INVESTIGATION OF CLASSIFIERS FOR EARLY-STAGE BREAST CANCER BASED ON RADAR TARGET SIGNATURES
INVESTIGATION OF CLASSIFIERS FOR EARLY-STAGE BREAST CANCER BASED ON RADAR TARGET SIGNATURES
复制标题
DOI：
10.2528/pier10051904
复制
发表时间：
2010-01-01
影响因子：
6.7
通讯作者：
Glavin, M.
中科院分区：
文献类型：
--
作者：
Conceicao, R. C.;O'Halloran, M.;Glavin, M.
超宽带（UWB）雷达作为检测早期乳腺癌的一种手段已经得到了广泛的研究。这种成像模式的基础是微波频率下正常和癌性乳腺组织之间的介电对比。然而，基于乳腺内恶性肿瘤和良性肿瘤之间的介电相似性，在微波图像中区分这些类型的组织可能是有问题的。因此，重要的是要研究替代方法来分析和分类介电散射在乳房内，考虑到其他肿瘤的特性，如形状和表面纹理的肿瘤。良性肿瘤往往具有光滑的表面和椭圆形，而恶性肿瘤往往具有粗糙和复杂的表面，具有针状或小叶。因此，一种分类方法是基于它们的雷达目标特征（RTS）对散射体进行分类，所述RTS携带关于散射体尺寸和形状的重要信息。在本文中，高斯随机球（GRS）被用来模拟良性和恶性肿瘤的形状和大小。主成分分析（PCA）用于从肿瘤的RTS中提取信息，而八种不同的肿瘤分类器组合在性能方面进行了分析，并在两种可能的方法方面进行了比较：线性判别分析（LDA）和二次判别分析（QDA）。
Ultra Wideband (UWB) radar has been extensively investigated as a means of detecting early-stage breast cancer. The basis for this imaging modality is the dielectric contrast between normal and cancerous breast tissue at microwave frequencies. However, based on the dielectric similarities between a malignant and a benign tumour within the breast, differentiating between these types of tissues in microwave images may be problematic. Therefore, it is important to investigate alternative methods to analyse and classify dielectric scatterers within the breast, taking into account other tumour characteristics such as shape and surface texture of tumours. Benign tumours tend to have smooth surfaces and oval shapes whereas malignant tumours tend to have rough and complex surfaces with spicules or microlobules. Consequently, one classification approach is to classify scatterers based on their Radar Target Signature (RTS), which carries important information about scatterer size and shape. In this paper, Gaussian Random Spheres (GRS) are used to model the shape and size of benign and malignant tumours. Principal Components Analysis (PCA) is used to extract information from the RTS of the tumours, while eight different combinations of tumour classifiers are analysed in terms of performance and are compared in terms of two possible approaches: Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA).