- 學位論文
電腦輔助偵測應用於阿茲海默症之系統開發
The Development of Computer-Aided Detection System for Alzheimer's Disease
摘要
失智症盛行率逐年上升，預估 2017(106)年到2061(150)年失智人口比例將有2%升到3%。阿茲海默症現行藥物治療主要在於保留或改善認知功能以延遲疾病惡化，如果治療時機越早則效果越好。在臨床上利用臨床失智評估量表評估出阿茲海默症時已經較為嚴重，利用 CT 影像來檢視是否有萎縮的現象是及早發現的方法。目前醫師以手動的方式評估影像中的萎縮程度較不客觀且費時，故本研究主要運用影像處理之方法來計算腦室與全腦比，並分析失智評估量表(CDR)與腦室與全腦比例之間的相關性，以有效提供醫師之參考。 本論文主要是開發計算腦室與全腦比例之電腦輔助偵測系統幫助醫師分析阿茲海默症患者的嚴重程度。研究步驟主要包含：(1)透過區域成長之影像前處理，去除雜訊並分割出完整的腦組織；(2)藉由影像加強技術進行影像邊緣強化，將腦室以及全腦區分；(3)藉由特徵之設定選取適當影像參數分析，將腦室與全腦面積計算；(4)將腦室與全腦比例參數結果呈現；(5)計算腦室與全腦之比值與CDR相關性。在驗證方面，使用77組影像，其中47組影像作為訓練組，30組影像為測試組，並輔以病人之CDR進行系統效能評估。經過訓練及測試所發展成功之電腦輔助偵測系統，期望所提供之量化參數可輔助醫師評估阿茲海默症嚴重程度。 初步結果顯示：CT影像經由統計分析結果可以得到腦室與全腦比例及CDR分數均為正比關係。在阿茲海默症等級之判讀上，扣除誤判及CDR分數為3之影像，CT影像的系統判斷能力，準確率88%、靈敏度81.82%、專一性92.86%、 KAPPA 0.75，系統應用於阿茲海默症輔助診斷有一定效能。此外，腦部影像手動分割及系統分割重疊率92.17%，系統亦提供醫師使用便利及加快診斷速度之介面。 透過初步結果顯示，藉由影像處理所提供之腦室與全腦比例及CDR分數有正比之關係，本系統可供阿茲海默症早期診斷使用。在未來數據量足夠的情況下，除了提供MRI影像與CT影像評估與CDR之間的關聯性，可以搭配判讀其他症狀以減少誤差率，未來也能減少病患所要做的診斷步驟以降低醫療的浪費。
並列摘要
According to the prediction of the Taiwan Alzheimer Disease Association, over 2% and 3% people will suffer to Alzheimer disease in 2017 and 2061 respectively. The earlier medical treatment of Alzheimer Disease (AD), the better for control AD and its disease progress. In clinical, it is too late after assess the level of AD by Clinical Dementia Rating (CDR). The method to get the ventricles and brain ratio in CT is a better choice. However, the assessment of brain atrophy ratio is not objective and time-consuming manually. An automatic system can provide the ventricles and brain ratio information for doctor is necessary. In this project, we develop the computer aided detection (CAD) system by image processing methods to calculate the ventricles and brain ratio. According to the preliminary study, steps of this project including (1) To remove noise and separate the full area of brain through image preprocessing; (2) to detect the ventricles and brain by edge detection; (3) to calculate the volume of ventricles and brain; (4) to calculate the ratio of ventricles and brain; (5) to calculate the ratio of ventricles and brain and then to find the correlation with CDR. Finally, 80 sets of images were used which including 30 groups of images as the training group and 50 groups of images as the test group, for system effectiveness evaluation by supplemented with the patient's CDR. After trained and verified, we expect our system could provide the useful data to aided diagnosis AD level for doctor. Preliminary results and statistical analysis showed that the ventricles and brain ratio and CDR scores are proportional directly. The CT image diagnostic system except misjudgment and CDR score 3, demonstrates an accuracy of 88%, sensitivity of 81.82%, specificity of 92.86%, and kappa of 0.75. These indicate that the system for the diagnosis of Alzheimer's disease is effective and accurate. In addition, the average segmentation split overlap rate for brain images in manual and developed system method is 93.14%, respectively. The system also provides ease of use interface for physician faster diagnosis. The results in early diagnosis showed that the atrophy ratio of CT and MRI images and CDR scores are highly relatively. In the future, we can make use of CT imaging and MRI imaging as a primary diagnostic tool for early detection and reduce the patients’ need for other diagnostic procedures, thereby reducing medical waste.