腦內微出血之電腦輔助偵測診斷系統的開發

The Development of CAD System for Cerebral Microbleeds

潘家鴻(Chia-Hung Pan)

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(2015年)

https://doi.org/10.6840/cycu201500944

摘要

腦內微出血至今已經被證實與許多神經性疾病具有關聯性，醫師藉由磁振造影的各切面來評估總出血點數量、面積作為診斷依據，故本研究之目的為發展一套電腦輔助偵測系統，減少篩選時間並量化重要的參數以協助醫師對病患往後的觀察與治療。
首先利用區域成長法去除腦組織以外的資訊。第二，可以透過臨界值法的閥值來對具有低訊號特性的微出血點做辨識，並且排除掉不必要的影像資訊。第三，利用以醫師手動圈選的出血點與鈣化點做T-test進行篩選後，找出具有鑑別能力的特徵參數並進行支持向量機的訓練與測試。且對照系統分類結果與組織病理檢查之結果後，可對系統整體的診斷能力進行評估。第四，利用手動點選的方式去除出血點以外的結構組織以防系統運算錯誤。第五，以ROI圈選針對個體或是整體的方式來計算出血點數量與出血面積以估算出體積。本研究利用40張影像並以ROI圈選107個含有腦內微出血點與鈣化點之影像(81個出血點、26個鈣化點)來進行系統訓練與測試，訓練組共採用20張影像，而測試組則為20張影像。
實驗結果顯示: 比較系統與醫師手動圈選交叉相比之出血面積平均誤差百分比為4.98%，總出血點誤差百分比為2.12%。系統針對出血點與鈣化點之特徵分析，所得到結果之訓練組的靈敏度100%、準確度:96.3%、專一性:84.6%、Kappa值:0.94; 測試組所得結果為靈敏度:100%、準確度:98.1%、專一性:92.3%、Kappa值:0.95。
本研究已開發出電腦輔助偵測診斷系統，針對腦內微出血之MRI影像進行影像處理與分析，計算出血點總合、面積與體積，減少判讀影像與篩選的時間並當下提供有效資訊以助醫師藉進行正確的診斷。在未來數據量足夠的情況下且有更好的方法讓系統更加精準，且根據出血點分佈位置以評估與各種神經性疾病的關連性，開發一套全自動的電腦輔助偵測診斷系統可在及時當下提供最正確的有益資訊，相信未來不久將可完全應用於臨床診斷上以提升醫療品質與效率。

關鍵字

腦內微出血 ； 影像處理 ； 磁振造影 ； 電腦輔助偵測診斷系統

並列摘要

The relation between Cerebral Microbleed (CMB) and neurological diseases has been verified, and the physicians calculate the number and surface areas of micro bleeding points by means of diagnosing MRI images. Thus, the purpose of this research is to develop a computer-aided detecting system (CAD) for reducing the time of diagnoses and quantifying the significant parameters during diagnosis.
First of all, we utilized region growing method to get rid of the information beyond cerebral tissues. Secondly, we identified the low frequency micro bleeding points by the threshold method, and then manually eliminate extra image information. Third, after the physicians manually selected the bleeding and calcified points, we figured out the significant parameters through T-test and support vector machine (SVM), furthermore, compared the result of our CAD system with pathological reports to evaluate the performance of this CAD system. Fourth, we manually removed the tissues beyond the bleeding points for avoiding the system errors. Finally, the volume of bleeding points was evaluated by the means of stacking up all the surface areas of bleeding points together by region of interest (ROI) method. In this study, we use 40 images which including 107 ROI points (81 bleeding points and 26 calcified points) were used to evaluate the performance of our CAD system and 20 images and 20 images were used as train set and test set.
The result shows that the differences between system selection and physician manual selection as below: the error rates of the surface areas are 4.98%. The system to features analysis for cerebral microbleed and calcified points and get the result as: for train set the sensitivity, accuracy, specificity, and Kappa value is 100%, 96.3%, 84.6%, and 0.94; for test set the sensitivity, accuracy, specificity, and Kappa value is 100%, 98%, 92.3%, and 0.95, respectively.
To sum up, the developed CAD system in this study can process and diagnose the MRI images with CMB; in addition, calculating the number, surface area and volume of bleeding points are available in this CAD system as well. As a result, reduced the diagnoses time and provide physicians with objective information, and also help doctors to diagnose more precisely. In the future, we hope there are more MRI images that allow us to improve the preciseness of the CAD system, and also find the correlation between the neurological diseases and the position of bleeding points. Moreover, we also look forward to decorate the CAD system become a fully automatic one and provide the reliable information immediately. We believe that this technique will be applied in clinical medicine in the near future for improving the quality and efficiency of medicine.

並列關鍵字

Cerebral Microbleed ； Image processing ； Magnetic Resonance Imaging ； Computer aided Detection System

參考文獻

[1]行政院衛生署網址

[28]張嘉麟,以影像處理技術評估腦內微出血於電腦斷層影像之研究，中原大學醫學工程系碩士論文，中壢，民國102年七月。

[39]吳浣樺、林炳文、蘇五洲、陳海雯，”組織切片證實為胰臟癌的放射治療效果”，放射治療與腫瘤學，2004, (11):81-89

[40]賴宣宇，”電腦輔助診斷系統於乳房腫瘤X光與超音波影像之整合應用”，中原大學醫學工程系碩士論文，中壢，民國94年七月

[2]Poels, M.M., Vernooij, M.W., lkram, M.A.mHofman et al. “ Prevalence and risk factors of cerebral microbleeds: an update of the Rotterdam scan study” Stroke 2010; 41: S103–S106.