肝臟組織特性之超音波辨識

Ultrasonic Tissue Characterization of Liver

葉文俊(Wen-Chun Yeh)

指導教授 ： 李百祺

國立臺灣大學/電機資訊學院/電機工程學研究所/博士(2005年)

https://doi.org/10.6342/NTU.2005.03048

摘要

纖維化與脂肪變性是肝臟常見的兩種組織學變化。纖維化是因肝臟的發炎所造成，程度範圍可以從最輕的門脈區分布到最嚴重的肝硬化，而肝硬化本身亦有輕微與嚴重之差異。纖維化程度偵測十分重要，臨床醫師可據此決定慢性肝炎病人的治療方針。脂肪變性為脂肪積聚在肝細胞內的組織學現象。因為嚴重的肝臟脂肪變性可引起肝功能異常，所以脂肪變性程度的偵測也很重要。肝臟切片是判斷纖維化與脂肪變性程度最精確的方式，但是此種侵入式檢查可能會引起嚴重的併發症，因而需要一個非侵入式又能精確偵測肝臟組織特性的方法。肝臟廣泛性的纖維化結構可造成彈性特性及超音波影像的變化。超音波因可使用彈性影像量測肝臟組織彈性，及二維影像擷取影像特徵，故是準確偵測肝臟組織特性很有潛力的工具。至於脂肪變性會使肝臟超音波影像亮度增加，亦可造成超音波影像特徵的變化，但是傳統超音波無法解析組織的微小變化(如脂肪油滴的分佈)，高頻超音波(≧20 MHz)因有更高的解析度，故有機會看到脂肪變性的更多細節，對脂肪變性程度作更精確的鑑別。本論文先對肝臟超音波彈性影像的重要基礎工作–肝臟彈性的量測–作研究，再進一步探討傳統超音波偵測肝臟纖維化程度及高頻超音波偵測肝臟脂肪變性程度的能力。
為了評估纖維化病變對肝臟彈性的影響及研究超音波彈性影像的潛在價值，我們以19個手術切下的人類新鮮肝臟檢體及一個肝臟腫瘤檢體進行以循環壓縮放鬆的方式作一維的彈性測量。以電子秤量到的壓力除以步進馬達設定的應變即可得到彈性模數，然後每個檢體接受病理檢驗來區分纖維化的程度(0-5級)。結果顯示除了某些中度纖維化的檢體外，彈性模數會隨著纖維化的程度而漸增，纖維化的程度和彈性模數有明顯相關。此外，前項研究中的16個人類肝臟檢體與4個新檢體並接受超音波掃瞄所得之二維影像，以評估超音波對肝臟纖維化程度之分類能力。以灰階值共現(gray-level concurrence)與不可分小波轉換(nonseparable wavelet transform)兩種方法所得之超音波影像特徵，用支持向量機制(support vector machine)這個分類器來分辨纖維化程度並以病理檢驗結果測試其準確度。每件肝臟檢體都接受病理檢驗並依其纖維化程度分級(同前項研究分0-5級共6級)，這6級再組合成4種分類方式(2, 3, 4或6類)。結果顯示2, 3, 4與6類的最佳分類準確度是91%, 85%, 81% 與 72%，證實纖維化程度可用這種非侵入式方法來作分類，雖然類分越多其表現越差。
再者，為了評估高頻超音波對肝臟脂肪變性程度之分類能力，以25-MHz單晶體探頭對另19個手術切下的人類肝臟檢體作掃瞄，以分辨其脂肪變性程度。最後以病理檢驗結果驗證其正確性。每件肝臟檢體都接受病理檢驗並依其脂肪變性程度分成0 到3 級。這4級再組合成3種分類方式(2, 3或4類)各作分類。結果顯示2，3與4類的最佳分類準確度是90.5%，85.8% 與82.6%。與傳統超音波(7 MHz)作比較(最佳分類準確度各是81.6%，75.8% 與74.2%)，高頻超音波之分類準確度較高。
總之，以上研究顯示超音波是肝臟組織特性辨識的優秀工具。本論文之第一項研究發現肝臟的硬度和纖維化的嚴重度有很好的相關，而且為超音波肝臟彈性影像的臨床應用提供很好的基礎。第二項研究證實肝臟纖維化程度也可用非侵入式的傳統二維超音波影像來作分類。而第三項研究則指出，肝臟脂肪變性程度需要更高解析度的高頻超音波才能有較高之分類準確度。然而高頻超音波要應用於臨床肝臟掃瞄，先需克服其基本的限制，本論文亦針對這些限制可能之克服方法，作一探討。在超音波組織辨識這個研究領域，本論文的主要貢獻在於對各種肝臟組織特性的辨識上，已有系統地作了許多先驅性研究﹕在彈性影像方面，本論文是第一個以新鮮人類肝臟作彈性量測的研究﹔在影像特徵辨識方面，本論文則成功的將支持向量機制這種分類器應用在超音波影像辨識的研究﹔在高頻超音波應用方面，本論文則進行了第一個人類肝臟二維高頻超音波影像之分析。相信這些研究成果，有助於超音波肝臟組織特性辨識的臨床應用。

關鍵字

肝臟彈性 ； 彈性影像 ； 脂肪變性 ； 纖維化 ； 組織特性 ； 肝臟 ； 灰階值共現 ； 不可分小波轉換 ； 支持向量機制 ； 分類 ； 超音波 ； 高頻超音波 ； 肝臟病理

