向量量化應用在超音波影像的探討

Abstract

由於超音波陣列探頭在接收訊號的過程中，相近通道所收到的訊號通常具有高的相關性，因此利用這種特性將同一時間內相鄰的二個甚至更多個通道所接收到的訊號組成一個二維或多維的向量，再將此向量以最近似的編碼向量表示，這種方法也被稱為向量量化法，可以比傳統單一通道量化方式更有效率。本研究除了用Field II模擬超音波掃描仿體的訊號外，也以超音波探頭實際收到的實驗資料作為測試對象，將這些數據經過向量量化後除了比較信號至量化雜訊比外，還比較B-mode影像品質、相位偏移修正的準確度及彩色都卜勒流速偵測的準確度等。實驗結果指出向量量化法比起均勻量化法更可以有效降低量化雜訊，尤其當位元數愈小，且在相同位元長度的情形下，其所提供的信號至量化雜訊比大約可以提高10dB左右。另外在B-mode影像上，若等效的位元數愈低時，其影像品質就會愈差，然而當影像退化仍在可接受的範圍內時，平均每點的位元數大約可以降低到2位元左右。而在彩色都卜勒的應用上，由於對流速準確度的要求比起B-mode影像的要求還要更高，因此位元數大約可以從16 bits降低為一半，至於相位修正的準確度在量化過後都還可以有效修正回來。

Because the signals received by adjacent elements of an ultrasound transducer array is often highly correlated, these signals can be grouped into multi-dimensional vectors and vector quantization (VQ) can be applied to improve the quantization efficiency over conventional scalar quantization (SQ). In this research, both simulation data and experimental data are used to test this hypothesis. The signal-to-quantization noise ratio (SQNR), B-mode image quality, accuracy in phase aberration correction and accuracy in blood velocity estimation are assessed. Results show that there is about a 10dB quantization noise reduction by VQ compared to conventional SQ, particularly when bit length is short. The image degradation in B-mode is minimal. Nevertehless, velocity estimation accuracy is affected by corruption of the phase data, thus restricting the bit length reduction by only a factor of two.