多級中值濾波改善非線性參數估測之雜訊問題

Abstract

「非線性參數」是聲學和超音波學中一個很重要的參數，一般表示為B/A，它代表一個介質對聲波造成的非線性影響之程度，當一個介質之B/A 值愈大，則聲波在該介質中傳遞時所產生的二次諧波之振幅也會愈大。 過去一些研究已經證實許多病變的組織其B/A 值都會較正常組織來的高，組織諧波成像技術利用聲波在病變及正常組織中產生的二次諧波振幅之差異，使病灶能呈現在超音波B-mode 影像上。這個技術雖然已經發展且商業化很久一段時間了，然而有些臨床的研究卻認為此技術在診斷上的幫助並不顯著，主要原因是病變可見度太低的關係。 如前所述，因為B/A 參數可以直接反映組織的健康或病變狀態，因此有愈來愈多的研究想要藉由估測組織的非線性參數做為超音波成像訊號，這種新的超波音波成像技術被認為將會比組織諧波成像有更好的病變可見度，因此很有機會取代傳統的超音波成像技術。 而最近被提出用來估測B/A 參數的方法：「延伸比較法」，僅證明了在超音波場可以估測正確，如果想利用B-mode 影像訊號來估測B/A，則必須先解決斑紋雜訊的問題，因此本研究提出了多級中值濾波的架構，當輸入訊雜比夠高時，這種濾波方式對改善B/A 估測的雜訊問題非常有效，而估測得到的B/A 影像其對比度也比組織諧波好，但如果輸入訊雜比太低，則改善的效果非常有限。

The “nonlinearity parameter” is an important parameter for ultrasonics and acoustics. “B/A” is commonly used to represent this parameter. It can be used to quantify the level of nonlinear response of a medium affected by sound wave interactions. The magnitude of the 2nd harmonic generation is proportional to B/A. Previous studies have demonstrated that many pathological tissues have a higher B/A than do healthy tissue. Tissue harmonic imaging ( THI ) exploits this difference, allowing lesions to be presented on ultrasound B-mode images. Although this technique has been commercially used for a long time, some studies have considered it to be no help in diagnosis because of its poor lesion visibility. As mentioned above, B/A can directly reflect the healthy or pathological states of tissues. For this reason, increasingly many studies have attempted to use the nonlinearity parameter as a new imaging signal for medical ultrasound imaging. A new technique with lesion visibility better than that of THI is needed, so the technique in question presents our field a rare opportunity to replace traditional ultrasonic technology. The newest proposed estimation method, “Extended Comparative Method” ( ECM ), has been proposed to estimate B/A correctly in the ultrasound field. ECM can also be applied to B-mode image signals only if speckle noise is suppressed. To address this issue, we propose multiple median filtering as a solution. This filtering method is highly effective when the input signal-to-noise ratio ( SNR ) is sufficiently low, but the performance of this method happens to be restricted by low input SNR.