非感染性皮膚水泡病及自體免疫疾病之跨領域臨床診斷決策支援系統

Cross-domain Clinical Diagnostic Decision Support System for Non-Infectious Blistering Diseases and Autoimmune Diseases

張英睿(Ying-Jui Chang)

指導教授 ： 李友專

臺北醫學大學/醫學院/醫學資訊研究所/碩士(2001年)

摘要

非感染性皮膚水泡病及自體免疫性疾病是皮膚科及免疫風濕科領域中常見且重要的疾病，此兩類疾病在臨床症狀及實驗室檢查方面，有許多共通性及重疊性，此亦造成診斷此類疾病的不確定性。皮膚科醫師及免疫風濕科醫師常需面對跨領域的病人，如能給予快速且正確的診斷，則對於治療和預後有重大的影響。 臨床診斷決策支援系統(或稱專家系統)可以幫助臨床醫師面對不同臨床症狀、檢驗數據的病人時，作出最適切的診斷建議及鑑別診斷，以便給予診斷以及後續的治療方針。對於醫學生而言，臨床診斷決策支援系統亦可以作為學習的輔助工具。本研究計劃建構一跨領域知識庫，並利用已開發的web-based機率性推論之診斷決策支援系統架構，以multi-membership Bayesian formulation的知識表現法，並推導出cross-domain Bayesian formulation跨領域貝氏運算，以建構一跨領域臨床診斷決策支援系統，嘗試解決以往之醫學診斷決策支援系統應用範圍局限於狹小的特定科別領域的缺點，幫助醫師能對病人病情有全面性考慮與了解；對醫學生與醫師診斷思考邏輯學習過程能有實際的全面性訓練，其他科別領域(如家庭醫學科、急診醫學科、一般內科)的醫師也能由此系統得到診斷及繼續教育上的實質幫助。 欲建造跨領域醫學診斷決策支援系統，其主要步驟可分為(1)知識表現法之建立，(2)Web-based介殼系統(含推論引擎及使用者介面程式)之建立，(3)知識工程之過程及醫學知識庫之建立，(4)知識有效化及系統評估。其中之(1)、(2)項為先前已開發完成之”機率性皮膚病理診斷決策支援系統”架構；(3)知識工程之過程及知識庫則必須另行擴充建立。除了原有的11個全身性水泡疾病框架，90個診斷條件或疾病項目，共有171個代表Apriori、TPR、FRP的機率數據外，自體免疫疾病共有6個疾病框架，98個診斷條件或疾病項目，共有78個代表Apriori、TPR、FRP的機率數據被建立。經由推導得出的跨領域貝式運算公式，亦可以將不同領域間之Apriori、TPR、FRP的機率數據互相轉換，此亦是決策支援系統發展上的一大突破。 從學術文獻期刊中取得20個臨床病例作為測試自體免疫疾病知識庫的測試對象；另取得10個臨床病例作為測試跨非感染性皮膚水泡病及自體免疫性疾病領域知識庫的測試對象。以本系統所計算出的診斷，準確率達90%(9/10)。參與計劃的領域專家認為此系統符合可用性。 雖然診斷決策支援系統的發展已有近四十年的歷史，也有許多系統被開發出來，然而這些系統均只能在單一的電腦上操作，雖然已有部分的網際網路版本被開發，卻只能在特定區域的電腦使用。至今為止，就吾人所知本研究是唯一的web-based的跨領域診斷決策支援系統。

關鍵字

皮膚水泡病 ； 自體免疫疾病 ； 決策支援系統 ； 機率性推論 ； 跨領域貝氏運算 ； 知識工程 ； 知識庫 ； 網際網路

並列摘要

Non-infectious blistering diseases and autoimmune diseases are not rarely seen in the dermatology and rheumatology fields. There share some common clinical findings and laboratory test results, which yield certain degree of uncertainty. Dermatologists and rheumatologists often cope with patients in multi-domain, if they could make fast and correct diagnosis, it will influence the treatment strategy and outcome prognosis. Clinical decision support system can help medical students and clinicians in learning diagnosis and treatment decisions for patient care. Capitalizing on development in artificial intelligence and decision science, clinical decision support systems tend to emulate the decision-making process of human medical experts. We used multi-membership Bayesian formulation and deducted a “cross-domain Bayesian formulation” to develop our system. In order to construct clinical diagnostic decision support system, there are four main steps: (1) knowledge representation, (2) web-based system shell (including inference engine and user interface), (3) knowledge engineering and knowledge base construction, (4) knowledge validation and system evaluation. We used previous well-established “Probabilistic Dermatopathological Diagnostic Decision Support System” as mainframe; which contained a knowledge base for blistering diseases. A knowledge engineering technique was implemented to build a knowledge base for autoimmune diseases. The knowledge engineering process included one rheumatological expert, one coordinator and one knowledge engineer. This group built up a data dictionary and knowledge base. Programming engineer developed the interface engine and user interface to access these two knowledge bases in a web-based architecture. Published case-report articles are used for the final evaluation. Beside the previous established knowledge base for blistering diseases, the knowledge base for autoimmune diseases represents 6 disease frames with 98 findings in the dictionary. Each disease frame consists of 5~10 findings, which represents clinical, laboratory test results and radiologic clues for the diagnosis of a disease. By mean of the mathematical formulation called “cross-domain Bayesian formulation”, values of Apriori, TPR and FPR could be transferred in the different domains. It was also a mark progression in the development of decision support system. Clinical diagnostic decision support systems have been developed for over four decades, but most of them could be used on solitary platform. As the rise of Internet, there are several web-based clinical decision support systems created, but they could be accessed under limited circumstanced. We proposed this web-based cross-domain clinical diagnostic decision support system for non-infectious blistering diseases and autoimmune diseases, and we believe this system maybe very well be the first such probabilistic cross-domain decision support system in a web-based fashion.

並列關鍵字

bullous disease ； autoimmune disease ； decision support system ； probabilistic inference ； cross-domain Bayesian formulation ； knowledge engineering ； knowledge base ； Internet

參考文獻

1.Razzaque Ahmed A. Diagnosis of bullous disease and studies in the pathogenesis of blister formation using immunopathological techniques. Journal of Cutaneous Pathology. 1984 Aug; 11(4): 237-48.

2.Goldberg SH, Bronson D. Blistering diseases. Diagnostic help for primary care physicians. Postgraduate Medicine. 1991; 89(2): 159-62.

6.Vander AD, Van Stoecker W. An expert diagnostic program for dermatology. Comp Biomed Res 1984; 17(3): 241-7.

7.Haberman HF, Norwich KH, Diehl DL, Evans SJ, Harvey B, Landau J. et al. DIAG: a computer-assisted dermatologic diagnostic system--clinical experience and insight. Journal of the American Academy of Dermatology. 1985; 12(1 Pt 1): 132-43.

8.Potter B, Ronan SG. Computerized dermatopathologic diagnosis. Journal of the American Academy of Dermatology. 1987; 17(1): 119-31.