乳腺辅助诊断系统中可疑肿块分割方法研究

乳腺癌是女性最常见的恶性肿瘤之一,早发现早治疗是防治的关键。而计算机辅助诊断技术能够有效地对具有乳腺癌重要特征的可疑肿块进行分割、检测和分类,从而提高影像医生的诊断效率和准确率。综述了目前出现的一些较好的乳腺可疑肿块分割方法,对这些算法进行了深入的研究,对它们的优点和性能进行了对比和分析。最后展望了基于肿块分割方法可能提高精准度的一些途径。     

基于特性模型与神经网络的乳腺图像肿块自动检测技术

钼靶X线摄影是最常用的乳腺癌早期诊断手段。该文针对乳腺图像中的肿块提出了一种基于特性模型与神经网络的计算机辅助诊断技术。它首先建立两种特性模型分别描述脂肪组织和腺体组织中的肿块;然后对脂肪中的肿块采用迭代阈值法进行检测,对腺体中的肿块采用小波域黑洞检索法进行标记;接着采用一种基于Canny算子和能量场约束以及ANFIS控制的填充膨胀方法分割疑似肿块;最后使用一种MLP分类器剔除假阳性。实验结果表明,该算法在面对特性迥异的多种肿块时可取得较高的检测精度,并保证较低的假阳性率。     

Automatic mass segmentation on mammograms combining random walks and active contour

Accurate mass segmentation on mammograms is a critical step in computer-aided diagnosis (CAD) systems. It is also a challenging task since some of the mass lesions are embedded in normal tissues and possess poor contrast or ambiguous margins. Besides, the shapes and densities of masses in mammograms are various. In this paper, a hybrid method combining a random walks algorithm and Chan-Vese (CV) active contour is proposed for automatic mass segmentation on mammograms. The data set used in this study consists of 1095 mass regions of interest (ROIs). First, the original ROI is preprocessed to suppress noise and surrounding tissues. Based on the preprocessed ROI, a set of seed points is generated for initial random walks segmentation. Afterward, an initial contour of mass and two probability matrices are produced by the initial random walks segmentation. These two probability matrices are used to modify the energy function of the CV model for prevention of contour leaking. Lastly, the final segmentation result is derived by the modified CV model, during which the probability matrices are updated by inserting several rounds of random walks. The proposed method is tested and compared with other four methods. The segmentation results are evaluated based on four evaluation metrics. Experimental results indicate that the proposed method produces more accurate mass segmentation results than the other four methods.     

Statistical analysis of mammographic features and its classification using support vector machine

This study aims at designing a support vector machine (SVM)-based classifier for breast cancer detection with higher degree of accuracy. It introduces a best possible training scheme of the features extracted from the mammogram, by first selecting the kernel function and then choosing a suitable training-test partition. Prior to classification, detailed statistical analysis viz., test of significance, density estimation have been performed for identifying discriminating power of the features in between malignant and benign classes. A comparative study has been performed in respect to diagnostic measures viz., confusion matrix, sensitivity and specificity. Here we have considered two data sets from UCI machine learning database having nine and ten dimensional feature spaces for classification. Furthermore, the overall classification accuracy obtained by using the proposed classification strategy is 99.385% for dataset-I and 93.726% for dataset-II, respectively.     

Improved watershed transform for tumor segmentation: Application to mammogram image compression

In this study, an automatic image segmentation method is proposed for the tumor segmentation from mammogram images by means of improved watershed transform using prior information. The segmented results of individual regions are then applied to perform a loss and lossless compression for the storage efficiency according to the importance of region data. These are mainly performed in two procedures, including region segmentation and region compression. In the first procedure, the canny edge detector is used to detect the edge between the background and breast. An improved watershed transform based on intrinsic prior information is then adopted to extract tumor boundary. Finally, the mammograms are segmented into tumor, breast without tumor and background. In the second procedure, vector quantization (VQ) with competitive Hopfield neural network (CHNN) is applied on the three regions with different compression rates according to the importance of region data so as to simultaneously reserve important tumor features and reduce the size of mammograms for storage efficiency. Experimental results show that the proposed method gives promising results in the compression applications.     

Effect of the display medium of a mammogram on microcalcification—diagnostic performance and the subjective preference of radiologists according to differing professional experience

