Analysis of Transform-Based Features on Lateral View Breast Thermograms

Abstract

Breast cancer is the most serious type of diseases; hence, it requires an efficient diagnosis for an early cure. Thermography for breast screening has achieved breakthroughs in detecting the abnormalities at earlier stages. In this paper, breast thermograms acquired in lateral views are considered for applying transform-based feature extraction. It is necessary to transform the acquired thermograms into the frequency domain to pick out the detailed feature that delineates abnormal from the normal ones. This paper categorizes the features under three major types of transforms namely wavelet, curvelet and contourlet transform. The transformed thermograms are then subjected to feature extraction where statistical values are measured and compared with spatial domain gray level co-occurrence matrix values. By analyzing the features using ANOVA and independent t test, the result significantly shows that the lateral view thermograms outperform than the frontal views. Finally, the features are used to classify the normal and abnormal thermograms of frontal and lateral views using pattern classifiers. According to the results, the classification performance is measured with the highest accuracy of 91%, 87% of sensitivity rate and 90% of specificity rate using the AdaBoost algorithm.