Hybrid of Salp Swarm Optimization Algorithm and Grasshopper Optimization Algorithm (SSOAGOA) for Feature Selection

Abstract

Transrectal Ultrasound (TRUS) imaging is the appropriate methodology for early recognizable proof of prostate malignant growth. TURS image contains speckle noise, tracker, low dissimilarity and at the early stage differentiate between the normal and abnormal of prostate lesion is a difficult task. The misdiagnosis or late identification may direct unnecessary biopsies. The computational-based classification system gives a second assessment of the radiologists. Objects are depicted as a set of measurable features in pattern recognition. The feature selection method selects the important features from the original features set to reduce the data visualization and improve the efficiency of classification with high accuracy. Using Ant Colony Optimization (ACO) method, the ultrasound images are enhanced, and the identified region of interest (ROI) with Ant Colony Optimization- Boundary Complete Recurrent Neural Network (ACO-BCRNN). Then, from ROI, features are extracted by using VGG-19 transfer learning techniques. The study focused on an optimization-based method of feature selection is undergone such as Salp Swarm Optimization Algorithm (SSOA), Grasshopper Optimization Algorithm (GOA), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Wolf Optimization Algorithm (WOA), and Ant Colony Optimization (ACO). These algorithms may produces outcome with highly essential information loss. To address the above issue, the novel automated detection system is required  for identifying the prostate cancer. Therefore in this article, the hybridization of the Salp Swarm Optimization Algorithm and Grasshopper Optimization Algorithm (SSOAGOA) is proposed as a feature selection method to select relevant and essential features for the discrimination of prostate cancer with high efficiency and accuracy. The selected features are used for classifying prostate cancer on TRUS images. The SSOAGOA method performance is assessed by using evaluation parameters through Support Vector Machine (SVM), Grid Search (GS), and Extreme Machine Learning (ELM). The comparative analysis of various feature selection methods such as SSOAPSO, SSOAGA, SSOAACO, and SSOAWOA are performed. The results indicate that the proposed method earned highest accuracy with all three classifier as 97.21%, 99%, and 100% respectively and comparatively superior to others.