運用資料探勘技術檢測病患罹患疾病診斷之異常

Using Data Mining Techniques to Detect Careless Diagnosis for Patient Diseases

病患就醫的診斷資料中隱藏著醫療人員的經驗與知識，若能從中找出病患顯示症狀與罹患疾病之間的關聯性，對提升醫療準確性及降低診斷疾病過程中的疏忽，必定可以提供相當有用的助益。本研究以病患每次就醫的診斷資料為探勘資料來源，分別以某一病患的顯示症狀、及罹患疾病為探勘目標，利用資料探勘(data mining)技術中的分群化(clustering)從以下兩方面找出症狀與罹患疾病之間關聯性，以做為檢測病患疾病診斷之異常的依據：一是以病患顯示症狀為一群組的中心點，本研究設計一個分群化方法，將與中心點滿足最小症狀相似度的診斷資料歸屬於群組中，從群組中找出病患顯示症狀最可能罹患疾病，做為檢測病患罹患疾病診斷之異常的依據；二是以病患罹患疾病為一群組的中心點，文中設計一個分群化方法，將與中心點滿足最小疾病相似度的診斷資料歸屬於群組中，從群組中找出病患罹患疾病最可能顯示症狀，做為檢測病患顯示症狀問診之異常的依據。文中根據提出的方法，以南部某一醫學中心病患每次就醫的診斷資料為例，設計與建置一個檢測病患疾病診斷探勘系統。

Medical staff experience and knowledge are hidden in the patients’diagnostic data. If we can find out the correlation between the symptoms and the diseases, it can provide very useful medical diagnosis reference information for improving medical accuracy and educing negligence in diagnosing diseases. This paper uses the diagnostic data of each patient's medical treatment as the source of the mining data, and a diagnostic data contains a patient’s symptoms and diseases. Let a patient’s symptoms and diseases as the targets of mining, respectively. We use clustering techniques to find out the correlation between the symptoms and the diseases from two aspects, and as a basis for detecting careless diagnosis for patient diseases. We develop two mining methods to detect careless diagnosis for diseases from two aspects, respectively. One is the patient’s symptoms as the center point of a group, a clustering method is proposed to assign diagnostic data that meets the minimum symptom similarity to the center point to the group. We find out from the group that the patient is most likely to have the diseases. As a basis for detecting whether the patient with diseases are carelessly diagnosed. The other is the patient’s diseases as the center point of a group, an clustering method is proposed to assign diagnostic data that meets the minimum disease similarity to the center point to the group. We find out from the group that the patient is most likely to show the symptoms. As a basis for detecting whether the patient with symptoms are careless inquiry. According to the proposed methods, we take the patients’diagnosis data in a medical center in southern Taiwan as an example to design and build a diagnostic mining system for disease detection.