急性缺血型腦中風(Acute Ischemic Stroke, AIS)是由大腦動脈阻塞而引起的腦部症狀。若能在黃金時間內恢復血流，未完全成為梗塞區(Infarct)的缺血區(Ischemia)還是有機會恢復。目前的臨床治療方法包括靜脈內血栓溶解治療(Intravenous Thrombolysis, IVT)或動脈內取栓手術(Endovascular Thrombectomy, EVT)。而電腦斷層影像(Computed Tomography, CT)是判斷急性缺血型中風病灶(Lesion)的重要依據。在病灶的標註上，主要依賴醫師的判讀，但是人工標註的方式耗時耗力，且存在主觀的不一致性，尤其在CT上，取栓治療前的缺血區更是肉眼較難判讀的。此研究的目標是建立自動分割急性缺血型腦中風缺血區的模型，客觀地協助臨床醫師重複確認該腦部區域是否有沒注意到的病灶。我們使用了70筆從聯新國際醫院(Landseed International Hospital)收集的醫學影像，經由影像前處理後，透過採用雙通道的三維卷積神經網路(Convolutional Neural Network, CNN)架構的DeepMedic進行腦遮罩分割模型及缺血區分割模型的訓練。缺血區分割模型的結果與黃金標準相比，其Dice達到73.8%。此研究結果在臨床上可以搭配磁振造影 (Magnetic Resonance Imaging, MRI)一同判斷，以提供更全面及詳細的診斷。;Acute Ischemic Stroke (AIS) is a brain symptom caused by the obstruction of cerebral arteries. If blood flow is restored within the golden hours, ischemic areas that have not fully become infarcted still have the potential to recover. Current clinical treatments include Intravenous Thrombolysis (IVT) and Endovascular Thrombectomy (EVT). Computed Tomography (CT) is a crucial tool for identifying the lesions in acute ischemic stroke patients. Lesion delineation primarily relies on clinicians′ judgment, but manual delineation is time-consuming and labor-intensive, with subjective inconsistencies, especially in identifying ischemic areas on pre-thrombectomy CT scans. This study aims to develop a model for automatic segmentation of ischemic areas in acute ischemic stroke, aiding clinicians in objectively confirming overlooked lesions. We used 70 medical images collected from Landseed International Hospital. After image preprocessing, we trained brain mask segmentation model and ischemic segmentation model using DeepMedic, a dual-channel 3D Convolutional Neural Network (CNN) architecture. The ischemic area segmentation model achieved a Dice coefficient of 73.8% compared to the gold standard. These results can be used clinically alongside Magnetic Resonance Imaging (MRI) to provide more comprehensive and detailed diagnoses.
