自動化識別用於數位腦血管攝影醫材管路之智能系統

目的：腦中風是造成死亡與失能的主因之一，數位腦血管攝影是診斷和治療腦中風的醫學影像重要工具。醫材管路種類繁多，在選用上必須藉由醫師專業知識和經驗的判斷做選擇，此過程需耗費相當時間，然而腦中風病人的黃金治療時間常常相當短暫。因此，本研究以提出電腦輔助系統用於識別醫材管路在腦血管攝影檢查，以減少操作者對醫材管路使用時的差異，期望能加速診斷治療的流程。方法：本研究中，所有管路影像擷取來自我們所屬醫療機構。依據一名放射科醫師與一名護理師專業共識標準，收集醫材管路分為H1影像、JB2影像以及VTK影像之三類別各250張影像。CNN基本架構由卷積層、最大池化層和全連接層所組成。模型效能以10等分交叉驗證進行評估。管路辨識耗費時間以回溯性調查方式，回溯於腦中風診斷治療流程中，在數位腦血管攝影檢查中所使用管路（H1,JB2,VTK）數量，以及每次辨識所耗費時間，作為評估。結果：結果顯示AlexNet模型整體準確率達到82.6±3.5%，其接收者操作特徵曲線下方面積為0.954；Transferred AlexNet模型整體準確率達到87.3±2.4%，其接收者操作特徵曲線下方面積為0.971。過去於腦中風病人的數位腦血管攝影檢查中花費於管路辨識的時間約為30－70秒。結論：臨床使用上，此開發的卷積神經網路模型作為電腦輔助識別系統，可以加強工作人員的辨識能力，提升工作流程速度，並提供放射科醫師在執行介入性治療時更有一致性建議。

關鍵字

腦中風；醫材管路；電腦輔助辨識

並列摘要

Purposes: Stroke is one of the leading causes of disability and death. Digital cerebral angiography is an important imaging modality for diagnosis. The medical catheter to be used is chosen manually by experienced radiologists, but this can be time consuming. Furthermore, there is often a limited therapeutic time window for stroke patients. Therefore, this study proposed a computer-assisted system for medical catheter recognition at cerebral angiography examinations, which would reduce inter-operator variabilities and facilitate rapid treatment in the diagnostic procedures. Methods: All of the medical catheter images used in this retrospective study were obtained from our institution. Taking the consensus of a radiologist and a nurse as the reference standard, the collected catheter images were classified into three categories (250 H1 images, 250 JB2 images, and 250 VTK images). A convolutional neural network (CNN) was composed of convolutional layers, maximum pooling layers, and fully connected layers. The proposed CNN's performance was evaluated using ten-fold crossvalidation. Retrospective investigation was used to determine, for each therapy, the number of medical catheters (H1, JB2, VTK) and the time spent for catheter selection. Results: In the results, the proposed AlexNet without pre-trained parameters achieved an overall accuracy of 82.6±3.5%, with an area under the receiver operating characteristic curve (AUC) of 0.954. Using transfer learning, the transferred AlexNet achieved an overall accuracy of 87.3±2.4%, with an AUC of 0.971. The expected time saved for catheter selection in each cerebral angiography examination is about 30-70 seconds using this model. Conclusions: For clinical use, the developed CNN models serve as a computer-assisted recognition system that can improve personnel's workflow productivity and provide consistent suggestions to radiologist for interventional therapy.

並列關鍵字

Stroke ； Medical Catheter ； Computer-Assisted Recognition

參考文獻

高淑娟 (2021)．腦中風患者身體功能與健康相關生活品質之追蹤研究。https://doi.org/10.6834/csmu202100026

王駿發、歐陽諺、蔡安朝、吳韋慶 (2020)．可抗遮蔽人臉偵測與辨識系統．電機工程會刊( 創刊號1) ，1-7。https://doi.org/10.6949/KHCIEE.202007_(1).0001

謝明儒、陳盈如、湯頌君、陳健驊、林稜傑、薛承君、李俊泰、張谷州、連立明、陳龍 (2020)．2020 台灣腦中風學會與台灣急診醫學會急性缺血中風病人的院前緊急處置與急診診斷治療指引．台灣中風醫誌，2(4)，257-275。https://doi.org/10.6318/FJS.202012_2(4).0001

錢駿、鍾芷萍 (2018)．缺血性腦中風的分類．臨床醫學月刊，81(2)， 72-76。https://doi.org/10.6666/ClinMed.201802_81(2).0014

陳幼梅、蔡淳娟 (2020)．從全人健康照護觀點談護理診療思辨教育之智慧創新．台灣擬真醫學教育期刊，7(1)，30- 44 。https://doi.org/10.6582/JTSSH.202006_7(1).0003

國際替代計量

自動化識別用於數位腦血管攝影醫材管路之智能系統

全文下載

主題瀏覽