesion classification is made to identify relative merits. As a result, the paper concluded that the neural network trained by artificial bee colony optimization algorithm based classifier outperforms all other explored classifiers for the examined dataset of breast DCE -MR images.

Hybrid ANN optimized artificial fish swarm algorithm based classifier for classification of suspicious lesions in breast DCE-MRI

“…In clinical diagnosis, the signs of abnormality observed by expert radiologists are very diverse. They include the size, contrast, intensity and density [1,[3][4][5][6][7][8][9][10][11][12][13][14][15].…”

“…Significant research on breast DCE-MRI lesion classification methods have already been made to automatically predict lesions such as artificial neural networks, linear discriminant analysis, logistic regression and support vector machines [1,3]. The artificial neural networks based classifier have been one of the most popular approaches for investigating the classification of malignant and benign breast DCE-MR lesions [1,[3][4][5][6][7][8][9][10][11][12][13][14][15]. The performance of any classifier depends on type of features used, training dataset provided and the type of classifier.…”

“…Texture feature has been proven to be useful in differentiating normal and abnormal pattern. For breast DCE-MRI suspicious lesion classification statistical texture features are widely used [3,4,[8][9][10][11][12][13][17][18][19][20].…”

“…In this proposed work, a set of 18 features were extracted from the ROI, 14 GLCM (gray level co-occurrence matrix) statistical texture measures as defined in [21] and 4 GLHM (gray level histogram moments) features defined in [3,11]. These features are used because they have proven improvement in classifier performance [3,11].…”

“…These features are used because they have proven improvement in classifier performance [3,11]. The features extracted are: entropy, skewness, difference moment, energy measure, difference variance, sum average, standard deviation, difference average, difference entropy, kurtosis, correlation, inertia, variance, inverse difference moment, mean, sum entropy, sum variance and information measure of correlation.…”