Classification of mammographic image based on texture features with Random Forest method for identification of breast tumors

Abstract

Breast cancer is one of the most common sorts of cancer in ladies around the world. One method that can be used to detect breast cancer is to use medical imaging. Purpose: This study was conducted to identify the types of breast tumors by combining feature extraction and classification using the Random Forest method. The research data comes from the DDSM repository, which consists of 20 benign and 20 malignant images aged 40 to 50 years. The research stages consist of preprocessing, feature extraction, and classification. The preprocessing and feature extraction stages use MATLAB R2021a, while the classification process uses the WEKA application. Texture feature extraction methods include Histograms and GLCM (Gray-Level Co-Occurrence Matrix). The histogram feature extraction results show that the benign image has a higher level of brightness and contrast and is symmetrical compared to the malignant image. Meanwhile, the malignant image has a more random or irregular histogram than the benign image. Then, the average value of GLCM texture features resulting from benign images is higher than malignant images. The texture feature-based breast tumor classification process using the Random Forest method from the Training Set stage obtained an accuracy of 100%. Meanwhile, at the cross-validation stage with variations of 5-Folds, 10-Folds, and 15-Folds, the same value was obtained for an accuracy of 95%. This shows that the Random Forest classification method can be used to identify breast tumor types with more accurate results and does not depend on an individual's ability to read medical imaging results.