使用可適性組織聲速偵測與自動化深度增益補償最佳化超音波影像

Abstract

超音波成像利用探頭發射聲波進入人體內，而與人體內的各種介質發生交互作用，而利用反射回的信號重建人體內影像，陣列超音波系統由於可以進一步調整探頭單元間的時間差而有更好的成像品質與應用而被廣泛地應用在醫學診斷上。然而在目前影像系統中，人體內組織不均勻的特性卻會對成像過程造成影響，舉凡組織聲速、折射、散射、衰減係數…等因素均會隨著不同的組織而有不同的物理效應，進而影響到成像的品質與診斷正確率。本論文將對於偵測平均聲速與亮度衰減補償兩方面提出自動化校正方法，期待藉由提升影像品質而增進診斷正確率。在聲速偵測方面，使用平均聲速成像在超音波成像系統中可降低因組織不均勻而造成影像品質下降的影響，影像品質包括了空間與對比解析度以及病理位置的判定…等，在目前成像系統中，往往採用預設好的聲速(例如:1540m/s)進行成像，而在此預設聲速與真正平均聲速之間的誤差會導致影像品質降低進而影響到診斷效果，因此本論文以平行化架構同時分析不同聲速下的影像品質來進行自動化聲速預估與校正，此分析方法為利用不同位置的子孔徑成像並計算誤差而推估正確聲速。而在亮度衰減補償方面，超音波成像過程之聲波傳遞的過程中，舉凡組織種類、入射角度、有效頻率與孔徑的下降…等均會導致在超音波影像之亮度與動態範圍隨著深度增加而有非規則降低的現象，診斷上可能會造成病理誤判，而在成像系統中，往往以固定預設的線性補償來降低此影響，加上依靠操作人員的診斷與經驗經由調整不同深度的亮度增益與動態範圍才可得到較為真實較為可信的無衰減影像，此過程十分費時與不精確，故此論文亦提出了自動化亮度增益補償方法以改善此人工操作之問題，此補償方法包含了先利用統計分區標定出主要組織以判斷其衰減曲線，接著對其安排了二維查找表與線性組合計算各取樣點的亮度補償，亮度補償後為了處理可能造成信雜比下降的傷害，此方法於最後加上了隨深度變化的選擇性對比加強器以適當調整影像對比，此方法可有效增進影像均勻度使影像品質提升。最後，本論文將此兩種自動化演算法於平行化軟體平台上(CUDA)加以實現，並達到在診斷前以短時間內進行最佳化影像之目標。

In ultrasound array imaging, using correct sound speed is one of the most important tasks to ensure beam-forming accuracy. However, due to variations in body type and tissue inhomogeneities, the actual sound speed ranging from 1400m/s to 1600m/s is hard to be accurately applied during beamforming. In most current systems, the sound velocity is pre-determined as a constant that would influence whether or not the image representation of the anatomy is reliable for diagnosis. In this research, a global mean sound speed estimation method is presented. It utilizes a sub-aperture imaging technique to form two images with different sound speed values, and choose the optimal one with the minimum mean error between these images as the estimated sound speed value. This optimal sound speed effectively provides higher contrast-to-noise ratio on experimental phantom images as well as clinical images. The other part of the research is regarding tissue attenuation that causes brightness changes in the image and thus an adaptive depth gain compensation method is needed. In current systems, incorrect compensation may be presented and the operators needs to adjust the gain and dynamic range manually. Therefore, in this research, an adaptive depth gain compensation algorithm is proposed with the purpose of automatically providing better image uniformity. This compensation algorithm first calculates attenuation curves from main tissue regions of images, and based on these curves, a 2d look-up table for linear composition could be generated to adapt the brightness pixel-by-pixel. Furthermore, in order to decrease influences of signal-to-noise ratio degradation by compensation, an adaptive contrast enhancement is added to last stage of this method. Finally, the computations of these two algorithms can be finished within 1 second on GPU platform.