近年來,資料探勘技術已趨成熟,也變得越來越流行,它能為來自於不同領域的大量數據,進行大量數據資料的分析,分類和有效預測的重要工具。而對於醫療方面的議題,如果能夠提早檢測出相關病症或是發現可能疑似案例,將是十分重要的,同時將增加成功治療的機會。這種檢測通常是設計成一個二元分類的問題,因此過去已有許多分類方法被用於檢測潛在的醫療問題。 睡眠呼吸中止症候群(Sleep Apnea Syndrome, SAS)是一種在睡眠中呼吸暫停及低通氣反復發作的疾病。SAS在亞洲中年男子盛行率介於25%至27%之間,而對亞洲的中年婦女來說盛行率則介於10%到16%之間。阻塞性睡眠呼吸中止症(Obstructive sleep apnea, OSA)是睡眠呼吸中止症候群中最常見的一類。患有 OSA的人會增加心臟疾病及睡眠相關事故的風險,甚至嚴重的 OSA患者,由於缺乏足夠的氧氣攝入量,因而可能導致死亡。 如需診斷OSA則必須進行夜間睡眠多項生理監測儀(Polysomnography, PSG)的測試。然而,PSG診斷只限於某些城市地區,加上它非常費時且昂貴,必須有專業的技術人員在患者睡眠中記錄多項生理訊號。 最近出現的一些OSA相關輔助診斷的研究,但這些前人文獻中分別只用了83、86和110個實際案例及採用複雜的臨床特徵來構建預測的模型,正確率成效有限,而且採用複雜之特徵,對於建立之預測模型進行普及化將有較高的挑戰。本論文研究的目標是要根據一大型臨床數據庫,使用較少及非侵入性的檢查,運用7種決策樹演算法找到診斷OSA最佳的分類模型,足以區別高低風險群組的病患,使得高風險病患能及早進行進一步的檢查,低風險病患則避免PSG檢查,造成醫療資源的浪費。
In recent years, data mining technologies have matured and become more popular, providing a useful tool to analyze, classify and predict large amounts of data gathered from various fields. Early detection of medical problems is important to increase the chance of successful treatment. Such detection is often formulated as a binary classification problem. Various classification methods have been developed for the detection of a potential medical problem. Sleep Apnea Syndrome (SAS) is a condition characterized by repeated episodes of apnea and hypopnea during sleep. The prevalence of OSA ranged from 25% to 27% in middle-aged men and from 10% to 16% in middle-aged women in Asian groups. Obstructive sleep apnea (OSA) is the most common category of SAS. OSA increases people’s risk of having heart diseases and sleep related accidents. Severe cases of OSA syndrome might cause patient deaths due to lack of sufficient oxygen intake. OSA is diagnosed with an overnight sleep test called polysomnography (PSG). However, the availability of PSG evaluation is relatively limited to urban areas. Moreover, it is very time-consuming, expensive and tedious task consisting of expert visual evaluation all ten minutes pieces of approximately eight-hour recording with a setting of many channels. There are some studies related to diagnosis of OSA in literature, but only 83, 86 and 110 patients and complex clinical features were used to building their models, and their performance had an accuracy of 74.2%~92.5%. The goal of this thesis is to find the best classifier of decision tree model using lower and noninvasive examination features on the diagnosis of OSA based on a larger clinical database. It will reduce overhead costs and examination duration for candidates or patients. According to the general physical check-up, the results can provide useful information to distinguish a patient who is the high risk patient or the low risk patient. Therefore, the decision result can help many low risk patients to avoid referring more complicated advanced examinations, it also can reduce medical overhead. Based on our proposed prediction model, the experiment results have shown that our model is more simple, accurate and reliable.