| Summary: | 碩士 === 國立中正大學 === 資訊工程所 === 94 === The correlation between tumor growth and angiogenesis is an important phenomenon that helps physicians to analyze the disease. Tumor angiogenesis has been widely studied in recent years and proved as a key role in tumor growth, invasion, and metastasis. In the past researches, only the vessel information is used to diagnose the tumor. However, the location of vessel related to the tumor, for example whether the vessel is inside the tumor or not, is proposed in this paper. In order to extract the new proposed features, the tumor regions should be segmented for capturing the information in and out tumor. This paper presents a computer-aided diagnostic (CAD) system that analyzes the relation between blood vessels and tumor using 3-D power Doppler ultrasound (US) for breast cancer. These 3-D power Doppler ultrasound datasets could be decoded into two kinds of sequential images, grey and vessel images. Grey images are applied by a fuzzy unit, a defuzzier unit, and connected component labeling techniques to determine tumor VOI (volume of interest). Vessel images are used to capture vessel blood voxels corresponding to the segmented tumor. This study of 3-D assessment of the correlation between blood vessels and tumor using the volumes of tumor and vessels, the ratio of vessel voxels and volume voxels, the ratio of vessel voxels inside tumor region, the ratio of vessel voxels outside tumor region, the ratio of vessel voxels around tumor region. Finally, these extracted features for all training datasets are put into the neural network and then the training process is computed. The trained neural network is then used for breast tumor diagnosis. For 223 solid breast rumors including 115 benign and 108 malignant cases, the accuracy of neural network using proposed features is 85.20 %.
|
|---|
