應用類神經網路建構腦血管動脈粥狀硬化預測模型

A Predication Model for Carotid atherosclerosis using Artificial Neural Networks

腦血管疾病（簡稱腦中風）一直為台灣的主要死因，因為腦中風的治療與照顧的進步，腦中風的致死率逐漸下降，自1981年的死因首位降至2004年的第三位，但因台灣社會人口的快速老化、腦中風的年齡層發生率變化不大，使得腦中風的患者並未減少，腦中風將仍是台灣中、老年人最重要的身心障礙原因。頸動脈狹窄是導致梗塞性腦中風發生的重要致病機轉，而頸動脈狹窄多是因動脈粥狀硬化所引起，頸動脈狹窄引起腦梗塞部分是因為嚴重狹窄導致腦血流不足，或是因為動脈硬化斑破裂產生的栓子阻塞遠端腦血管。本研究以台北市某教學醫院之健檢資料為母體，抽取大於45歲共7,198位個案數，使用25項與腦血管動脈粥狀硬化相關的變數，比較類神經網路及邏輯斯迴歸兩種模型的預測績效，建構最佳預測模型。最後本研究發現，類神經網路的預測模型優於邏輯斯迴歸的預測模型，為本研究最佳模型，AUC（ROC曲線下面積）為0.841、Sensitivity（敏感度）為77.2%、Specificity（特異度）為74.4%、Accuracy（正確率）為76.8%。

Cerebrovascular disease has been the main cause of death in Taiwan, Stroke treatment and care because of progress, decline in stroke mortality, the first since the death in 1981, fell to third place in 2004, but Taiwan society rapidly aging population, the incidence of stroke in the age group changed little, so did not reduce stroke in patients with cerebrovascular disease will continue to be in Taiwan, the elderly, the most important cause of disability. Carotid stenosis is caused infarction occurred in the important stroke pathogenesis, and carotid artery stenosis is the result of many caused by atherosclerosis, carotid artery stenosis caused by cerebral infarction in part because of severe stenosis led to insufficient cerebral blood flow, or generated because of atherosclerotic plaque rupture in the distal embolic cerebral vascular obstruction. In this study, a medical center in Taipei health check information for the parent, taking a total of 7,198 bit older than 45 years the number of cases, the use of 25 and cerebrovascular atherosclerosis-related variables, Comparison of neural network and logistic regression the forecasting performance of two models to construct the best predictive model. Finally, the study found that the neural network prediction model is superior to logistic regression prediction model, the optimal model for this study, AURC is 0.841, Sensitivity was 77.2%, Specificity was 74.4%, Accuracy 76.8%