利用紋理分析於腦部電腦斷層中的中大腦動脈影像評估血管動脈硬化

Texture Analysis of Middle Cerebral Artery CT Images in Atherosclerosis Evaluation

王裕翔(Yu-Siang Wang)

指導教授 ： 蘇振隆

中原大學/工學院/生物醫學工程研究所/碩士(2011年)

https://doi.org/10.6840/cycu201101068

摘要

動脈粥狀硬化為導致國人罹患心血管疾病、腦血管疾病及週邊血管疾病的主要原因之一。通常病患都已經患有臨床相關症狀才進行治療，錯過了最好的治療時機，所以導致國人因為此類相關疾病的死亡率也相對提高許多。而目前血管硬化疾病主要是經由電腦斷層影像上的特徵變化以及利用週邊血管檢查數據來做為主要診斷，因此本研究主要目的為利用腦部斷層影像特徵和週邊血管檢查訊號，輔以病患資料建立一套評估血管硬化系統。
本研究採用曾經接受腦部斷層檢查及週邊血管檢查的病人，並記錄病患週邊血管檢查訊號結果、人口學資料、臨床資料等。在病患影像方面，結合紋理量化分析以及影像增強技術進行分析，(1)影像增強：利用伽瑪校正，將腦部斷層影像透過非線性對比增強演算法，以改善局部對比度。(2)紋理量化分析：中大腦血管動脈區域使用空間灰階共生矩陣於影像並擷取其紋理特徵值，利用中大腦血管動脈硬化特徵與周圍腦組織分析其差異性。(3)局部二位元圖形：透過此方法增加影像紋理的敘述能力，幫助紋理特徵有更好的解釋能力。最後，整合血管動脈硬化特徵參數、 週邊血管檢查參數及病患個案資料以主成分分析法進行統計分析，並且以倒傳遞類神經網路評估血管硬化。
研究系統以週邊血管檢查訊號為基礎，各別找出腦部斷層影像和病患資料中重要的參數資訊，分別以20組正常病例以及200組血管動脈硬化病例進行倒傳遞類神經網路訓練，並且以14組正常病例以及57組血管動脈硬化病例進行測試。經過統計分析以及類神經網路評估後，可以發現本研究中腦部斷層影像特徵擷取，其Accuracy為94.37%、Sensitivity為96.49%、Specificity為85.71%、Kappa value為0.756；以及使用局部二位元圖形方法於腦部斷層影像特徵擷取，其Accuracy為98.59%、Sensitivity為98.25%、Specificity為100.00%、Kappa value為0.936，兩者評估結果以使用局部二位元圖形於腦部斷層影像有較好的鑑別能力。
本研究利用空間灰階共生矩陣以及局部二位元圖形於腦部斷層影像並且量化影像紋理特徵，發展出一套評估血管硬化之系統，提供臨床醫師利用影像以及病患個案資料參數來進行分析；透過紋理特徵參數分析的量化資訊以及整合病患個案資料，能提供醫師於診斷血管硬化更多輔助。

關鍵字

局部二位元圖形 ； 空間灰階共生矩陣 ； 電腦斷層影像 ； 動脈粥狀硬化 ； 倒傳遞類神經網路

並列摘要

Atherosclerosis, one of main reason leads to suffer Cardio-vascular disease, cerebro-vascular disease, and peripheral vascular disease. Patients usually treatment after have clinical symptoms. Therefore, the mortality increased in this type of related diseases. The most commonly diagnosed for Atherosclerosis is based on computer tomography (CT) images and peripheral vascular measurement. The purpose of this study is to improve the ability to detect Atherosclerosis in CT images by combining analytic system of quantizing with contrast enhanced images and peripheral vascular measurement information.
This study used texture quantification and image enhancement techniques to analyze the brain CT images within Atherosclerosis. Steps of study include: 1. the use of gamma correction to improve contrast of brain CT image; 2. to calculate texture features of middle cerebral artery (MCA) territory after using spatial grey level dependence method (SGLDM); and 3. The use of local binary pattern (LBP) to support texture features usefulness. After principle component analysis (PCA), some meaningful parameter of patient records, texture features, and texture features with local binary pattern were chosen. Finally, use neural network (BPN) to evaluate this system.
This study based on peripheral vascular measurement information to classify cases. 20 normal and 200 abnormal cases were trained, and 14 normal and 57 abnormal cases tested thru back propagation neural network. The accuracy, sensitivity, specificity, and kappa value for to detect Atherosclerosis texture features with/without local binary pattern were 98.59%/94.37%, 98.25%/96.49%, 100.00%/85.71%, and 0.936/0.756, respectively. The system evaluation of results, texture features with local binary pattern have better performance.
A system including SGLDM and LBP for quantizing analysis, and enhancement images for clinician to diagnosis was developed. This multi-function system can improve the accuracy to detect Atherosclerosis in CT images, and avoid the false diagnosis to patients.

並列關鍵字

Atherosclerosis ； Computer tomography ； Spatial grey level dependence method ； Back propagation neural network ； Local binary pattern

參考文獻

[24]黃綠宣，“結合紋理分析與影像增強技術於電腦斷層影像中輔

1185–1187, 1971.

[4] L. Drouet, “Atherothrombosis as a systemic disease.” Cerebrovasc

[5] P. Libby, “Current concepts of the pathogenesis of the acute

coronary syndromes.” Circulation, 104:365–372, 2001.