Combining the advantages of radiomic features based feature extraction and hyper parameters tuned RERNN using LOA for breast cancer classification

Abstract

This paper proposes Combining the Advantages of Radiomic features based Feature Extraction and Hyper Parameters tuned Recalling Enhanced Recurrent Neural Network (RERNN) using Lizard optimization Algorithm (LOA) for Breast cancer Classification. Here, breast cancer images are taken from the real time dataset collected from VPS hospital and then the images are preprocessed using Altered Phase Preserving Dynamic Range Compression (APPDRC) to remove the noises. Then the radiomic features, such as morphologic features, grayscale statistic features and Haralick texture features have been extracted utilizing Entropy Based Local Binary Pattern (ELBP). These extracted features have presented to Recalling Enhanced Recurrent Neural Network (RERNN) classifier. Hence, Lizard optimization Algorithm (LOA) is utilized to optimize the Recalling Enhanced Recurrent Neural Network (RERNN). The proposed approach is executed in MATLAB platform, then the performance is compared with different existing approaches. This approach is applicable in real time applications for screening the abnormalities of the breast cancer at initial stage, thus, determining the proper treatment to be given to the patient for decreasing deaths caused by breast cancer. The novelty or aim of this paper is to diagnose the breast cancer in early stage by extracting the radiomic features and to classify the types (Malignant, Benign, normal) of breast cancer with high accuracy by reducing the computational time and error rate. The simulation outcomes demonstrate that the proposed FE-APPDRC-ELBP-RERNN-LOA attains the accuracy of45.75%, 37.64%, 24.64 %is higher than the existing methods.