Sonography Images for Breast Cancer Texture Classification in Diagnosis of Malignant or Benign Tumors

This work aims at selecting useful features in critical angles and distances by Gray Level Co-occurrence Matrix (GLCM). In this project, images were labeled based on physician opinion in two groups (malignant or benign). These labeled images were used in classification analysis. Images were opened and read in Matlab software. The tumors were cropped in rectangular shape manually; then graycomatrix and GLCM have been calculated in 4 angels (0, 45, 90 and 135 degree) and 4 distances (1, 2, 3 and 4) for cropped tumor images. Since each angle and distance pair include 22 features, each image had 352 final features (22 features * 4 angles * 4 distances =352). At the final step, features were classified using Kmeans method into 2 classes of malignant and benign; then the confusion matrix was made and qualitative comparison was used to select important features and critical distances and angles in each one. Some special features, angels and distances which had the best classification result and high percentages of accuracy were selected as useful features. These finding suggested that texture parameters can be useful to help in distinguishing between malignant and benign breast tumors.