Automatic breast thermography images classification based on deep neural networks

Breast thermography is a screening tool which is capable of detecting cancer at an early stage. The main objective of this work is using the full power of deep neural network (DNN) and exploring its ability to learn the discriminative features of input data. The transfer learning and data augmentation are performed to solve the problem of lack of labled data. To improve the accuracy, the support vector machine (SVM) classifier will hybrid with the convolutional neural network (CNN) instead of using the deep model as endtoend. The performance is verified by the kfold crossvalidation. The proposed techniques are trained and evaluated on DMRIR dataset to classify the thermographic images to normal and abnormal groups. The proposed technique of employing AlexNet hybrid with SVM achieves the best performance, producing 92.55% accuracy, 95.56% sensitivity, 89.80% precision, 92.63% F1 score.