Breast cancer diagnosis using evolving deep convolutional neural network based on hybrid extreme learning machine technique and improved chimp optimization algorithm

Abstract

Today, diagnostic systems based on artificial intelligence play a significant role in confirming doctors' recommendations. These systems are becoming effective tools in clinical treatment. In this paper, we propose a new method for identifying atypical breast cancer based on the ZFNet network for breast mammography images. Initially, the Wiener and CALHE filters are used in order to evaluate the effectiveness of the preprocessing step. Following this, a pre-trained ZFNet was modified and trained on the CBIS-DDSM dataset. Furthermore, an extreme learning machine (ELM) was used to replace the remaining few layers at the very end. Moreover, a method was presented for estimating the optimum number of layers that should be replaced in the structure. In the end, ELM was developed in order to enhance the classification performance. This was accomplished by utilizing an improved version of the Chimp Optimization Algorithm (called SWChOA), along with four other benchmark meta-heuristic optimization algorithms. These algorithms were WSO, COA, AVOA, and AHA. Accuracy, precision, specificity, sensitivity, Matthew's correlation coefficient (MCC), and F1-score were the six popular metrics that were used to investigate the diagnostic performance of our method and to compare it to the performance of other well-known methods. Precision-recall curves and area of interest (ROI) curves were also utilized in this investigation. The approach known as ZFNet-SWChOA-ELM was shown to have the highest level of performance based on the findings of investigations that made use of a 10 hold-out validation.