使用超音波散射統計參數影像評分肝纖維化程度：理論分析與臨床研究

Abstract

肝硬化的早期診斷預測方式，在現代醫學中一直是眾多人所關注的研究議題，在目前肝硬化的診斷方式中，病理切片檢驗被視為是一黃金標準，但可能發生副作用及取樣誤差，近年來許多研究者開始投入非侵入式診斷方法的研究。超音波技術目前在各個領域中皆有廣泛的應用，在臨床醫學上更是不可或缺的非侵入式診斷工具。傳統的超音波灰階影像為一種定性影像，無法提供組織內部的散射子特性，因此發展定量影像和參數的評估方式逐漸成為主流。基於上述原因，本研究透過演算法對不同肝臟纖維化程度病人的肝臟掃描訊號進行定量影像的成像及定量參數的分析計算，並觀察各種參數與隨著肝纖維化病情嚴重程度的趨勢變化，以達到診斷及評分肝臟纖維化程度的目的。 由於超音波逆散射訊號隨機干涉，造成影像中存在斑紋現象，而這種斑紋現象的特性與掃描組織中散射粒子密度存在一特定關係，針對健康和病變組織掃描所得到的斑紋也會有所差異。本研究使用醫用超音波儀器收取臨床超音波掃描訊號，並使用Nakagami統計分佈來描述逆散射訊號，計算其 Nakagami參數影像。我們同時引入五種定量參數(Nakagami-m參數、紋理參數、包絡訊號強度和衰減係數)來區分肝臟纖維化的程度。 實驗結果顯示，隨著肝纖維化評分的增加，在Nakagami參數影像中反應出逆散射訊號之統計分佈由Rayleigh分佈趨近於pre-Rayleigh分佈，因此可提供一視覺化的方式來評估肝臟纖維化階段；而在定量參數計算的部分，Nakagami-m參數在早期肝纖維化分類中，隨著纖維化的嚴重程度而逐漸下降，且診斷效能良好(AUC F≥1:0.96、AUC F≥2:0.95、AUC F=3:0.97)，其餘參數則在各肝纖維化階段均無鑑別力。 為了更進一步瞭解影響Nakagami-m參數計算之變因，本研究進行了數個後續探討，包括感興趣區域(ROI)、掃描位置和脂肪肝對Nakagami-m參數計算的影響，而得到下列結論：感興趣區域(ROI)之最適大小介於7倍脈衝長度和9倍脈衝長度之間，而當大小設定為8倍脈衝長度時，Nakagami-m參數有較好的診斷效能；肝臟之掃描位置之計算結果在統計上並無明顯的偏差存在(p>0.05)；在帶有脂肪肝的病人中所收到的影像，其亮度有較高的現象，導致Nakagami-m參數之計算結果偏高，而無法反應其組織內部的散射子特性。 最後，本實驗室所原創的ACRA方法提供了一種定量參數來代表病變的組織和正常組織的差異程度，並與早期肝纖維化程度的分級具有高度的相關性。整體研究成果在肝纖維化的診斷方法上具有發展潛力且有高度的臨床應用價值。

The strategy of assessing and identifying early stage liver fibrosis has been viewed as an important issue in modern medicine. Liver histological diagnosis based on biopsy is the gold standard for liver fibrosis assessments nowadays, but sampling errors may occur when using this method. Thus, researchers have focused on developing non-invasive diagnosing method as a tool for the assessment. Ultrasonic technology has been widely applied in various fields, and has become the front-line non-invasive diagnosing tool in clinical medicine. Traditional ultrasound gray-scale image is a kind of qualitative image, and cannot provide the information of scatterers inside the tissue. Therefore, development of quantitative imaging or parameter has gradually become the mainstream. For the above reasons, we performed quantitative analysis of liver B-scan signal on patients with different stage of liver fibrosis by using our algorithm, and investigate the quantitative parameters along with the severity of fibrosis in this study. The characteristic of ultrasound speckle pattern in B-scan image, which results from the wave interference phenomenon of backscattering signal, was considered to be associated with the density of scatterers in tissue, which can be used in differentiating between healthy and diseased tissues. In this study, clinical ultrasound scanning signals were obtained by medical ultrasound equipment. Backscattering signals were described by Nakagami statistical distribution, and a Nakagami-model-based image has been calculated. Five kinds of quantitative parameters including Nakagami parameter, texture properties, mean intensity and attenuation coefficient, were also introduced for assessing the degree of fibrosis. Analysis results showed that the global statistics of backacattered signal changed from a Rayleigh distribution to a pre-Rayleigh distribution when the fibrosis score increased. It means that Nakagami image can be used for distinguishing different degrees of fibrosis. Calculation results of quantitative parameters revealed that Nakagami parameter decreased with the fibrosis score, and has outstanding performance in scoring early stage fibrosis (AUC F≥1:0.96、AUC F≥2:0.95、AUC F=3:0.97), while other parameters showed limited performance in staging fibrosis. In order to further understand variables that affect the calculation of Nakagami parameter, several investigations had been done in this study, including the size effect of region of interest (ROI), the effect of scanning position and fatty liver. Results showed that the optimum size of ROI lies between 7 times pulselength and 9 times pulselength. In addition, Nakagami parameter has better performance when setting ROI-size to 8 times pulselength. Analysis results also showed that no significant statistical deviation exists in calculation when changing scanning position in human liver. It also found that the brightness of scanning image is higher in patients with fatty liver, caused higher calculation results of Nakagami parameter, which leads to the failure of reflecting scatterer properties in the tissue. Last but not least, the ACRA method proposed by our lab provides a quantitative parameter to represent the degree of difference between normal and abnormal tissue. And results of the method demonstrated that ACRA parameter is highly correlated to the stage of early fibrosis. It is concluded that current findings of this study has great potential and clinical application value in diagnosing liver fibrosis.