Optimal breast tumor diagnosis using discrete wavelet transform and deep belief network based on improved sunflower optimization method. (July 2020)

• Record Type: Journal Article
• Authors: Shen, Lizhen; He, Mingfang; Shen, Ning; Yousefi, Nasser; Wang, Chao; Liu, Guoqing
• Is Part Of: Biomedical signal processing and control. Volume 60(2020)
• Publication Date: 2020-07
• Subjects: Diagnosis -- Breast cancer -- GLCM -- DWT -- DBN -- Sunflower optimization method -- Improved; 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.2020.101953
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: New algorithm is proposed for automatic diagnosis of breast cancer. Deep belief network (DBN) is adopted for the classification. New improved optimization algorithm is proposed. The optimized algorithm is used for optimize the DBN weights. Abstract: Breast cancer is one of the most widespread types of cancer among women, but it does not necessarily mean pre-death, such that timely diagnosis of it can make the patient get to survive. Due to the significance of breast cancer, early diagnosis of abnormal areas in breast helps to cure this cancer in the initial steps. This study presents a new computer-aided diagnosis system for the early detection of breast cancer. The proposed method contains five important stages including noise reduction, image segmentation, mathematical morphology, feature extraction based on the combination of discrete wavelet decomposition and GLCM, and finally classification based on Deep Belief Network (DBN). To improve the DBN efficiency, it is optimized by an enhanced version of the sunflower optimization algorithm. Simulation results are applied to the MIAS database and the achievements have been compared with three different methods. Simulation results showed that the rate of accuracy, specificity, and sensitivity for the proposed model are achieved 91.5%, 72.4%, and 94.1%, respectively for the MIAS benchmark which gives better achievements toward the previous methods.