Improving early prostate cancer diagnosis by using Artificial Neural Networks and Deep Learning

Prostate cancer could be diagnosed by routine controls such as biopsy. But considering prostate biopsy side effects, using automated tools along with some selected features in early diagnosis of this cancer seems necessary. Even though production of this tool previously has been done, but the importance of the issue binds us to increase its accuracy as much as possible. Using Deep Learning to enhance medical diagnosis is an important matter in areas of research. Deep Learning Artificial Neural Networks are classification algorithms that can be used for classification. In this movement, we are going to improve existing classifier based expert system for early diagnosis of the organ to attain informed decision without biopsy by using some definite features. 50 data used in this paper are collected from Imam Reza hospital (Tehran). Classifying training input data, we have used following classifiers: Scaled conjugate gradient (SCG), Broyden-Fletcher-Goldfarb-Shanno (BFGS), and Levenberg- Marquardt (LM) training algorithms of Artificial Neural Networks (ANN); and AlexNet which is one of the CNNbased methods of Deep Learning. The proposed system was designed based on AlexNet function which had the best performance among existing methods. In fact, this paper is going to state how deep learning could be used for early diagnosis of cancer and Deep Learning advantages of SVM in cancer diagnosis as well. In the end, the predictive accuracy of the mentioned method of Deep Learning has been compared with that of gained by use of SVM and ANN. Deep Learning achieved classification accuracy is 86.3%, while for SVM was 81.1% and for ANN 79.3%. But sensitivity and specificity didn’t have considerable changes.