比較陰道滴蟲於不同自動化尿液機台以AI輔助之檢出率

Comparison of the Detection Rates of Trichomonas Vaginalis in Different Automated Urinalysis Analyzer with AI Approaches

傳統上，陰道滴蟲（Trichomonas vaginalis），可經由尿液檢體經顯微鏡檢查而發現，不僅快速、方便、且可在臨床症狀出現前檢出，近年來尿液機台自動化，雖降低人工鏡檢比例，但本部陰道滴蟲陽性率由0.6‰驟降至0.14‰。本研究探討兩種自動化尿液機台以AI（a rtificial intelligence）輔助後，陰道滴蟲檢陽性率的表現：(1)自動化以影像判讀機台搭配陰道滴蟲檢測模型（The detection models of T. vaginalis）人工鏡檢率2.6%，陰道滴蟲的陽性率為0.373‰。(2)自動化尿液以流式細胞儀配合優化後軟體判讀，人工鏡檢率0.5%，陰道滴蟲的陽性率為0.36 3‰。自動化尿液機台以AI輔助後皆提升陰道滴蟲陽性率。本研究對於使用自動化尿液機台，提升陰道滴蟲陽性率得到以下結論：(1)使用自動化影像法判讀若無AI，可用尿液化學與尿沈渣特定結果攔截(2)使用影像法可搭配陰道滴蟲檢測AI模型(3)使用流式細胞儀式系統，配合優化後軟體可提示有陰道滴蟲。未來將以流式細胞儀搭配陰道滴蟲檢測模型運算，因流式細胞儀陽性率（0.363‰），近似於陰道滴蟲檢測模型（0.373‰），但人工鏡檢率（0.5%）低較為便利。期許利用雙AI模式，在陰道滴蟲檢驗上既便利又能得到更好的效能。

Traditionally, Trichomonas vaginalis can be detected in urine through microscopic examination, which is not only rapid and convenient but can also be detected before clinical symptoms appear. In recent years, with the automation of urine testing, although it has reduced the proportion of manual microscopic examinations, the positivity rate of Trichomonas vaginalis has drastically decreased from 0.6‰ to 0.14‰. This study explores the performance of two automated urine analysis systems assisted by Artificial Intelligence (AI) in detecting Trichomonas vaginalis : (1) An automated image analysis system combined with a T. vaginalis detection model resulted in a manual microscopic examination rate of 2.6% and a vaginal Trichomonas detection rate of 0.373‰. (2) An automated urine analysis system using flow cytometry with optimized software for interpretation yielded a manual microscopic examination rate of 0.5% and a vaginal Trichomonas detection rate of 0.363‰. The use of AI assistance in automated urine analysis systems improves the detection of vaginal Trichomonas. This study concludes the following for enhancing Trichomonas vaginalis detection using automated urine analysis systems: (1) Using automated image system analysis without AI, one can set up the system to withhold the report when specific range of urine chemistry and sediment results are detected. This approach ensures that the results are flagged for manual review before generating a final report. (2) Image analysis can be combined with a T. vaginalis detection AI model. (3) The use of a flow cytometry system, coupled with optimized software, can indicate the presence of Trichomonas vaginalis. Future efforts will involve utilizing a flow cytometry system with a T. vaginalis detection model, as the Trichomonas detection rate (0.363‰) of flow cytometry method is similar to that of the T. vaginalis detection model with image analysis system (0.373‰), but with a lower manual microscopic examination rate (0.5%), making it more convenient. The aim is to achieve both convenience and improved efficiency in Trichomonas vaginalis detectioin through the use of a dual AI model.