使用超音波參數影像與紋理分析評分肝臟纖維化程度

Abstract

肝臟纖維化是一緩慢且長期的過程，若置之不理有很高機率會導致肝硬化。但現在的醫學發展，除了侵入式的病理切片檢驗外，尚無有效診斷出纖維化程度的定量技術。超音波是一種非侵入式的診斷工具，它具有即時影像的優點，目前在纖維化檢測中已成為第一線的臨床診斷工具。然而傳統超音波灰階影像是一種定性影像，對於纖維化的特徵描述並不明顯，需要有經驗的醫生才能判讀，而初期的纖維化甚至連醫生都無法察覺。考量以上因素，於是本研究決定朝著發展超音波定量影像的方向進行，並以統計參數來描述組織特性，達到診斷肝臟纖維化的目的。 超音波經由探頭發送訊號至組織內部，並接收回波的隨機逆散射訊號，藉此分析其中隱含的組織訊息。我們引入Nakagami統計分佈來描述組織特性，發現其Nakagami參數m值可定量纖維化程度，而Nakagami影像也較傳統灰階影像更能夠分辨纖維化。在動物實驗中，我們將老鼠注射DMN藥物，誘發肝臟纖維化，並計算每個纖維化階段之m值，同時與病理切片分數做比較。結果顯示，纖維化程度與m值呈現正相關，即使病理分數同為0分的肝臟，m值仍隨著纖維化程度上升而增加，意味著m值的靈敏度比病理切片分數為高，並且能成功辨識初期的纖維化與否， 符合臨床診斷的需求。 我們同時引入紋理分析來描述肝臟纖維化，並計算4種影像參數與m參數做比較，分別是：對比值、相關性、能量均勻性和均質性，發現不論趨勢或動態範圍，紋理分析參數皆不如m參數描述來的完整，因此利用m參數來判別纖維化程度是合理且適當的。

Liver fibrosis is a tardy and long-time process, and it has high probability to induce liver cirrhosis if we pay no attention on it. However, the development of medicine nowadays lacks an effective quantitative technique to diagnose fibrosis stages except biopsy, an invasive examination. Ultrasound is a non-invasive diagnose tool. It has the advantage of real-time and has became the front-line diagnose tool on detecting fibrosis stage. But the traditional ultrasound gray-scale image is a qualitative image. It cannot describe the characteristic of fibrosis clearly and always need the experienced doctor to judge. In fact, even experienced doctors cannot perceive the initial stage of fibrosis with traditional images. Consider the above reasons, so we decide to develop the quantitative ultrasonic image and use statistic parameters to describe fibrosis stages, and finally achieve the purpose of fibrosis diagnosis. The ultrasound transducer emits signals into tissues, and receives the random backscatter signals of echoes. We can obtain the hidden information of tissue by analyzing these random signals. To describe the characteristic of tissues, we introduce Nakagami statistic distribution, and find the Nakagami parameter – m can quantitate fibrosis stages. Besides, Nakagami images are more effective to distinguish fibrosis stages than traditional gray-scale images, too. In animal experiments, we injected DMN to rats to induce liver fibrosis, and calculated the value of m at each fibrosis stage, and compared m with the biopsy scores. The result shows that fibrosis stages are positive related with the value of m. Even some biopsy scores are the same of zero, the value of m still increases with the fibrosis become serious step by step. It means the sensitivity of m is higher than biopsy score. And the value of m conforms the need of clinical diagnose with detecting initial stage of fibrosis successfully. At the same time, we also introduce the texture analysis to describe fibrosis stages and calculate four types of texture parameters to compare with the value of m. They respectively are contrast, correlation, energy, homogeneity. We find that no matter trend or dynamic ranges, the description of texture parameters about fibrosis is less complete than the value of m. So we use the value of m to distinguish liver fibrosis stages is reasonable and adequate.