Investigations on Statistical Classification Methods for Use in Breast Thermography

摘要

Recent studies have stated that thermography has shown to be very promising as an auxiliary tool in the task of early detection of breast cancer, which is a fundamental factor to increase the chances of cure of the patient. The present work aims to analyze methods of images classification of breast thermographic and to evaluate the results obtained with the purpose of investigating the feasibility of the use of infrared thermography for the detection of breast cancer. Initially, the thermographic image is obtained and processed. Then, the feature extraction occurs that is based on the temperature ranges obtained from the thermogram, determining the input data for the classification process. Eight statistical classifiers were evaluated. Finally, 93.42% of accuracy, 94.73% of sensitivity and 92.10% of specificity were obtained for the Cancer class in a binary analysis (Cancer versus Non-cancer) and in a multiclass analysis (Malignant, Benign, Cyst and Normal), 63.46% of accuracy, 80.77% of sensitivity and 86.54% of specificity were obtained for the Malignant class.