Using Forest Optimization Algorithm for Feature Selection on Mammograms

• Main Author: 賴仁宏
• Other Authors: 葉進儀
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/968dc2

碩士 === 國立嘉義大學 === 資訊管理學系研究所 === 106 === Breast cancer ranks first in the incidence of women cancer in Taiwan, with more than 10,000 women suffering from breast cancer every year. Nearly 2,000 women die due to breast cancer, which is equivalent to six women who lose their precious lives every day because of breast cancer, which pose a threat to women's health. Therefore, in addition to promoting breast self-examination and correct eating habits, the best way is early diagnosis, early detection and early treatment, the current early diagnosis of mammography is viewed as the most effective way to reduce patient mortality and extend the life of patients. The survival rate of the first stage of breast cancer after treatment survival rate is as high as 97.5%. In mammography, the diagnosis of benign and malignant tumors is a challenging task for experts. In this paper, the Mammographic Image Analysis Society (MIAS) and The Digital Database for Screening Mammography (DDSM) databases were used for testing in reality to classify benign and malignant breast tumors. The steps of this study include Region of Interest (ROI) extraction, feature extraction, feature selection, classification, and performance appraisal. This study calculated 270 features and applied Forest Optimization Algorithm (FOA) for feature selection. Four heuristic algorithms were added to compare performance and supported vector machines (SVM) were used for classification. Through experiments, FOA is compared with the other heuristic algorithms and traditional feature selection methods. Experimental results show that FOA is superior to the other algorithms in performance.