DCE-MRI and DWI Analysis for Breast Tumor Biomarker
Date Issued
2016
Date
2016
Author(s)
Cheng, Tsung-Chi
Abstract
There are several breast cancer therapies at this day. Doctors will design different treatments for patient according to the cause of breast cancer and molecular biomarkers. Several biomarkers are used in estimate the prognosis, such as estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). The main purpose of this study is to predict the biomarkers with features extracted from dynamic contrast-enhance MRI (DCE-MRI) and diffusion-weighted MRI (DWI). There were four types of biomarker, ER, PR, HER2, and triple negative (TN, ER-/PR-/HER2-), using in this paper as classification targets. The DCE-MRI uses contrast material to enhance particular organs, tissues, or tumors, and uses fast and continuous imaging to trace the variation of contrast enhancement. The DWI represents the different mobility of water molecules between normal tissue and diseased tissue. In the proposed computer-aided classification system, the tumor was indicated by a user-specified volume of interest (VOI) and segmented by the confident connected method. Then we applied a registration method to obtain the DWI tumor. After the tumor images of DCE-MRI and DWI were obtained, seven categories of features were extracted to improve the classification performance, including ADC features, region features, shape features, texture features, ranklet texture features, vascular features, and kinetic curve analysis. The ADC features used the apparent diffusion coefficient (ADC) to quantify the water diffusion within tissue. The region features were used to quantify the heterogeneity and randomness of the tumor. The shape features including compactness, margin, and ellipsoid fitting model were used to quantify the three dimensions (3-D) shape information of the tumor, and the texture features based on the grey level co-occurrence matrix were also used to quantify 3-D texture information of the tumor. The ranklet texture features were extracted after applying the ranklet transformation. The vascular features were used to extract the morphology features of vessel. At last, after using the fuzzy c-means clustering to find the representative kinetic curve of the tumor in DCE-MRI, the representative kinetic curve was used in the kinetic curve analysis to quantify temporal features. In the experiment of classification of ER tumors, 78 biopsy-proved tumors with 47 ER positive tumors and 31 ER negative tumors were used to evaluate the performance of the proposed classification system. Its accuracy, sensitivity, specificity, and Az value were up to 80.76% (63/78), 82.98% (39/47), 77.42% (24/31), and 0.8006. In the second experiment of classification of PR, 78 biopsy-proved tumors with 27 PR positive tumors and 51 PR negative tumors were used to evaluate the performance of the proposed classification system. Its accuracy, sensitivity, specificity, and Az value were up to 79.49% (62/78), 70.37% (19/27), 84.31% (43/51), and 0.7911. In the experiment of classification of HER2 tumors, 78 biopsy-proved tumors with 36 HER2 positive tumors and 42 HER2 negative tumors were used to evaluate the performance of the proposed classification system. Its accuracy, sensitivity, specificity, and Az value were up to 80.77% (63/78), 77.78% (28/36), 80.95% (34/42), and 0.8501. In the experiment of classification of TN tumors, 78 biopsy-proved tumors with 14 TN positive tumors and 64 non-TN tumors were used to evaluate the performance of the proposed classification system. Its accuracy, sensitivity, specificity, and Az value were up to 80.77% (63/78), 71.43% (10/14), 82.81% (53/64), and 0.7043.
Subjects
DCE-MRI
DWI
breast
estrogen receptor
progesterone receptor
human epidermal growth factor receptor 2
triple negative
apparent diffusion coefficient
heterogeneity
GLCM
ellipsoid
ranklet transformation
vessel
kinetic curve
SDGs
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-105-R02944032-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):0e29d568dc6c12526a039c1766555c53