以影像處理技術評估腦內微出血於電腦斷層影像之研究

The Application of Image Processing Methods for Evaluating Cerebral Microbleeds in Computed Tomography Images

張嘉麟(Chia-Lin Chang)

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(2013年)

https://doi.org/10.6840/cycu201301039

摘要

未施打顯影劑之電腦斷層影像無法及時於偵測腦內微出血提供詳細的臨床診斷資訊，且臨床醫師對於有腦內微出血患者所採取的治療方式必須仔細評估，以避免因進行溶栓或抗凝藥物治療時，導致患者有出血轉化的風險。因此，本研究之主要目的是以影像處理技術評估未施打顯影劑的電腦斷層影像上微出血等相關資訊並分析其紋理特徵，找出具有統計上的顯著差異的參數，進而提供未來進一步研究的參考。
本研究先利用影像前處理擷取大腦組織，再分別透過對數轉換、Gamma轉換與直方圖等化等三種方法增強腦組織影像，並比較其於微出血病灶之突顯效果。接著進行病灶與非病灶之ROI手動圈選，再分別進行鄰近灰階相依矩陣與灰階共生矩陣等2種紋理分析，並統計10個病灶ROI與20個非病灶ROI之紋理特徵參數。最後，提供6張原始影像與6張增強後的影像，並透過本研究設計之問卷請專科醫師進行病灶辨識評估。
研究結果顯示，影像增強方法以直方圖等化的效果較佳。紋理特徵參數統計後結果顯示當病灶與非病灶的尺寸相近時，Contrast_0、Homogeneity_0、Homogeneity_90與Homogeneity_135等4個參數對病灶分類具有統計上的顯著差異；而當病灶與非病灶的尺寸不相近時則是以NNU、SM、EN、Homogeneity_0、Homogeneity_90與Homogeneity_135等6個參數對病灶分類具有統計上的顯著差異。另外，部分腦組織因陳舊性出血而導致圈選ROI時面積會產生誤差，其誤差率最大達22.43%；因此排除此誤差後，紋理特徵參數之統計結果顯示Energy_135、Entropy_0、Entropy_45、Entropy_90與Entropy_135等5個參數對病灶分類具有統計上的顯著差異。本研究增強後的影像提供給專科醫師偵測微出血病灶之最佳偵測正確率可達85.71%。
總體而言，本研究以直方圖等化法增強未施打顯影劑的電腦斷層影像可初步偵測出腦內微出血，並找出具有統計上顯著差異之紋理特徵參數，提供未來微出血輔助偵測研究有正面的幫助。

關鍵字

電腦斷層影像 ； 微出血 ； 鄰近灰階相依矩陣 ； 灰階共生矩陣

並列摘要

Non-contrast Computed Tomography (NCCT) images cannot provide detailed information on the clinical diagnosis in detecting cerebral microbleeds. In order to avoid risk of hemorrhagic transformation during thrombotytic or antithrombin therapy, clinicians must be carefully to assess the treatment for patients with cerebral microbleeds. Therefore, the purpose of this study was to assess cerebral microbleeds on non-contrast CT images with image processing techniques, and find statistically significant parameters by texture analysis, and move further to provide a reference of future research.
In this study, firstly, image pre-processing was used to extract brain tissue, and then brain images enhanced through logarithmic transformation, Gamma transformation and histogram equalization, also compared it to highlight the effect of microbleeds lesions. Followed by ROI manually selecting of the lesion and non-lesion, and then neighboring gray level dependence matrix and gray level co-occurrence matrix were used as texture analysis. Furthermore, 10 regions of interest (ROIs) as lesions and 20 ROIs as non-lesion were selected, and extracted the 30 texture parameters by statistics. Finally, 6 original images and 6 enhanced images were provided to assess the lesion through questionnaire by specialist physicians in this study.
The results show that histogram equalization was the best enhancement in this study. In texture analysis part, the results of statistics show that Contrast_0, Homogeneity_0, Homogeneity_90, and Homogeneity_135 were statistically significant in lesion classification when the size of lesion was similar to non-lesion; the NNU, SM, EN, Homogeneity_0, Homogeneity_90, and Homogeneity_135 were statistically significant in lesion classification when the size of lesion were not similar to non-lesion. In addition, some brain due to old hemorrhage caused by the ROI selected would produce error, the maximum of error rate was 22.43%. Therefore, after excluding this error, the results of statistics show that Energy_135, Entropy_0, Entropy_45, Entropy_90, and Entropy_135 are statistically significant in lesion classification. This study provided specialist physicians enhanced images as microbleeds lesions selecting and the best accuracy rate of detection was 85.71%.
In summary, this study could detect cerebral microbleeds on non-contrast CT images with histogram equalization and find statistically significant texture parameters. This study is positively helpful to provide future research of aided detection as a reference.

並列關鍵字

Gray Level Co-occurrence Matrix (GLCM) ； Non-Contrast Computed Tomography (NCCT) ； Neighboring Gray Level Dependence Matrix (NGLDM) ； Cerebral Microbleeds (CMBs)

參考文獻

[14] 劉詩維, 羅彬峯, 簡甄余, "利用磁敏感加權影像來偵測腦部微出血的評估與分析," Taiwanese Journal ofApplied Radiation and lsotopes, vol. 5, 2009.

[24] 曾林維, "利用紋理分析於電腦斷層影像偵測急性缺血性中大腦動脈中風," 碩士, 中原大學, 台灣, 2006.

[25] 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助診斷急性缺血性腦中風," 碩士, 中原大學, 台灣, 2008.

[26] 胡學成, "利用紋理特徵於電腦斷層影像預測急性缺血性腦中風中大腦動脈出血轉化之輔助診斷系統," 碩士, 中原大學, 台灣, 2011.

[27] 王裕翔, "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," 碩士, 中原大學, 台灣, 2011.