急性腦中風輔助診斷雲端平台之開發

腦中風為一種急症，若沒有即時接受適當治療，可能造成中至重度殘障的後遺症。由於急性腦中風其診斷及治療的複雜性與時效性，需神經專科醫師治療，但非都會區醫院急診可能無神經科醫師 24 小時值班，故研究開發一套輔助診斷急性腦中風之平台，將病患診斷資訊及影像儲存於雲端，供第一線臨床人員準確且有效的依據與處置建議。而傳統的醫院資訊管理中， PACS 系統的影像存儲通常採用定期不斷地進行存儲擴充的方式，但影像越來越精細需要更大的空間儲存，因此需利用雲端存儲影像及資料。而本研究專為腦中風患者設計之雲端平台，可透過系統查看腦中風之影像及資訊，將影像處理平台下載後，透過 ROI 圈選出影像中病灶的位置，並搭配自動病灶預測平台作為參考，以提供醫師診斷建議。在研究中，使用 NAS 架設網頁伺服器，以 WordPress 進行網頁設計，並以 phpMyAdmin 來編寫 PHP 處理 Web 上的 MySQL，進行腦中風資料庫的建立及管理，使用介面以網頁呈現，設置帳戶權限來維護資料安全性，根據使用者權限，可新增、查詢、修改、刪除病患之腦中風資訊。平台並使用出血性腦中風 25 組CT影像、ac-ASPECTS 25 組 CT 影像、pc-ASPECTS 15 組 CT 影像及腦內微出血 20 組 CT 影像，測試其資料上傳、搜尋功能、權限安全及影像處理平台之功能。結果顯示本研究在網頁搜尋病患資訊之驗證，準確率可達 100%。找六位使用者測試本系統，並收集使用結果，以單獨使用急性腦中風電腦輔助診斷雲端平台為舊系統，而加入自動病灶預測平台參考病灶相對位置為新系統，在效率部分新系統可明顯減少使用者圈選時間；在以 ac-ASPECTS 及pc-ASPECTS分數協助判定後續處理之準確率舊系統比較分別為 66.7% 與56.7%，在新系統則分別為 80%與 73.3% ，可明顯看出新系統之效能較佳。由結果顯示本研究開發急性腦中風輔助診斷雲端平台，除了能幫助醫師更有效率的判斷，並提供客觀的資訊輔助經驗較不足的醫師。由於目前影像處理分析還未能在線上處理，必須將平台下載至本機，較為不便，未來希望能使系統線上化，將安全性增強，並進行跨院區使用測試，讓本平台能實際運用於不同院區中。

關鍵字

急性腦中風；雲端平台；影像處理；特徵參數；人工智慧

並列摘要

Stroke is a kind of emergency, if not treated immediately, it may cause the sequelae of moderate to severe disability. Due to the complexity and timeliness of diagnosis and treatment of acute stroke, neurologists are required to treat it, but there may not be any neurologists on duty 24 hours a day in non-metropolitan hospitals. The development of a platform to assist in the diagnosis of acute stroke by storing patient diagnostic information and images in the cloud for frontline clinicians to make accurate and effective recommendations for treatment. In traditional hospital information management, image storage in PACS systems is usually done on a regular basis with continuous storage expansion, but increasingly detailed images require more space for storage, so cloud storage of images and data is required. In this study, a cloud-based platform was designed for stroke patients to view stroke images and information through the system. After downloading the image processing platform, the location of the lesions in the images was selected through ROI and the automatic lesion prediction platform was used as a reference to provide diagnostic recommendations to doctors. In the study, a web server was set up using NAS, WordPress was used for web design, and phpMyAdmin was used to write PHP to handle MySQL on the web for the creation and management of the stroke database. According to the user's permission, the patient's stroke information can be added, queried, modified, and deleted. The platform also uses 25 CT images of hemorrhagic stroke, 25 CT images of ac-ASPECTS, 15 CT images of pc-ASPECTS, and 20 CT images of Micro-bleeds in the brain to test its data uploading, search function, security, and image processing platform functions. The results of this study showed that the accuracy of the web-based patient information search was 100%. Six users were asked to test the system and the results were collected. The old system used the Acute Stroke Computer Assisted Diagnosis Cloud Platform alone and the new system used the Automatic Prediction Platform to refer to the relative location of the lesion. The efficiency of the new system is obvious that can reduce the time for users to circle the selection. The accuracy rates of ac-ASPECTS and pc-ASPECTS scores to assist in the follow-up process are 66.7% and 56.7% for the old system, and 80% and 73.3% for the new system, respectively. The results of this study show that the development of a cloud-based platform for acute stroke diagnosis will not only help doctors to make more efficient judgments, but also provide objective information to assist doctors with less experience. As the image processing and analysis cannot be done online yet, the platform has to be downloaded to the local computer, which is rather inconvenient. In the future, we hope to make the system available online, enhance security, and conduct cross-hospital testing, so that the platform can be used in different hospitals.