應用人工智慧於孕期憂鬱症之評估研究

A Study of Applying Artificial Intelligence for the Assessment of Maternal Depression

王冠智(Kuan-Chih Wang)

指導教授 ： 張俊郎

國立虎尾科技大學/管理學院/工業工程與管理研究所/碩士(2010年)

摘要

近年來自殺一直為台灣十大死因之一，根據研究統計有87%的自殺死亡個案生前患有憂鬱症，而患有憂鬱症之患者有15％死於自殺，目前憂鬱症已經是一種常見的心理疾病，以目前醫療而言，憂鬱症是無法自我判斷，必需藉由專業醫師的診斷，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，便顯其重要性。

一般女性在懷孕時期雖受到較完善的照顧，但因為生理與心理變化，容易引起心情上的起伏，導致罹患孕期憂鬱症，若沒有及早發現治療，則會併發產後憂鬱症，嚴重者甚至會傷害自己與嬰兒。本研究目的藉由人工智慧中之決策樹與類神經網路來建構孕期憂鬱症輔助診斷系統，找出影響孕期憂鬱症之危險因子與憂鬱症狀，提供醫師未來進行醫療診斷時，能有效減少人為判斷失誤。

研究結果顯示，應用倒傳遞類神經網路預測，是否罹患孕期憂鬱症之準確度為83.33％，醫療評估用指標，接受器操作曲線面積為0.819，研究發現因子中以工作情況、教育程度、籍貫、懷孕週數、其他疾病、睡眠時間，影響孕期憂鬱症最為顯著，而應用決策樹歸納出兩條憂鬱症規則，其準確度為86.67％，接受器操作曲線面積為0.861，研究發現憂鬱患者中較容易出現不開心到哭與想要傷害自己之症狀，本研究建立之預測系統可提供給醫師，作為診斷孕期憂鬱症之參考，幫助容易罹患孕期憂鬱症之高危險群，提前診斷與防治，對於醫師在臨床診斷確實有實質上之助益。

關鍵字

孕期憂鬱症 ； 資料探勘 ； 倒傳遞類神經網路 ； 決策樹 ； 接受器操作曲

並列摘要

Suicide has been one of the ten major leading causes of death in Taiwan recently. According to the statistics, 87% of the cases suffered from depressions prior to their death, while 15% of the patients with depressions died from committing suicides. Depression is a very common emotional problem at the time being. Since one cannot determine on his/her own and professional physicians’ diagnosis and assessment is crucial in determining whether one has depression or not, how to effectively construct a set of comprehensive auxiliary diagnosis system to help doctors in their clinical assessment is becoming more and more important.
Although women receive more cares during their pregnancy in general, due to the physical and psychological changes, they are more prone to suffer emotional swifts leading to prenatal depressions. When left alone without early treatment and intervention, it is very likely for these women to have postpartum depressions where in the worst case, they harm themselves and the infants. In this study, we aimed to use the decision tree and artificial neural network of the artificial intelligence to construct a diagnosis system in screening prenatal depressions. The objective is to find out the risk factors and symptoms that affect prenatal depressions for physicians to use as a reference, hoping to effectively reduce human judgmental errors in future medical assessment.
The results in this study indicate that when using BPN to predict, accuracy of the risk factors for prenatal depressions is 83.33%, and the ROC area of the health care evaluation index is 0.819. Among all factors, the most significant ones include employment status, education background, birth place (city), pregnancy weeks, other diseases, and hours of sleep; these factors have great impact on prenatal depressions. Furthermore, with the use of decision trees, two rules of depression symptoms are generated at 86.67% in the accuracy and 0.861 for the ROC area. The results also reflect that depression patients are more likely to feel like crying and harming themselves when they are unhappy. The predictive system established in this study can provide a reference for physicians to diagnose prenatal depression, help identify the group at high risk so preventative measurements and treatments can be provided and thus practically beneficial in enhancing the clinical diagnosis accuracy.

並列關鍵字

Prenatal Depression ； Data-Mining ； Back-Propagation Neural Network (BPN) ； Decision Tree ； Receiver Operating Characteristic Curve

參考文獻

11.蕭雅尤、林名男、劉昱志、謝昌成（2009），「產後情感性疾患與產後憂鬱症危險因子」，基層醫學，24卷，4期，頁126-130。

5.徐禎美、陳彰惠（2008），「孕期憂鬱的盛行率和預測因子」，實證護理，4卷，2期，頁149-156。

9.楊子慧、李明濱、張秀如（2005），「產後憂鬱症」，北市醫學雜誌，2卷，7期，頁598-604。

8.陳淵渝、林志強、諶立中（2004）．「女性憂鬱症的治療驅勢」．台灣醫學，8卷，2期，頁253-257。

27.Heh, S.S. (2001). “Validation of the chinese version of the edinburgh postnatal depression scale: detecting postnatal depression in taiwanese women”, Nursing Research China, Vol. 9, No. 2, pp.105-113.