Cancer detection using convolutional neural network optimized by multistrategy artificial electric field algorithm

Abstract

AbstractRecently, image processing schemes are widely used to improve disease detection performance in many medicinal fields. Cancer is considered as one of the most deadly disease and early diagnosis of cancer is the complicated task in the field of medicine. In this paper, we present the two pretrained convolutional neural network (CNN) based on ensemble models such as VGG19 and VGG16 for cancer diagnosis that classifies both normal and abnormal images. The dilemma associated with CNN hyperparameter tuning complicates while diagnosing cancer. Hence, we propose multistrategy based artificial electric field (M‐AEF) algorithm for hyper‐parameter tuning in CNN thereby finding the optimal values. The exponentially decaying learning rates are more helpful to train CNN and prevent it from a local minimum. Thus, random minority over‐sampling and random majority under‐sampling address the imbalanced issue present in the dataset. The images are obtained from three different datasets namely the Kaggle dataset, International Collaboration on Cancer Reporting (ICCR) dataset, and cancer programming dataset for cancer detection. The experimental results are executed in MATLAB software and various performance analyses are carried out. Finally, the proposed method demonstrated better and higher cancer detection performance than other methods.