DCE-MRI Analysis for Breast Tumor Biomarker

• Main Authors: Cheng-Yang Chen, 陳政揚
• Other Authors: 張瑞峰
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/96328191434838426645

碩士 === 國立臺灣大學 === 資訊網路與多媒體研究所 === 102 === Breast cancer has become the women leading causes of the death in recent years. Recently, different therapies are used to cure the breast cancer and the doctor will decide the treatment plan according to molecular biomarkers. There are several biomarkers which can be used in prognosis or predictive, such as estrogen receptor (ER), progesterone receptor (PR), or human epidermal growth factor receptor 2 (Her2). In this paper, the main purpose is to use the features extracted from the dynamic contrast-enhance MRI (DCE-MRI) to predict the biomarkers. There are two types of biomarkers using in this paper as targets to classify the tumor which is ER and triple negative (ER-/PR-/Her2-). DCE-MRI is a method which records the signal intensity of the tumor by using the contrast agent. In the proposed DCE-MRI computer-aided classification system, the tumor is indicated by user and the tumor is segmented by a region growing based algorithm. After segmenting the tumor, four categories of features are used to improve the classification performance, including region features, texture features, shape features, and kinetic curve analysis. The region features are used to quantify the heterogeneity and randomness of the tumor. The shape features including compactness, margin, and ellipsoid fitting model are used to quantify the 3 dimensions (3-D) shape information of the tumor, and the texture features based on the grey level co-occurrence matrix are also used to quantify 3-D texture information of the tumor. At last, after using fuzzy c-means clustering to find the representative kinetic curve of the tumor, the representative kinetic curve is used in the kinetic curve analysis to quantify temporal features. In the experiment of classification of ER tumors, 71 biopsy-proved tumors with 48 ER positive tumors and 23 ER negative tumors are used to evaluate the performance of the proposed classification system. Its accuracy, sensitivity, specificity, and Az value are up to 83.10% (59/71), 83.33% (40/48), 82.61% (19/23), and 0.7651. In the second experiment of classification of triple negative tumors, 65 biopsy-proved tumors with 55 triple negative tumors and 10 non-triple negative tumors are used to evaluate the performance of the proposed classification system. Its accuracy, sensitivity, specificity, and Az value are up to 83.08% (54/65), 80.00 (8/10), 83.64 (46/55), and 0.8565.