從X光影像預測肺結核纖維化等級

Abstract

肺結核在全球仍然是主要死亡原因之一。若超過兩到三星期沒有治療，在臨床上或是傳染病學上是危險因子。因為結核病可能涉及多個器官系統，如呼吸、心臟、中樞神經、肌肉、胃腸和泌尿生殖系統。因此即時診斷肺結核是非常重要的，若延誤治療，會造成嚴重的發病率以及結合分枝桿菌的傳播，甚至死亡。行政院衛生署疾病管制局所提供的肺結核診斷流程圖中，對疑似肺結核的病人，如咳嗽2~3星期，都進行胸部X光檢查。從肺結核診斷流程圖中可以知道，疑似肺結核病人診斷的第一個步驟就是胸部X光檢查；若以確認診斷是肺結核患者，都有不正常之胸部X光影像。因此，從這邊可以知道胸部X光檢查在疑似是肺結核或是肺結核的診斷上有極重要的角色，故疑似是肺結核或是肺結核病人都需要做正面胸部X光檢查。而且做胸部X光檢查效率快，不必花費大量時間和金錢。因此有一套肺結核胸部X光分級預測系統以及找出如何分辨肺結核等級的特徵，可讓醫師在第一時間更加了解病人狀況給予初步的診斷與藥物等級的治療。 在目前的相關文獻中，我們是第一個提出一套肺結核X光影像分級預測系統的新方法，可以對肺結核患者的胸部X光影像進行分級預測。而肺結核胸部X光影像擷取肺部程式，在目前的相關論文中有人研究過，但X光影像所擷取出來的肺部並非完整，品質上有改善的空間。因此，我們提出的肺部擷取程式，能針對胸部X光影像將肺部完整取出。 在本實驗結果中，肺結核胸部X光影像是依據行政院衛生署署立雙和醫院胸腔內科醫生標記的等級去分成3類，經過ROI+GrowCut的影像前處理將肺的遮罩完整擷取出來，之後再將肺結核胸部X光影像肺部遮罩和原始肺結核胸部X光影像結合 (Mask the Original Image) 或是經過直方圖等化的影像結合 (Mask the Histogram Equalization Image)，即完成影像前處理的部分，並利用本實驗室用Matlab GUI所撰寫的特徵值萃取(feature extraction)程式，最後使用繼承式雙目標基因演算法 (Inheritable Bi-objective Genetic Algorithm, IBCGA) 搭配支持向量機 (Support vector machine, SVM)以及使用繼承式雙目標基因演算法(Inheritable Bi-objective Genetic Algorithm, IBCGA)搭配K-Nearest Neighbor (KNN)，在大量的參數中挑選較少的參數並得到最大的適應性。IBCGA_SVM實驗結果中的準確率最高為60.68%，從2277個特徵中挑選了15個特徵，而IBCGA_KNN實驗結果中的準確率最高為59.72%從483個特徵中挑選了19個特徵。

Tuberculosis (TB) remains one of the major causes of death from a single infectious agent worldwide. Tuberculosis may involve any of a number of organ systems , like the respiratory, cardiac, central nervous, musculoskeletal, gastrointestinal, and genitourinary systems, and timely diagnosis of the disease is paramount. The potential risks associated with diagnostic delays are significant, and include prolonged morbidity, increased mortality, and Mycobacterium tuberculosis transmission. Centers for Disease Control provide tuberculosis diagnostic flowchart, patients with suspected tuberculosis, like cough for 2 to 3 weeks into the conduct chest X-rays. From the tuberculosis diagnostic flowchart can know, suspected diagnosis of tuberculosis patients in the first step is chest X-rays. If confirm the diagnosis of tuberculosis patients have abnormal chest X-ray images. Therefore, from here can know that the chest X-rays have a very important role in the diagnosis of tuberculosis or suspected tuberculosis, suspected tuberculosis or tuberculosis patients need to do a positive chest X-ray. And the efficiency of the chest X-rays do not have to spend a lot of time and money. Therefore have a tuberculosis chest X-ray classification prediction system and figure out how to distinguish between tuberculosis grade features, Allow doctor understand the patient status and preliminary diagnosis and drug levels of treatment. We are the first new method of prediction system of a tuberculosis X-ray image classification, classification prediction of tuberculosis in patients with chest X-ray image. Tuberculosis chest X-ray image capture lung program was studied in the current paper, but the lung X-ray image capture is not complete, the quality can improvement. Therefore, our proposed lung segmentation program, for chest X-ray image the lungs complete remove. The results of this experiment, tuberculosis chest X-ray image divided into three categories based on Taipei Medical University-Shuang Ho Hospital and Pulmonary Medicine doctor. After ROI + GrowCut image processing mask integrity of the lung to retrieve. Then, tuberculosis chest X-ray images lung mask combined with chest X-ray images of the original tuberculosis image or chest X-ray images of the histogram equalization tuberculosis image, to complete the part of image pre-processing and use feature extraction program. Finally use Inheritable Bi-objective Genetic Algorithm (IBCGA) and Support vector machine (SVM). Selection of fewer parameters in a large number of parameters and maximum flexibility and accurate the largest IBCGA_SVM experimental results in the highest rate of 60.68%.