Deep Learning Based Anomaly Detection

Deep Learning Based Anomaly Detection

Magnetic resonance imaging (MRI) has been a prevalence technique for breast cancer diagnosis. This paper introduces a semi-supervised method for extracting breast tumors in a set of real MRIs of different types of breast cancer patients. We call the proposed method as Semi-supervised Tumor Segmentation (SSTS), and apply it to both mass and non-mass lesions. We have trained 225 classifiers with respect to different settings of threshold parameters that need to be set in SSTS. We will show the performance of SSTS for extracting the infiltrating ductal carcinoma (IDC) and the ductal carcinoma in situ (DCIS) tumors based on a set of real MRIs of 21 breast cancer patients; and how different settings of the parameters will influence the extraction results. We additionally implement five state-of-the-art intensity-based image segmentation algorithms that can be compared with SSTS on breast tumor extraction.