資料探索以不同的演算法應用於醫學資料庫中可能獲得良好的診斷結果。醫院病例資料系統藉由醫療資訊整合結合現有之軟硬體技術,可使系統所蒐集之資料相容性提高,並加入資料探索技術使系統在臨床上更具實用性。本論文將延續先前的研究成果,將資料探索技術運用於不同類型的資料庫探討其可行性,以及提供醫學影像傳輸標準傳輸介面轉換,提高系統的整體適用性。 論文中採用的演算法分別為決策樹、貝氏網路及到傳遞類神經網路,分別對子宮頸抹片檢查資料、X-ray肺部腫瘤偵測結果資料與中醫舌診影像資料等三種醫學資料庫,透過演算法學習與測試運算,以演算法的診斷能力、診斷結果的解釋能力為指標,評估演算法應用於不同類型資料庫的可行性。過程中並將沒有標準格式化之醫學影像本研究為透過開發之DICOM影像轉換介面,轉換為DICOM影像格式並存入影像管理系統。 初步結果顯示對於子宮頸抹片檢查資料為建立資料庫與資料庫使用者介面,已獲取完整異常抹片病人病歷資料。將獲得的資料經決策樹、貝氏網路與倒傳遞類神經網路等演算法,其診斷正確率分別為73.91%、62.31%與57.97%。在X-ray肺部腫瘤偵測系統決策樹所獲得的結果較所採用倒傳遞類神經網路所獲得的結果為佳。Left_R、Left_Tong、Left_LR、Right_R、Mid_R、Mid_Coat診斷參數提供84.99%之正確率,將可廣泛應用於上消化道疾病的診斷。此外研究發現另外加入舌中、舌左的平均溫度可提升診斷的正確率為97.49%。 在整體而言決策樹是三種演算法中表現最為平均且系統有良好的信賴度,此外貝式網路則可在上消化道疾病病患舌診影像分析資料獲得最佳的診斷結果。未來可提高收集的資料量,進一步改善系統所獲得的結果,使系統更具臨床意義,並且達到輔助醫生診斷的目標。
The application of data mining for medical database which three difference algorithms application for the medical database, could obtain the best diagnosing performance. Furthermore, IHE (Integrating the Healthcare Enterprise) system tried to make the data which was collected by the system more compatible and combine function of data surveying and knowledge knowing. There were two objectives in our study. One was apply the method of data mining into different kind of database and evaluate its practicability. The other was provided DICOM transfer interface to improve system integrating practical. In this thesis, we introduced differentiate algorithms application for cervical smears test, X-ray of lung nodule detection result and Chinese medicine of the tongue diagnosis medical database. There were three algorithms which were Bayesian Network (BN), Decision Tree (DT) and Back Propagation Neural Network (BPN) included in this research. The system provided DICOM transfer interface to DICOM 3.0 part 10 format, and storage in system of medical manage image. Primary results showed that it could obtain complete patient histories with abnormal cervical smear by building up the database and the user interface of database. Furthermore, was the data calculate by DT, BN and BPN algorithms, the accurate rate was 73.61%, 62.31% and 57.97%, respectively. In X-ray of lung nodule detection system, result of data after analysis observed that the obtaining effect was better with DT than with BPN. Upper gastrointestinal disease diagnose which using Left_R, Left_Tong, Left_LR, Right_R, Mid_R, Mid_Coat could obtain 84.99% of accuracy by using BN method. It would apply extensively to diagnose in upper gastrointestinal. Besides, the investigation indicate adding the mean temperature of the middle of the tongue and the left of the tongue else could promote 97.49% of the accuracy of diagnose. In whole, DT performed the most equally and system with good confidence of three algorithms, and BN could obtain the best diagnosing results in tongue diagnose image analysis data of upper gastrointestinal patients. In the future, we fried to increase amount of collection data in order to improve the obtaining results from system and make the system possess more significant. It might achieve the goal of assisting doctors diagnosing.