椎動脈狹窄(>50 %)所造成的中風診斷，目前仍然是神經放射科中的一個比較難以明確評估的問題。相同的，對椎動脈狹窄(>50 %)的病患是否應該接受氣球擴張術及血管支架植入手術，對手術風險和治療成效的評估，目前也是現代血管微創介入性神經放射線手術中需要深入了解和探討的課題。另外，椎動脈的曲折程度也是增加手術難度的因素，因此在術前對椎動脈狹窄的解剖位置和手術進路可能性，進行有效且便捷的分析至關緊要。由於腦中風的臨床表現會因為不同部位的腦部損傷而不同，初期診斷對非神經專科醫師而言，也不是一件簡單的工作。目前由電腦斷層血管造影所產生的影像，僅適用於放射科醫師，對於大多數沒有接受過專業放射學訓練的急診室醫師而言，這些電腦斷層血管造影所產生的影像無法立即地被判讀。因此我們假設，隨著人工智慧的發展與GPU製程技術的進步，現今對於深度學習網路的應用亦趨成熟，使得我們可以透過這項新穎的數據分析技術，協助醫師評估椎動脈的狹窄程度和曲折程度訊息，提高椎動脈狹窄的診斷正確率並提供更精確的術前風險分析。 本研究之目的即是將電腦斷層血管造影與深度學習的優點相結合，透過深度學習網路對電腦斷層血管造影進行自動去除骨骼的處理，獲得去骨之後的電腦斷層掃描投影影像，並建立3D椎動脈模型。透過3D影像投影技術與3D椎動脈模型，可以協助急診醫師在沒有專業的放射科醫師的協助下進行初步的解讀，如此便能節省時間，並改善急診中風病患的檢傷分類，儘早安排缺血性中風病患接受治療，減低預後不良程度。同時，3D動脈模型也可以協助醫師對於頸動脈和椎動脈狹窄做更精準的血管微創介入性手術的術前評估。 ;The neuroradiological diagnosis of ischemic strokes caused by vertebral artery stenosis (>50 %) remains a challenging task. Similarly, the risk of endovascular neuroradiological operations and the efficacy of treatment outcome require further investigation. For example, the endo-vascular neuroradiological procedures such as angioplasty and stenting to treat vertebral artery stenosis for acute ischemic stroke (AIS) should be investigated in larger studies. In addition, the tortuosity of vertebral arteries is a factor which increases the operation difficulty; therefore, preoperative route planning is essential identify possible unanticipated problems and to de-velop contingency plans for these difficult cases. The initial diagnosis of strokes is not easy for non-subspecialists due to the diversity of brain injuries. Currently, the images generated from the computed tomography angiography (CTA) protocols are designed for radiologists only. They are presented to clinicians in the form of stacks of 2D images, which are not immediately comprehensible to most emergency physicians who do not have the subspecialty radiology training in AIS. On the other hand, with the advances in artificial intelligence and the manu-facturing cost drop of GPU, the application of deep learning network has become readily available. Here, we hypothesize that using this novel data analysis technology can improve the diagnostic accuracy of vertebral artery stenosis and preoperative risk assessment. The purpose of this study is to apply the deep learning technology to process CTA images for generating bone subtracted CTA automatically. By removing bone from raw CTA images us-ing deep convolution neural networks (CNN), 3D data processing techniques such as volume rendering and marching cubes surface modeling can visualize 3D vertebral arteries on 2D screen. Therefore, clinicians in the setting of emergency rooms can evaluate vertebral artery stenosis and identify AIS patients more accurately without the assistance of neuroradiologists. This could potentially save time and improve the treatment outcome for AIS patients. In addi-tion, 3D vertebral artery model could assist clinicians for improving the precision of preopera-tive evaluation the of endovascular interventional neuroradiologic procedures.
