Identification of Breast Abnormality from Thermograms Based on Fractal Geometry Features

Abstract

Breast cancer is a common cause of mortality among women globally. Thermography is a prognostic modality currently under research that has shown potential of providing early information on developing breast cancer based on temperature changes. In this work, we have extracted fractal texture features, namely Hurst coefficient, fractal dimension and lacunarity, from the segmented breast region for abnormality identification. These features are fed to classification algorithms like support vector machine (SVM), logistic regression, k-nearest neighbours (KNN) and Naïve Bayes to classify thermograms into two classes—healthy and sick. The best accuracy obtained for identification of breast abnormality using fractal features is 94.53% (Naïve Bayes classifier) as compared to accuracy of 92.74% obtained using texture and statistical features (PCA-SVM classifier) in our previous study. This further opens the scope for thermography to be used clinically for breast examination on a large scale.