Textural and Shape Features for Lesion Classification in Mammogram Analysis

Abstract

The efficient analysis of digital mammograms has an important role in the early detection of breast cancer and can lead to a higher percentage of recovery. The process of mammogram classification can be divided into two steps as follows: first, it has to be established if the tissue contains abnormalities, and, second, the nature of the lesion has to be determined. This second step of a computer-aided diagnosis system is important in order to select the best treatment for the patient and to achieve the highest chance of recovery. In general, digital mammogram analysis consists of preprocessing, feature extraction, feature selection and classification. Feature extraction is crucial in identifying informative characteristics that can differentiate between benign and malignant lesions. The two main types of feature extraction methods are shape features and texture features. In the current paper, we present several experiments in order to compare the performance of different feature extraction methods from the two types mentioned previously. As data, images from the Digital Database for Screening Mammography (DDSM) are used, which has precise ground truth for the cancerous tissue. For classification Decision Trees and Random Forest methods are used to evaluate the performance using the different extracted features. The experiments that were carried out show that shape features perform better than texture features to separate benign and malignant abnormalities. Also, some outliers were found causing a decrease in the accuracy of the system and achieving 66% test accuracy using shape features and Random Forest classifier.