超音波影像紋理的視覺特徵量化與反掃描轉換

Abstract

臨床醫學使用超音波影像來診斷肝臟疾病，很倚賴視覺所看到的特徵；然而這些特徵並沒有前後一致的定義。在本論文裏，幾種由Amadasun所提出的特徵被用來量化超音波影像中所看到的特徵。在量化這些特徵之前，會先利用「反掃描轉換」來降低因為超音波成像「數位掃描轉換」所帶來的影響；這個反掃描轉換必須考量特徵的性質而設定取樣頻率。本研究使用了300份海綿和肝臟的超音波影像來探討這些特徵，以及反向掃描轉換的必要性。本研究發現所使用的特徵必須根據所要觀察的組織特性，而Amadasun所提特徵中的「coarseness」和「busyness」正好可以反映fully developed speckle和肝臟組織的特性。

The visual features are important in clinically diagnosing liver diseases with ultrasound image texture. However, there is no consistent definition of these features. In this study, several textural features proposed by Amadasun are adopted for quantifying the visual features of ultrasound image texture. To cope with the distortion caused by the digital scan conversion (DSC) in ultrasound imaging, a back-scan conversion (BSC) algorithm is applied to homogenize the sampling format and sampling rate of ultrasound image texture before measuring these features. The effectiveness of this measure is investigated using 300 ROI’s of sponge and liver images. It is confirmed that BSC is an important preprocessing step in quantifying these features of ultrasound image texture. By this measure, the visual features may be quantified. The results show that the use of the features is dependent on what is looking for; that is, the distinction between tissue echotexture and fully developed speckles should emphasize on the “coarseness” of the echotexture, while the one between normal liver and cirrhosis should emphasize on the spatial intensity variation (busyness). This study shows a correlation between the tissue and the ultrasound echotexture.