Traditional machine learning algorithms for breast cancer image classification with optimized deep features. (March 2023)

• Record Type: Journal Article
• Authors: Atban, Furkan; Ekinci, Ekin; Garip, Zeynep
• Is Part Of: Biomedical signal processing and control. Volume 81(2023)
• Publication Date: 2023-03
• Subjects: Histopathological image classification -- Transfer learning -- Meta-heuristic algorithms -- Machine learning; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104534
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: For breast cancer diagnosis, computer-aided classification of histopathological images is of critical importance for correct and early diagnosis. Transfer learning approaches for feature extraction have made significant progress in recent years and are now widely used. To select the best representative features to classify breast cancer pathological images and avoid the curse of dimensionality, this work uses optimized deep features. The proposed approach firstly employs ResNet18 architecture for feature extraction to achieve deep features. Then, meta-heuristic algorithms namely Particle Swarm Optimization (PSO), Atom Search Optimization (ASO) and Equilibrium Optimizer (EO) algorithms, are employed to provide more representative features of breast cancer pathological images. To understand the effect of optimized deep features on classification, traditional machine learning (ML) algorithms are used. The experimental analysis of the proposed approach has been done on the public benchmark dataset BreakHis. Experimental results illustrate that, for features obtained from ResNet18-EO, the proposed approach achieves a 97.75% F-score by using the Support Vector Machine (SVM) with gaussian and radial-based functions (RBF). Highlights: An optimized deep feature extraction and selection method is proposed. The proposed model can deeply learn useful information from the histopathological images. ResNet18 extracted deep features are optimized with PSO, ASO and EO. BreakHis is used to show the generalization ability of meta-heuristic algorithms. Results demonstrate the performance advantages of metaheuristic algorithms.