建構頸動脈血管近遠端內膜中層厚度預測模式之研究

台灣地區罹患心腦血管慢性疾病的比例一直居高不下，其中動脈粥狀硬化現象會增加罹患心腦血管疾病的風險，亦是現代台灣人健康最大的威脅。因此，如何有效建構一套完善的輔助診斷系統，以協助醫師的臨床診斷，便愈顯其重要性。針對動脈粥狀硬化的有效臨床診斷指標普遍採取頸部超音波觀察頸動脈內膜中層厚度(Intima Media Thickness, IMT)的變化加以評估，內膜中層厚度增加表示有早期粥狀動脈硬化之徵兆，故近年來已將中層厚度技術視為診斷動脈粥狀硬化與心腦血管疾病的有效量化指標，亦是心腦血管疾病的獨立預測危險因子。唯有控制影響動脈硬化與血管疾病的潛在危險性因子，才能有效趨緩動脈硬化與血管疾病之併發症的延展性。本研究運用人工智慧(Artificial Intelligence, AI)技術，發展一套輔助診斷模式，收集個案醫院民眾之健檢數據，建構近遠端內膜中層厚度預測模式系統，透過倒傳遞類神經網路(Back Propagation Neural Network, BPN)與粒子群最佳化演算法(Particle Swarm Optimization, PSO)決定案例式推理(Case Based Reasoning, CBR)系統各變數因子的最佳權重。研究結果顯示，近端與遠端預測系統準確度分別高達為87.931%與90.517%，且醫學評估指標之受試者作業特徵曲線(Receiver Operating Characteristic Curve, ROC曲線)面積分別為0.897與0.852，使得運用粒子群最佳化演算法求解權重結合案例式推理系統有最佳的預測能力與診斷價值，其影響近端內膜中層厚度增厚現象之顯著因子有：心血管疾病史、睡眠習慣、糖化血色素、舒張壓、膽固醇、工作狀況、高密度膽固醇；而影響遠端內膜中層厚度增厚現象之顯著因子有：高血脂疾病史、低密度膽固醇、空腹血糖、高密度膽固醇、憂鬱量表、舒張壓、三酸甘油脂。本研究提供醫師未來進行醫療服務時，作為全方位的早期評估及判斷，同時可擬定早期防範的對策，降低發生的可能性，對於醫師之臨床診斷將有實質之助益。

關鍵字

動脈粥狀硬化；頸動脈內膜中層厚度；倒傳遞類神經網路；粒子群最佳化演算法；案例式推理；受試者作業特徵曲線

並列摘要

A high proportion of people suffer from chronic diseases of cardiovascular and cerebrovascular origin in Taiwan. Atherosclerosis will increase the risk of cardiovascular and cerebrovascular diseases, it is also the greatest threat to health for Taiwanese at present. Therefore, constructing an effective and comprehensive set of auxiliary diagnosis systems to help doctors with clinical diagnosis has become more important. Adopting ultrasonography technology to observe the condition of carotid intima-media thickness is an effective clinical technique to diagnose atherosclerosis. Since intima-media thickness can show signs of early atherosclerosis, the measurement result of intima-media thickness technology has been accepted as a reliable and valid quantification of early atherosclerosis, cardiovascular and cerebrovascular diseases in the last decade. Moreover, it is the main predictive risk factor of cardiovascular and cerebrovascular diseases. Thus, if physicians can control its effect upon the potentially dangerous factors of atherosclerosis, cardiovascular and cerebrovascular diseases, they will be effective in reducing the disease complications of atherosclerosis, cardiovascular and cerebrovascular. This study applies artificial intelligence technology to develop some auxiliary diagnosis models. Using a collection of health examination data from hospital cases to set up a predictive model system for the proximal part and distal part of intima-media thickness, we applied the back propagation neural network and particle swarm optimization to search the optimal weighting of the index for a case based reasoning system. The result shows the proximal part and distal part prediction system accuracy rate are up to 87.931% and 90.517%. Further, the receiver operating characteristic curve areas are 0.897 and 0.852 according to medical evaluation. We contend the model of particle swarm optimization search and the optimal weighting of the index with a case based reasoning system have the best predictive ability and diagnostic value. The results of this study find the effect of the thickening phenomenon of proximal part intima-media thickness are cardiovascular diseases, sleep habits, glycated hemoglobin, diastolic blood pressure, cholesterol, working conditions, and high density lipoproteins cholesterol. These variables are the most significant ones. Further, the effect of the thickening phenomenon of the distal part intima-media thickness are hyperlipidemia diseases, low density lipoproteins cholesterol, fasting plasma glucose, high density lipoproteins cholesterol, depression, diastolic blood pressure, and triglyceride. These variables are the main ones. This study can help doctors in making accurate and early assessments. Simultaneously, it is helpful as a diagnostic reference point for doctors, so patients can take the proper treatment for reducing the possibility of occurrence.