兩階段誘導式深度學習應用於多種類包裝藥品之圖像分類

Taipei Medical University Institutional Repository > 醫學院 > 人工智慧醫療碩士在職專班 > 博碩士論文 > Item 987654321/63346

請使用永久網址來引用或連結此文件: http://libir.tmu.edu.tw/handle/987654321/63346

題名:	兩階段誘導式深度學習應用於多種類包裝藥品之圖像分類 Two-Stage Induced Deep Learning for Diverse Packaging Drugs Image Classification
作者:	游育新 YOU, YU-SIN
貢獻者:	醫學院人工智慧醫療碩士在職專班林于翔
關鍵詞:	卷積神經網路;深度學習;藥品圖像分類;誘導式深度學習;兩階段誘導式深度學習 Convolutional neural network;Deep learning;Drug image classification;Induced deep learning;Two-stage induced deep learning
日期:	2023-07-14
上傳時間:	2023-12-15 14:37:21 (UTC+8)
摘要:	醫療疏失經統計為美國前三大死因，而用藥錯誤佔醫療疏失中相當大的比例，然而，用藥錯誤造成的原因大多屬於可避免之因素，這促使了世界衛生組織 (WHO) 發起"無傷害用藥運動" (Medication Without Harm Campaign) 以減少用藥錯誤對病患帶來的重大風險，目前用藥錯誤常見的防範措施包括高人字體 (Tall-man lettering)、自動配藥機和條碼管理系統，然而，這些傳統的防範措施各有其限制及缺陷。隨著人工智慧的蓬勃發展，近年來許多研究運用了先進的人工智慧技術，對於各種藥品進行自動化分類，研究結果顯示，運用人工智慧技術於藥品的自動化分類，無論是在分類準確率或推理速度，皆獲得了極大的進展。儘管人工智慧方法被證明可有效的運用於藥品自動化分類，但過往研究主要仍集中在對於無包裝之顆粒藥品、或者是單一包裝藥品進行自動化分類，然而，在實務上，醫療機構內存有上千種具不同包裝類型的相似藥品，在調劑過程中將大幅的增加配藥錯誤的風險。有鑑於此，本研究提出了一種新穎的兩階段誘導式深度學習 (TSIDL) 方法，以針對多種類不同包裝的相似藥品圖像進行自動化分類。實驗結果顯示，本研究所提出的兩階段誘導式深度學習方法，在108類不同包裝藥品圖像的分類任務之中，達到了99.39%的傑出分類準確率，除此之外，每張藥品圖像所需要的推理時間僅需3.12毫秒。以上實驗結果顯示了本研究所提出之兩階段誘導式深度學習方法，在未來的自動化配藥系統中具有實際應用的淺力，進而可有效降低用藥錯誤對病患帶來的重大風險。 Medical errors are the third leading cause of death in the United States, with medication errors being a significant preventable contributor. This has prompted the World Health Organization (WHO) to initiate the Medication Without Harm Campaign to prevent such errors. Among these errors, dispensing errors are crucial. Current prevention methods include tall-man lettering, automated dispensing machines, and barcode systems. In 2016, the National Library of Medicine (NLM) organized the Pill Image Recognition Challenge to encourage the development of algorithms for classifying pills. Furthermore, there have been numerous recent studies on the automatic classification of single-packaged drugs. Previous studies have mainly focused on investigating unpackaged pills or drugs with single-packaging. However, there are currently limited studies on the classification of drugs with diverse packaging, despite the presence of thousands of such drugs within healthcare institutions during the dispensing process, which significantly increases the risk of dispensing errors. In this study, we proposed a novel two-stage induced deep learning (TSIDL) framework to classify diverse packaging drugs. The results demonstrate that the proposed TSIDL method outperforms state-of-the-art CNN models in all classification metrics. Notably, it achieved a state-of-the-art accuracy of 99.39%. Moreover, this study also demonstrated that the TSIDL method achieved an inference time of only 3.12 ms per image. These results highlight the potential of real-time classification for similar drugs with diverse packaging and their applications in future dispensing systems.
描述:	碩士指導教授:林于翔委員:林于翔委員:彭徐鈞委員:劉文德
資料類型:	thesis
顯示於類別:	[人工智慧醫療碩士在職專班] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	90	檢視/開啟

在TMUIR中所有的資料項目都受到原著作權保護.

TAIR相關文章

資料載入中.....