This study was nvestigate the effect of different display media of mammographic images (visual display terminals [VDT] and a view-box) on the display-media diagnostic performance and subjective prefereness of radiologists according to differing professional experience. This study included 30 patients and 120 images (four images per patient) who underwent digital mammography at Changhua Christian Hospital in central Taiwan, using the General Electric digital mammography system. Biopsies indicated that 15 patients had microcalcifications, while the other 15 patients were normal. In this experiment, the interpreting physicians included of five males and one female. The physicians were divided into three groups (attending physicians, radiology residents and interns) according to varying levels of experience and expertise in digital mammography. Three different display media were used, including a medical monitor (BARCO V9601100), a general monitor (SAMSUNG 191T), and an unmasked view-box. A signal-blind test based on these three different display media was used to evaluate the six radiologists' microcalcifications-diagnosis performance and to obtain the area-under-curve (AUC) values from the receiver-operating-characteristic (ROC) curves, sensitivity, and specificity. A five-point Likert-type rating scale was used to evaluate the radiologists' subjective preference. The results of the analyses of variances showed that different professional experience settings had a significant effect on all AUC, sensitivity, and specificity. Attending and resident physicians performed significantly better on AUC, sensitivity, and specificity than interns. Different display-type settings had a significant effect on AUC and sensitivity; however, they had no significant effect on specificity. The physicians performed significantly better on AUC when the display types were a BARCO medical LCD and a SAMSUNG general LCD rather than a conventional unmasked view-box. The physicians performed significantly better on sensitivity when the display type was a BARCO medical LCD rather than a SAMSUNG general LCD or a conventional unmasked view-box. Different professional experience provided significantly different preference evaluating (F(2, 16)=6.50, p<0.05), and different display type settings had a significant effect on physicians' diagnosis performance (F(2, 16)=138.5, p<0.05). The results of a Duncan test demonstrated that the physicians' most preferred display media was a BARCO medical LCD. The findings of this research indicate that physicians' better diagnosis performance depends on both their professional experience and a high-resolution medical-level display media to interpret digital mammographic microcalcifications.     

Comparison of multiwavelet, wavelet, Haralick, and shape features for microcalcification classification in mammograms

We present an evaluation and comparison of the performance of four different texture and shape feature extraction methods for classification of benign and malignant microcalcifications in mammograms. For 103 regions containing microcalcification clusters, texture and shape features were extracted using four approaches: conventional shape quantifiers; co-occurrence-based method of Haralick; wavelet transformations; and multi-wavelet transformations. For each set of features, most discriminating features and their optimal weights were found using real-valued and binary genetic algorithms (GA) utilizing a k-nearest-neighbor classifier and a malignancy criterion for generating ROC curves for measuring the performance. The best set of features generated areas under the ROC curve ranging from 0.84 to 0.89 when using real-valued GA and from 0.83 to 0.88 when using binary GA. The multi-wavelet method outperformed the other three methods, and the conventional shape features were superior to the wavelet and Haralick features.     

基于标记脉冲耦合神经网络的乳腺肿块分层检测方法

乳腺X线图像中的肿块检测是乳腺癌早期诊断的重要手段。该文提出了一种新的肿块检测方法。将脉冲耦合神经网络(Pulse Coupled Neural Networks, PCNN)与标记符相结合设计了标记PCNN图像分层方法,继而利用多同心层(Multiple Concentric Layers, MCL)模型得到可疑区域。最后,借助肿块的形态学特征剔除假阳性区域得到最终的肿块。实验结果表明,该文方法在保证假阳性率(False Positive Rate, FPR)的同时,肿块真阳性率(True Positive Rate, TPR)达到92.08%。同时针对东方女性致密型乳腺案例中检测结果明显优于MCL方法和MCA方法。     

Quantitative analysis of morphological techniques for automatic classification of micro-calcifications in digitized mammograms

In this paper we present an evaluation of four different algorithms based on Mathematical Morphology, to detect the occurrence of individual micro-calcifications in digitized mammogram images from the mini-MIAS database. A morphological algorithm based on contrast enhancement operator followed by extended maxima thresholding retrieved most of micro-calcifications. In order to reduce the number of false positives produced in that stage, a set of features in the spatial, texture and spectral domains was extracted and used as input in a support vector machine (SVM). Results provided by TMVA (Toolkit for Multivariate Analysis) produced the ranking of features that allowed discrimination between real micro-calcifications and normal tissue. An additional parameter, that we called Signal Efficiency*Purity (denoted SE*P), is proposed as a measure of the number of micro-calcifications with the lowest quantity of noise. The SVM with Gaussian kernel was the most suitable for detecting micro-calcifications. Sensitivity was obtained for the three types of breast. For glandular, it detected 137 of 163 (84.0%); for dense tissue, it detected 74 of 85 (87.1%) and for fatty breast, it detected 63 of 71 (88.7%). The overall sensitivity was 85.9%. The system also was tested in normal images, producing an average of false positives per image of 13 in glandular tissue, 11 in dense tissue and 15 in fatty tissue.     

Approaches for automated detection and classification of masses in mammograms

Breast cancer continues to be a significant public health problem in the world. Early detection is the key for improving breast cancer prognosis. Mammography has been one of the most reliable methods for early detection of breast carcinomas. However, it is difficult for radiologists to provide both accurate and uniform evaluation for the enormous mammograms generated in widespread screening. The estimated sensitivity of radiologists in breast cancer screening is only about 75%, but the performance would be improved if they were prompted with the possible locations of abnormalities. Breast cancer CAD systems can provide such help and they are important and necessary for breast cancer control. Microcalcifications and masses are the two most important indicators of malignancy, and their automated detection is very valuable for early breast cancer diagnosis. Since masses are often indistinguishable from the surrounding parenchymal, automated mass detection and classification is even more challenging. This paper discusses the methods for mass detection and classification, and compares their advantages and drawbacks.