三維乳房腫瘤外形主成分分析應用於乳房腫瘤橫向探測實驗

Abstract

乳癌是全世界女性的第二死亡主因，早期檢測以及診斷可增加乳癌的治癒機率，降低乳癌死亡人數以及社會成本支出。乳房超音波影像是最常用來診斷乳癌的方法之一，為了增加乳房超音波影像診斷上的準確度，提供第二意見的資訊作為乳房超音波影像的輔助診斷已有相當多的研究，而本研究即是提供腫瘤的生物力學特性作為診斷之輔助。 一般來說，乳房腫瘤的硬度比臨近的正常乳腺組織還硬，惡性腫瘤的硬度亦高於良性腫瘤，此現象常被用於臨床觸診。然而進行觸診的過程中會因每位醫師的經驗不同或是施力不同造成判斷上的誤差，且無量化的標準。另外對於某些較深且較小的乳房腫瘤就無法以人工觸診來診斷。先前研究中有以橫向探測方式，利用裝有力量感測器的探頭壓過乳房腫瘤組織的仿體，找出具有描述腫瘤特性的力量參數群，並以模糊類神經網路瘤力學模型有效地預測腫瘤硬度，然而先前研究探討之腫瘤形狀為規則之球體，對臨床中腫瘤的不規則形狀未有完整的描述。本研究以超過兩百個超音波腫瘤影像當作樣本，並以主成分分析找出腫瘤外型的主要特徵值，取出第一個主成分特徵值做為基準並做出相對應的仿體，並進行橫向探測實驗，將腫瘤形狀變形因素加入之前研究的判斷模型，使系統更趨近於臨床應用。

Generally breast tumor tissue is harder than the adjacent normal breast tissue, malignant tumors are also harder than benign tumors, this phenomenon is often used in clinical palpation. However each physician's diagnosis has no quantitative criteria and the judgment is all subject to their expertise and judgment. In addition, it is very difficult to use palpation to diagnose the inner and smaller breast tumors. Thus there are increasing researches aimed for providing physicians secondary opinions, such as tumor hardness and mobility, in order to increase the diagnosis accuracy. In our previous studies, we used the probe of force sensor to press the breast tumor phantoms of different hardness, found the data responded from force sensor. Furthermore, we constructed a neural fuzzy network biomechanical model to predict tumor hardness. However, the model used in previous studies is inadequate because the model only consider the tumors with spherical shapes. Therefore the study focuses on modeling the tumors with irregular shapes. Data based of more than two hundred breast tumor ultrasound images samples have been used to construct a statistical model to describe breast cancer morphology. We adopt principle component analysis to identify the main features of tumor shapes. The preliminary result shows that the first principle component of the statistical shape model can effectively describe the tumor morphology and was chosen to made corresponding tumor phantoms. In the future a biomechanical model of breast tumor with irregular shape will be formulated based on the results.