超音波影像之乳房腫瘤特徵項強化研究

Research of Ultrasound Image Enhancement for Breast Tumor Teatures

翁士庭(Shih-Ting Wung)

指導教授 ： 陳正剛

國立臺灣大學/工學院/工業工程學研究所/碩士(2013年)

https://doi.org/10.6342/NTU.2013.02884

摘要

超音波掃描是常見的身體掃描診斷方式之一。超音波成像並不像X光攝影檢查術那般具有侵略性，且無放射問題。超音波成像也提供動態影像，利於放射科專家或醫師擷取即時影像圖或針對特定影像區域研究。然而，由於機器掃描以及超音波特性的限制，某些組織成像效果較差，尤其是具有較厚層皮下組織與脂肪的病患，也因此增加分析超音波影像之難度。在本研究中，我們便針對特定的腫瘤特徵項進行超音波影像的強化。

我們嘗試了一些傳統影像處理方式，例如直方圖均化[1]與其變形加權式直方圖分解[2]。另外，我們也針對特徵項開發新的影像處理，如加權式變異值、遮罩差異值等。這些影像處理方式可互相結合而生成其他影像效果。

在經過影像強化後，我們透過測試腫瘤的特徵值與自動輪廓圈選效果來檢驗影像強化之功效。以上實驗測試對象為台大醫院提供之 263筆乳房腫瘤影像，其中包含198筆良性腫瘤和65筆惡性腫瘤；另外再從其中隨機抽取良惡性腫瘤各五十筆進行自動輪廓圈選效能測試。

關鍵字

超音波影像 ； 影像強化 ； 異質性 ； 自動輪廓圈選 ； 直方圖均化 ； 加權式直方圖分解 ； 變異數加權

並列摘要

Ultrasound scanning has been one of the most popular ways for medical detection and diagnosis. Unlike x-ray imaging, such as chest-x-ray, mammography, etc., sonography is less invasive and is radiation-free. It also provides dynamic imaging that is helpful for radiology professionals to capture instant imaging or to focus on certain suspicious region. However, because of the machine limitation, the image of the histological structure located farther away from the skin is becoming vague, especially for those patients who have thicker subcutaneous tissue and fat. As a result, analyzing ultrasound image may sometimes become a difficult job. In this thesis, our purpose is to enhance ultrasound image for better analysis of tumor features.

We apply several conventional numerical methods to enhance ultrasound images, such as histogram equalized [1], weighted histogram separation [2]. We have also developed weighted variation, and absolute mask difference enhancement methods. These methods can be further combined with one another to generate a variety of new methods.

To verify whether these methods are effective, features of tumor interior heterogeneity and contour segmentation are tested. The proposed methods are validated with 263 Breast Cancer (BC) images collected at National Taiwan University Hospital including 198 benign and 65 malignant tumors for tumor features testing. 100 images were selected randomly from 263 images with 50 each benign and malignant tumors for contour segmentation testing.

並列關鍵字

ultrasound ； image enhancement ； interior heterogeneity ； auto tumor segmentation ； histogram equalization ； weighted histogram separation ； weighted variation

參考文獻

[1] Rafael C. Gonzalez, Rafael C. Gonzalez, “Digital Image Processing”

[2] Soo-Chang Pei, Yi-Chong Zeng and Jian-Jiun Ding, “Color Images Enhancement Using Weighted Histogram Separation”

[6] Stavros, A.T., et al., Solid breast nodules: use of sonography to distinguish between benign and malignant lesions. Radiology, 1995.(1), P.123~P.134.

[3] Website of Female Cancer Foundation “http://www.femalecancer.org.tw/

index.php?option=com_content&task=view&id=483&Itemid=31”