- 學位論文
使用對比調整演算法與紋理特徵的肺栓塞偵測系統
Pulmonary embolism detection system using the contrast enhancement algorithm and texture feature
摘要
傳統的肺動脈栓塞檢測需透過專業的醫師判斷。而近年CT影像技術進步讓影像的品質以及影像張數跟著增加，但這也無形中增加了醫師診斷時的疲勞。因此我們提出了一套使用對比調整演算法與紋理特徵的自動肺栓塞偵測系統。 首先我們利用三次曲線對肺部影像作對比增強，其目的是用以將肺動脈及肺葉中的細支血管突顯出來。接著將影像中的紋理以及亮度當作搜尋肺栓塞的特徵，在紋理的部分我們使用Laws’ Mask來抽取，而在亮度部分我們則使用對比增強指標MAD來抽取，最後透過類神經網路學習並辨識出肺栓塞的區域。從實驗結果顯示肺部CT影像在經過三次曲線補償後，肺動脈血管及細支血管清晰可見。最後利用10個資料集中共460張影像作測試，從測試結果中發現我們的系統辨識率已達到98%。
並列摘要
The traditional detection of PE needs to rely on the professional judgments of physicians. With advances in CT technology both of the image quality and the number of images are improved, but it is also virtually increase the fatigue when physician in the diagnosis. Therefore, we propose a an pulmonary embolism detection system to reduce the fatigue when physician in the diagnosis. First, we use a cubic contrast enhancement method to enhance branch vessels contrast in the pulmonary artery and lobar lung. It can highlight the PE in the branch vessels and help doctor to understand the PE degree. Next the texture and brightness feature in the image as the search characteristics of pulmonary embolism, in the part of the texture, we use Laws’ Mask to extract texture feature, and use Modification Average Distance to extract contrast feature. Finally we use the neural network to recognize pulmonary embolism with all features, extracted features. The experimental results showed that lung CT images after a cubic contrast enhancement method, the pulmonary blood vessels and branch vessels become clearly visible. Finally, we use 460 image obtained from 10 data sets to perfrom testing. We understand that the recognition rate of our proposed system is 98%.