基於知識本體與資訊萃取技術提升核醫科造影檢查報告正確性

Using Ontology and Information Extraction Techniques to Improve Nuclear Medicine Report Accuracy

目的：2014 年十大主要死因第一名為惡性腫瘤，而正子斷層攝影正是判斷惡性腫瘤以非侵入性最佳的方式，但在過去的研究當中鮮少討論將兩個不同領域的報告透過資訊科技技術，來幫助核子醫學科與解剖病理科醫師分析報告的差異性，並將其回饋至臨床作業改善醫療品質。方法：本研究以長庚紀念醫院高雄院區之核子醫學科正子斷層攝影報告與解剖病理報告為主開發報告稽核資訊系統，系統分為報告收集代理人、報告稽核代理人、報告警示系統與報告查詢系統，首先將非結構化萃取成結構化的報告，經報告差異知識本體比對，提供差異性的報告給核子醫學科醫師。結果：本系統自動判斷檢查報告與病理報告結果不一致或有疑慮時，能以紅燈、黃燈與綠燈的自動警示方式回饋給核子醫學科醫師，紅燈表示兩項檢查報告確實不同，黃燈表示系統無法判讀或醫師確認無法分析的報告，綠燈表示兩項檢查報告相同無誤，藉由過去的檢查與病理報告協助醫生修改判讀的思維，來提升報告的正確性。結論：本研究驗證了檢查報告及病理報告雖為不同領域，但經過資訊技術分析建立報告稽核資訊系統的可行性，並彰顯了導入系統後帶來的實質效益，未來希望將此系統擴展至其他醫學檢查報告，持續改善醫療品質及提升病人安全。

Objectives: Cancer (malignant neoplasms) ranked first among the top 10 causes of death in 2014. Positron emission tomography (PET) is recognized as the best non-invasive imaging technique for the detection of cancer. However, previous research contains little discussion on how to utilize information technology to help physicians of nuclear medicine and anatomic pathology analyze the differences between PET reports and anatomic pathology reports, and feed the analysis results back into clinical operations for better quality of treatment. Methods: We developed a report auditing system based on PET/CT scan reports and anatomic pathology reports in Kaohsiung Chang Gung Memorial Hospital. This system consists of a report extraction agent, a report check agent, a report alert system, and a report search system. This system first converts unstructured data into structuralized reports through a data extraction process. Later, through ontology-based comparisons between reports, this system can identify discrepant reports for nuclear medicine physicians. Results: The proposed system is able to automatically signal nuclear medicine physicians (using red, yellow, and green lights) when it detects incongruences between PET/CT scan reports and anatomic pathology reports, or it senses a need for human attention. A red light indicates discrepancy between reports; a yellow light shows that the system is unable to compare differences between reports or manual analysis by physicians is required; a green light suggests no difference between reports. Through comparison with past examination reports and pathology reports, this system can assist physicians in establishing a correct mindset for report interpretation and improving the accuracy of reports. Conclusions: This study shows that although examination reports and pathology reports are two separate disciplines, it is feasible to utilize information technology to develop a report auditing system to compare the reports. In this study, the effectiveness of this system is confirmed. In the future, this system can be extended to include other examination items, to continuously improve healthcare quality and patient safety.