| Summary: | 碩士 === 國立臺灣大學 === 資訊工程學研究所 === 98 === The breast cancer is always the main causes of death for women. In recent years, the sonoelastography has been used to measure the tumor strain. In the sonoelastography, the physicians need to lightly compress a tumor to obtain a dynamic elastographic image sequence. According to the displacement of the tumor, the tumor strain will be obtained on sonoelastography video. Finally, the physicians will choose the representative slice from the dynamic elastographic image sequence to diagnose the tumor. The purpose of this study is to use image quantification method to automatically choose a representative slice, and automatically segment the tumor contour to evaluate the features to diagnose the tumor. First, according to the uniformity inside the tumor (the signal to noise ratio, SNRe) or the contrast of the tumor and the surrounding normal tissue (contrast to noise ratio, CNRe), the two kinds of quality quantification methods will be used to select the representative slice. Then, the level set method is used to segment the tumor contour. Finally, the B-mode and elastography features by the tumor contour are extracted for diagnosis. Furthermore, the two kinds of features are combined to diagnose the tumors to improve the performance. In this study, 151 biopsy-proved sonoelastography composed of 89 benign and 62 malignant masses are used to evaluate the performance of the quantification methods and the representative slices selected by the proposed methods will be compared to the physician-selected slice. In the experiment result, as using elastography features, the diagnosis performance of accuracy is 82.12% (124/151) on representative slice of CNRe, 82.12% (124/151) on representative slice of SNRe, 82.78% (125/151) on the physician-selected slice; as using B-mode features, the diagnosis performance of accuracy is 80.79% (122/151) on representative slice of CNRe, 87.42% (132/151) on representative slice of SNRe, 84.11% (127/151) on the physician-selected slice; as combining the B-mode and elastography features, the diagnosis performance of accuracy is 86.09% (130/151) on representative slice of CNRe, 90.07% (136/151) on representative slice of SNRe, 89.40% (135/151) on the physician-selected slice. Therefore, the representative slice selected by SNRe and CNRe colud replace the physician-selected slice to reduce the physician’s load, and combining the B-mode and elastography features will increase the diagnosis performance.
|
|---|
