乳腺超声中微钙化的自动检测。

Automatic detection of microcalcifications in breast ultrasound.

机构信息

Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei 10617, Taiwan and Department of Computer Science and Information Engineering, National Taiwan University, Taipei 10617, Taiwan.

出版信息

Med Phys. 2013 Oct;40(10):102901. doi: 10.1118/1.4821098.

DOI:10.1118/1.4821098

PMID:24089929

Abstract

PURPOSE

In an ultrasound (US) image, the presence of microcalcifications within breast lesions is an important indicator of malignancy. The purpose of this study was to develop a novel automatic detection system to find microcalcifications inside a breast lesion using an US image.

METHODS

Breast US images from 103 cases with microcalcifications were obtained using an US system with a 6-14 MHz transducer, and 585 microcalcification foci marked on 103 breast US images by a radiologist were used as the ground truth. After segmentation of the lesion contour using the level set method, the microcalcification candidates inside the lesion were found using adaptive speckle reduction and top hat filters. Then, three criteria were used to identify the real microcalcifications, including the mean, single point, and brightness criteria.

RESULTS

The proposed method revealed microcalcifications within the lesions in all 103 cases. The sensitivity and the false positive (FP) rate for the detection of microcalcification foci were 80.3% (470/585) and 3.1 per case, respectively. The sensitivities and FP rates for the benign and malignant cases were 79.2% (243/307) with a FP rate of 3.5 and 81.7% (227/278) with a FP rate of 2.6, respectively.

CONCLUSIONS

The authors' proposed method has the potential to provide a tool to help physicians detect microcalcifications within breast lesions.

摘要

目的

在超声（US）图像中，乳腺病变内微钙化的存在是恶性的重要指标。本研究旨在开发一种新的自动检测系统，利用 US 图像发现乳腺病变内的微钙化。

方法

使用具有 6-14 MHz 换能器的 US 系统获得 103 例微钙化病例的乳腺 US 图像，并由放射科医生在 103 张乳腺 US 图像上标记 585 个微钙化灶作为金标准。使用水平集方法对病变轮廓进行分割后，使用自适应散斑减少和顶帽滤波器找到病变内的微钙化候选物。然后，使用三个标准来识别真实的微钙化，包括均值、单点和亮度标准。

结果

所提出的方法在所有 103 例病例中均能显示病变内的微钙化。微钙化灶检测的灵敏度和假阳性率（FP）分别为 80.3%（470/585）和 3.1 个/例。良性和恶性病例的灵敏度和 FP 率分别为 79.2%（243/307）和 FP 率为 3.5，81.7%（227/278）和 FP 率为 2.6。