Vascular Analysis Based on Tumor Contour for 3-D Power Doppler Breast Ultrasound

• Main Authors: Yan-hao Huang, 黃彥皓
• Other Authors: Ruey-feng Chang
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/22455669296793252534

碩士 === 國立中正大學 === 資訊工程所 === 95 === Since the Doppler ultrasound (US) is successfully applied for detecting the blood flow, the studies of tumor vascularity have played leading roles to diagnose diseases of breast recently. The tumor vascularity is critical for growth, invasion, and metastasis. In general, malignant tumors need more complex network of blood vessels to obtain sufficient nutrients for growing. In the past, researches about vascularity just count the number of vascular pixel or voxel to analyze the condition of tumor. However, there are some vascular characteristics, for example morphology and tortuosity, which can be employed to provide more important diagnosis information. In this paper, we demonstrate a computer-aided diagnostic (CAD) system for 3-D power Doppler breast US that can quantify vascular morphology and tortuosity inside and outside tumor to differentiate malignant and benign tumors. At first, the skeleton of blood vessels is extracted, and then ten morphological features and three tortuous features are calculated from the skeleton for diagnosis. Furthermore, because tumor vessels provide the passage for tumor infiltration and metastasis and the vessels of malignant tumor often penetrate into the tumor for direct providing nutrients, whether the vessels are inside or outside the tumor could be an important diagnosis characteristic. In order to add the vessel location information, the tumor should be segmented at first. After the extraction of tumor contour, the morphological and tortuous features could be calculated for the vessels inside and outside the tumor respectively. Finally, vascular features inside and outside the tumor are used as the inputs of multi-layered perceptron neural network to classify the tumors. Investigations into 221 solid breast tumors include 110 benign and 111 malignant cases. If the tumor contour information is not used to extract the vascular features, the accuracy of neural network using these features is 77.83% (172/221), the sensitivity is 56.76% (63/111), the specificity is 99.09% (109/110), the positive predictive value is 98.44% (63/64), and the negative predictive value is 69.43% (109/157). Moreover, if the vascular features are obtained by including the tumor contour information, the accuracy ratio of the neural network using 14 proposed features is 87.78% (194/221), the sensitivity is 81.08% (90/111), the specificity is 94.55% (104/110), the positive predictive value is 93.75% (90/96), and the negative predictive value is 83.20% (104/125).