並列摘要

Fibrosis and steatosis are two common histological changes of the liver. Hepatic fibrosis is caused by inflammation and the degree of fibrosis may range from fibrous expansion in the portal area to cirrhosis. Detection of fibrosis status is very important. By knowing the fibrosis status, proper treatment can be given to the patients of chronic hepatitis. Hepatic steatosis – the accumulation of fat within liver cells – is another common histological finding of liver. Detection of steatosis status is also important because severe steatosis of liver may induce elevated levels of serum liver enzymes. Liver biopsy has been used to accurately identify the fibrosis and steatosis. However, the needle biopsy procedure is invasive and sometimes may cause severe complications, hence not a routine diagnostic procedure. Therefore, a non-invasive method for liver tissue characterization is desired and ultrasound may be most feasible tool. A diffuse fibrotic structure changes the mechanical properties of liver and affects characteristics of the ultrasound image. Since the elasticity of liver potentially can be measured by ultrasonic elasticity imaging and the fibrosis grade can be classified using the image features, ultrasound is very promising for tissue characterization of liver. Besides, steatosis may increase the echogenicity of the liver and change the image features as well. However, the histological changes associated with steatosis, such as distribution of fatty droplets, cannot be resolved by conventional ultrasound. High-frequency ultrasound (≧20 MHz) with higher spatial resolution has the potential to reveal more details of steatosis and achieve higher classification accuracy. In this thesis, we measure the elastic properties of fresh human liver tissues, as the first step for elasticity imaging of liver. The capability of the ultrasound in determining the liver fibrosis is evaluated by conventional ultrasound and the steatosis by high-frequency ultrasound.
To evaluate the impact of fibrosis on elastic properties of human liver and to investigate potential benefits of ultrasonic elasticity imaging, 19 fresh human liver samples and one hepatic tumor (focal nodular hyperplasia) sample obtained during operation are studied. Simple 1D estimates based on the cyclic compression–relaxation method are preformed. Elastic modulus values are derived from the pre-determined strain and the stress values. Each specimen subsequently receives histological examination and a grade of liver fibrosis is scored from 0 to 5. Results show that the elastic modulus generally increases with the fibrosis grade though some discrepancies exist at the middle grades of fibrosis (score 1-3). It is concluded that severity of fibrosis has a good correlation with stiffness of liver.
On the other hand, B-mode images of 16 specimens in above-mentioned study and 4 new specimens (total 20 fresh human liver samples) are also obtained to evaluate ultrasound’s ability in determining the grade of liver fibrosis. Image features derived from gray-level concurrence and nonseparable wavelet transform are extracted to classify fibrosis using a classifier known as the support vector machine. Each liver sample subsequently undergoes histological examination and liver fibrosis is graded as 0-5 (i.e., totally 6 grades). The 6 grades are then combined into 2, 3, 4 and 6 classes. Classifications using the extracted image features by the support vector machine are tested and correlated with histology. The results reveal that the best classification accuracy of 2, 3, 4 and 6 classes are 91%, 85%, 81% and 72%, respectively. Thus, liver fibrosis can be noninvasively characterized using B-mode ultrasound even though the performance declines as the number of classes increases.
High-frequency B-mode ultrasound images of 19 new fresh human liver samples are also obtained to evaluate their usefulness in determining the steatosis grade. The images are acquired by a mechanically controlled 25-MHz single-crystal probe. A subsequent histological examination of each liver sample grades the steatosis from 0 to 3. The 4 grades are then combined into 2, 3 and 4 classes. The classification results are correlated with histology. The best classification accuracies of the 2, 3, and 4 classes are 90.5%, 85.8%, and 82.6%, respectively, which are markedly better than those for conventional ultrasound at 7 MHz (best classification accuracies of 81.6%, 75.8%, and 74.2%, respectively). These results indicate that liver steatosis can be more accurately characterized using high-frequency B-mode ultrasound.
In conclusion, these studies demonstrate that ultrasound is a promising tool for tissue characterization of liver. The first study shows that severity of fibrosis has a good correlation with stiffness of liver and it provides a good basis for the elasticity imaging of liver in clinical use. The second study shows that liver fibrosis can also be noninvasively characterized using conventional B-mode ultrasound. In the third study, it is shown that liver steatosis needs higher spatial resolution of ultrasound for more accurate characterization. Nevertheless, clinical applications of high-frequency ultrasound for liver will not be feasible unless the penetration limitation is overcome. The main contributions of this thesis include the following pioneer works in ultrasonic tissue characterization of the liver. In the first study, we were the first group in the world to measure elastic properties of fresh human liver samples and investigated the correlation with histological findings. In the second study, we successfully applied the support vector machine to the field of ultrasonic tissue classification. In the third study, we performed pioneering work of high-frequency liver imaging and tissue characterization. These research results will be instrumental to our future work.

並列關鍵字

Liver elasticity ； Elasticity imaging ； Support vector machine ； Nonseparable wavelet transform ； High-frequency ultrasound ； Liver pathology ； Ultrasound ； Classification ； Fibrosis ； Gray-level concurrence ； Tissue characterization ； Liver ； Steatosis

參考文獻

Adinolfi LE, Gambardella M, Andrena A rt al. Steatosis accelerates the progression of liver damage of chronic hepatitis C patients and correlates with specific HCV genotype and visceral obesity. Hepatology 2001;33:1358-1364.

Angulo P. Nonalcoholic fatty liver disease. N Engl J Med 2002;346:1221–1231.

Aklonis JJ. Introduction to polymer viscoelasticity, second edition. New York: John Wiley & Sons, 1983:7-9.

Bedossa P, Dargere D, Paradis V. Sampling variability of liver fibrosis in chronic hepatitis C. Hepatology. 2003;38:1449-1457.

Bercoff J, Tanter M, Fink M. Supersonic imaging: a new technique for mapping of the visco-elastic properties of tissues. 28th International Symposium on Ultrasonic Imaging and Tissue Characterization, Arlington, VA, May 28–30, 2003. Ultrason Imaging 2003.