Advanced Medical Imaging Analytics in Breast Cancer Diagnosis

Abstract

Modern imaging technique provides a fast, noninvasive means to study physiologic, metabolic, and molecular processes in the body. Imaging is the primary means in clinical cancer practice to facilitate diagnosis, prognosis, and treatment evaluation. While breast cancer contributes to 25% of morbidity in all cancer and is the second leading cause of cancer death in women, it is also one of the most treatable malignancies if detected early. In this chapter, we present an overview of research using advanced imaging analytics tools on Digital Mammography (DM) to improve the sensitivity and specificity of breast cancer detection. Currently, there are two dominating trends in the advanced imaging analytics field: texture analysis and deep learning. We implement three texture analysis algorithms: Gray Level Co-Occurrence Matrix (GLCM), Local binary patterns (LBP), and Gabor Filter and one deep learning network: ResNet. Gradient Boosted Tree (GBT) classifier is then developed on the features to diagnose the lesion as malignant vs. benign. The classifier using texture features from each texture analysis algorithm has an accuracy of 0.82, 0.72, and 0.72 for GLCM, LBP, and Gabor, respectively. If the texture features from different texture analysis algorithms are pooled together, the classifier has an accuracy of 0.81. The same classifier using features extracted from ResNet has an accuracy of 0.89 indicating the potentials of deep learning in medical imaging for disease diagnosis.