基於條件隨機域方法的語意式乳房超音波腫瘤偵測

Semantic Lesion Detection for Breast Ultrasound Based on Conditional Random Fields

李思賢(SZU-HSIEN LEE)

指導教授 ： 張瑞峰

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2014年)

https://doi.org/10.6342/NTU.2014.10526

摘要

電腦輔助乳房腫瘤偵測系統可以有效率地輔助放射科醫生檢查乳房超音波影像，所以電腦輔助乳房腫瘤偵測系統被廣泛運用在乳房超音波影像的腫瘤偵測。為了發展出用於腫瘤偵測的電腦輔助乳房腫瘤偵測系統，我們採用條件隨機域模型對被放射科醫師用語意式標籤標示過的乳房超音波影像做像素分類。當腫瘤標籤的像素被決定後，我們使用孔洞填充和連通分量方法把腫瘤的候選區域組成同區域。為了減少預測區域的偽陽性，在訓練基於向量支持機器的腫瘤偵測器時，使用形態學和材料特徵來計算腫瘤候選區域為真實腫瘤的可能性。我們蒐集103個案(63個不正常個案和40個正常個案)去評估本實驗的效果。經由實驗結果，本方法達到腫瘤偵測率95%和每張圖平均有0.24個偽陽性個數。結合兩種特徵組合的品質因數(FOM)達到0.89，與其他現行方法相比時有顯著性差異(p-value < 0.05)。總結，實驗結果顯示我們提出的電腦輔助乳房腫瘤偵測系統可以有效地應用在臨床上。

關鍵字

乳房腫瘤 ； 電腦輔助偵測 ； 條件隨機域

並列摘要

Computer-aided detection systems (CADe) have been widely developed for the detection of lesions in breast ultrasound (BUS) images due to the capability and efficiency of aiding the radiologists in reviewing the breast ultrasonography. To develop a CADe system for lesion detection, we propose to adopt the conditional random fields (CRFs) model to perform the pixel classification task of BUS images with semantic labels described by experienced radiologists. Once the lesion label of the pixels can be determined, we perform hole-filling and connected component labeling to group the regions of lesion candidates. To reduce the false positives of suspected regions, morphological and textural features are used for the evaluation of the likelihood of the lesion candidate while training a lesion detector based on support vector machine (SVM). We collect 103 cases (63 for abnormal cases and 40 for normal cases) to evaluate the effectiveness of the proposed method. As the result, the proposed method achieves the sensitivity of 95% with false positive rate at 0.24 FPs/image. The figure of merit (FOM) of the combination of two feature sets is 0.89 which significantly outperforms the recent state-of-the-art methods (p-value < 0.05). In summary, the experiment results support our proposed system to be applied in the clinical use of computer-aided lesion detection.

並列關鍵字

breast lesion ； computer-aided detection ； conditional random fields

參考文獻

[1] R. Siegel, J. Ma, Z. Zou, and A. Jemal, "Cancer statistics, 2014," CA: a cancer journal for clinicians, vol. 64, pp. 9-29, Jan 2014.

[3] W. Buchberger, A. Niehoff, P. Obrist, P. DeKoekkoek-Doll, and M. Dunser, "Clinically and mammographically occult breast lesions: detection and classification with high-resolution sonography," in Seminars in Ultrasound, CT and MRI, Aug 2000, pp. 325-336.

[4] S. W. Chan, P. S. Cheung, S. Chan, S. S. Lau, T. T. Wong, M. Ma, et al., "Benefit of ultrasonography in the detection of clinically and mammographically occult breast cancer," World J Surg, vol. 32, pp. 2593-8, Dec 2008.

[5] P. Crystal, S. D. Strano, S. Shcharynski, and M. J. Koretz, "Using sonography to screen women with mammographically dense breasts," American Journal of Roentgenology, vol. 181, pp. 177-182, Jul 2003.

[6] S. S. Kaplan, "Clinical utility of bilateral whole-breast US in the evaluation of women with dense breast tissue," Radiology, vol. 221, pp. 641-649, Dec 2001.