FGMM-based Computer Aided Diagnosis to Classify Benign/Malignant Breast Mammogram Images
Samaneh Aminikhanghahi · Open PRAIRIE (South Dakota State University) · 2014
Mammography is potentially a convenient screening method to be comfortable and effective in early detection of breast cancer, but its interpretation is difficult due to low quality images and the limitations of physicians’ experience. Recently, Computer Aided Diagnosis (CADx) in digital mammograms intends to provide assistance to physicians in image interpretation and increase breast cancer diagnosis accuracy. The main objective of this thesis is to enhance and introduce a new CADx system to classify the detected regions in wireless mammogram images into malignant or benign categories. In this approach, different sets of morphological and textural features were investigated and a correlation based feature selection algorithm was used to reduce the CADx system size and complexity. Also, after performing a comparative study to investigate the accuracy of different popular classifier method such as Support Vector Machine (SVM), Gaussian Mixture Model (GMM), and Fuzzy Logic System (FLS), a new classification method called Fuzzy Gaussian Mixture Model (FGMM) by combining GMM and FLS is evaluated and introduced. The experimental results are obtained from a data set of 300 images taken from the Digital Database for Screening Mammography (DDSM, University of South Florida) for different classes. Confusion matrix analysis is used to measure the performance of the CADx system. The results show that the proposed FGMM classifier has achieved an overall classification quality of 86.16%, with 93% accuracy, 90% sensitivity and 96% specificity, and outperformed other classifiers in all aspects. The experimental results obtained from the developed CADx in this research prove that the proposed technique will improve the diagnostic accuracy and reliability of radiologists’ image interpretation in the diagnosis of breast cancer. The resulting breast cancer CADx detection system is a promising tool to provide preliminary decision support information to physicians for further diagnosis.