breast mass detection with nonlinear model of thermography images

Detecting a breast mass is a common and stressful event for women. although most breast masses are benign, the risk of malignancy highlights the importance of appropriate screening. different imaging methods have different precisions and accuracies, so choosing an appropriate imaging method, especially for women with dense breast tissue, is very important. since vascular structure regional temperatures differ between normal and abnormal tissues, thermography can detect masses earlier than conventional imaging methods.237 cases including 152 healthy individuals and 85 cases with breast masses examined in this study. the raw recorded images of these cases are gray-levels which are given to a nonlinear transform to become colorful which increase the thermal contrast. then these color scaled images are given to convolutional neural networks. the used networks in this research is alexnet and googlenet. the extracted features are given to different classifiers as input. the used classifiers in this study are knn, svm and nb. the best result was achieved when googlenet and svm were used together. the results of this study have remarkable accuracy and sensitivity which are 95.8% and 100%, respectively. the developed system combining nonlinear color scaling and deep learning shows potential as an effective tool for early breast screening.