用於三維細胞膠之單探頭超音波反射式剪切波彈性影像

Abstract

細胞組織的彈性硬度與癌症的階段和病灶嚴重程度有很高相關性，臨床上能透過定量三維細胞膠微環境之彈性硬度來觀察癌症的治療效果，先前研究中有提出反射式剪切波（reflected shear wave）彈性影像的概念，有著量測誤差與耗材體積較大兩缺點待改善，故本研究最佳化了該反射式剪切波彈性影像系統。首先透過k-Wave模擬2D剪切波的傳遞與反射過程，來發展本實驗反射式剪切波系統的演算法，在實驗中利用20 MHz高頻單探頭打出2000週期的超音波來製造聲輻力來產生剪切波，搭配直徑12 mm圓柱狀1.5%水膠反射壁的設計，與待測物造成阻抗差異使剪切波產生反射波，利用已知的反射壁距離用飛時測距與峰值尋找…等演算法，獲得反彈剪切波訊號的回來時間差，計算待測物之平均剪切波速度進而定義其平均彈性硬度，在誤差上提出新版本演算法標準差0.02 m/s比舊版本縮小了約50%，並節省50%原先三維細胞膠培養時所需的藥品花費，達到了實驗兼顧經濟成本的效果。另一方面，本研究還針對未來實驗中癌症有彈性硬度非均質狀態的情況，進行重構介質彈性分布的討論，例如：三維細胞膠中有團聚的腫瘤球，因為本研究使用的單探頭系統無法像傳統電腦斷層使用反向投影演算法，故本研究透過生成各種的非均質彈性硬度分布介質的模擬資料，約15000筆，使單探頭繞3 mm的同心圓在36個角度下旋轉，以類斷層掃描的方法，收集不同位置下待測物的剪切波反彈訊號特徵，使用機器學習搭載深度多層的訓練模型，學習重構該介質的原始彈性硬度分布狀態與訊號上的關係，測試於模擬資料上模型回召率達到0.8有良好的表現，並套用至實際實驗中非均質仿體與非均質三維細胞膠的彈性硬度重構上。

The shear modulus of human body tissues has a high correlation with the stage of cancer. This study developed the single-transducer reflected shear wave elasticity imaging system（RSWEI）for 3D cell cultures to effectively quantify the shear modulus after cancer treatment. This study showed that the new system saved 50% of the cost and has a smaller standard deviation in measuring the shear modulus, the new algorithm was 0.02 m/s, while the previous version was 0.04 m/s, reducing the error by about 50%. In this study, k-Wave was used to simulate the transmission and reflection of 2D shear waves to develop the algorithm for the reflection shear wave system. In the experiment, a 20 MHz single-transducer was used to generate 2000 cycles of ultrasound to produce acoustic radiation force to generate shear waves, and a 1.5% agarose gel reflecting wall and the tested object was designed to cause impedance differences to make shear wave reflections. In homogeneous assumption, analyzing the elasticity imaging of the single transducer by the time of flight and find-peaks algorithms were used to obtain shear wave speed information. In addition, this study also discussed the reconstruction of the elasticity distribution of heterogeneous states generated by cancer characteristics, such as tumor balls with aggregated cell gels. Since this system cannot use filtered back-projection algorithm like traditional CT scans, simulated data with heterogeneous shear wave speed media were generated, and the single transducer was rotated around 36 angles to collect reflected shear wave signal characteristics at different positions, using machine learning to train the model and learn the relationship between the original medium and scanning signals. The model achieved good performance with a recall rate of 0.8 on simulated data and was ultimately applied to reconstruct the elasticity stiffness of heterogeneous phantoms and 3D cell gels in actual experiments.