Data mining technique for breast cancer detection in thermograms using hybrid feature extraction strategy

Breast thermography is capable of detecting cancer at an early stage. In this work, we have used for analysis 50 thermograms (25 each of normal and abnormal). The main objective of this work is to evaluate the use of discrete wavelet transform (DWT) and texture features extracted from thermograms in classifying normal and abnormal groups. One clinically significant texture and two DWT features were fed to Decision Tree (DT), Fuzzy Sugeno, Naive Bayes Classifier, K Nearest Neighbour, Gaussian Mixture Model and Probabilistic Neural Network classifiers to evaluate the best classifier. Our results show that, DT and fuzzy classifiers yielded a highest average accuracy of 93.30%, sensitivity of 86.70% and specificity of 100%. The proposed computer-aided diagnostic system can be used for an automatic classification of normal and abnormal breast thermograms which can aid the radiologists in their diagnosis.