乳房腫瘤之形狀力學模型建立

Abstract

乳癌為目前全球女性發生率最高的癌症，早期的檢測可增加乳癌的治癒率。乳房超音波影像為常用來進行診斷的方法之一，為了提升準確度，本研究團隊設計一個乳房超音波的輔助診斷系統，在得到影像資訊的同時，亦提供使用者腫瘤的生物力學特性。 系統中應用不同乳房組織硬度不同的特性，此現象常被用於觸診檢查之中。系統中探討探頭與腫瘤之間的力量關係，以主成分分析方法自力量中萃取特徵，作為支持向量回歸模型的輸入，建立腫瘤硬度比預測模型。 本研究中模型與過去模型的差異，在於對腫瘤外形的描述方法與建立預測模型方法，改善了過去模型在內插硬度比以及不規則腫瘤形狀預測不良的問題。在相同條件下，內插硬度比預測誤差降低了14%。而在改變腫瘤形狀描述方法後，對於不規則形狀的預測誤差亦降低了14%。相較於過去的模型，新的模型預測效果較佳，且更趨近於臨床的應用。

The stiffness of different breast tissues are different. And this phenomenon is often used in clinical palpation. But the result of palpation is subjective. Thus there are increasing researches aim for providing the physicians secondary opinions. And some study is related to the tumor biomechanical properties. In previous research, we used the force relation between probe and tumor. And extract some features from the force curves. Then we construct a neural network biomechanical model to predict tumor hardness ratio. However, the hardness prediction of the irregular shape tumor and the internal hardness tumor samples are bad. In this thesis, the problems of previous model are improved. I construct a support vector regression biomechanical model, and use the tumor shape statistic model to express the tumor appearance. Then the relative error of internal hardness prediction decrease 14%, and the error of irregular shape tumor hardness prediction also decrease 14%. Compared to previous model, the new model is closer to clinical application.