應用人工智慧於新生兒聽力障礙輔助鑑別診斷之研究

Using Artificial Intelligence for Assistance in Differential Diagnostic of the Newborn with Impaired Hearing

陳啟浩(Chi-Hao Chen)

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(2008年)

摘要

應用人工智慧於新生兒聽力障礙輔助鑑別診斷之研究

研究生：陳啟浩
指導教授：張俊郎 博士

國立虎尾科技大學工業工程與管理研究所

摘 要

據研究顯示每1000個新生兒中就會有1~6個有聽力障礙，若以聽閾30dB HL來定義聽力障礙，則出現率為每1000個新生兒中就有約2.5個是有永久性雙側聽力障礙。但如果是較嚴謹的聽閾50dB HL才認為有聽力障礙，則其出現頻率會降為每1000個新生兒中才有1個是聽力障礙。如果包含單側聽力障礙，則聽力障礙的出現率會增加1/3至1/2倍。而台灣地區的研究結果顯示，新生兒先天性雙側聽力障礙的比率約為1000人中有1.3人，單耳聽力障礙的比率為1000人中有3.8人。

聽力損傷影響所及往往不僅止於聽覺，若未能早期發現早期治療通常會連帶的影響到患者之語言、認知、學業成就等等，甚至有可能造成患者在未來的社經地位和情緒的發展。此外若以預防醫學的角度著眼，建立一套可以準確診斷出新生兒是否有聽力障礙的疾病預測模型，早期發現早期治療以避免醫療資源的浪費，確實有其必要性。

本研究以個案醫院之新生兒資料庫為研究對象，透過類神經網路及決策樹演算法進行預測模型的建構，研究結果顯示決策樹預測模型較優於類神經網路預測模型，其訓練準確率和測試準確率分別為95.86%及96.43%，本研究建構之模型對於醫師臨床之診斷及新生兒聽力障礙之早期療育將有實質的助益。

關鍵字：新生兒聽力障礙、決策樹、類神經網路

關鍵字

新生兒聽力障礙 ； 決策樹 ； 類神經網路

並列摘要

Using Artificial Intelligence for Assistance in Differential Diagnostic of the Newborn with Impaired Hearing

Student：Chi-Hao Chen
Advisor：Dr. Chung-Lang Chang

Institute of Industrial Engineering and Management
National Formosa University

ABSTRACT

The research shows that there are one to six neonates would have the hearing loss problem among 1000 neonates. If we use hearing thresholds 30dB HL to define hearing loss, there are 2.5 infants would have permanent loss to bilateral hearing among 1000 neonates. However, if we use more strict definition, 50dB HL, to judge, the probability of having hearing loss would be lower to 1/1000. If including unilateral hearing loss, the probability of hearing loss would increase about one-third to one-half times. The results conducted from Taiwan show that the ratio of neonatal bilateral congenital hearing loss is 0.0013, and the ratio of unilateral hearing loss is 0.0038.
The effects of hearing loss do not just only hurt hearing. If the patient did not accept therapy in early days, it would affect the patient’s language, cognitive ability, academic performance, even development of socialization and morale. In addition, from a point of view of Preventive Medicine, set up a disease forecasting model that could diagnose accurately whether newborn baby has hearing loss or not is prominent. “Discover early, cure early” could prevent from medical squander.
The subjects were the neonates in database of case hospital. We proceeded to construct the forecasting models by Neural Network (NN) and Decision Tree. The result shows that the forecasting model of decision tree is superior to the model of Neural Network, and degree of accuracy and test are 95.86% and 96.43% separately. The models that this study constructed are beneficial to the clinical diagnosis.
Keyword：Newborn hearing loss、Decision Tree、Neural Network

並列關鍵字

Newborn hearing loss ； Decision Tree ； Neural Network

參考文獻

參考文獻

1.王俊傑、沈炳臣、毛彥傑、朱明輝、鐘裕亮、王俊勝、張永鵬、康淵 (2008)，「結合時序分析與類神經網路應用於旋轉機械之故障診斷」 ，先進工程學刊，3卷，1期，頁75-83。

2.朱明輝、張永鵬、王俊傑、彭善謙、康淵 (2005)，「類神經網路使用多重振動信號於馬達故障診斷」 ，中原學報(自然科學及工程系列)，33卷，3期，頁 339-350。

3.江衍銘、張斐章 (2006)，「以遙測資訊建立類神經網路雲分類系統於雨量推估及洪水預測」 ，農業工程學報，52卷，4期，頁39-47。

4.池文海、陳瑞榮、張賜福 (2007)，「運用類神經網路於台電服務品質滿意度之探討－以宜蘭縣住宅用戶為例」 ，中華管理學報，8卷，1期，頁51-69。