急性缺血型中風之缺血半影與白質病變分割及量化;Segmentation and Quantification of Ischemic Penumbra and White Matter Hyperintensity Lesions in Acute Ischemic Stroke

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Electrical Engineering > Electronic Thesis & Dissertation > Item 987654321/95764

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95764

Title:	急性缺血型中風之缺血半影與白質病變分割及量化;Segmentation and Quantification of Ischemic Penumbra and White Matter Hyperintensity Lesions in Acute Ischemic Stroke
Authors:	卓祐晟;CHO, YU-CHENG
Contributors:	電機工程學系
Keywords:	急性缺血型中風腦缺血區;腦梗塞區;白質病變;缺血半影;自動化分割;深度學習;實質腦遮罩提取;影像對位;Acute ischemic stroke brain ischemic region;Acute ischemic stroke brain infarct area;White matter hyperintensities;Penumbra;Automated segmentation;Deep learning;Brain mask extraction;Image registration
Date:	2024-07-29
Issue Date:	2024-10-09 17:15:33 (UTC+8)
Publisher:	國立中央大學
Abstract:	醫師對於急性中風介入治療的決策上，仍需花上不少時間在影像上的判讀，且無法以肉眼快速精準的量化出缺血半影區，因此在搶救病人的決策上，仍有改善空間。本研究提出了一套基於深度學習自動化分割的系統，來實現能夠針對急性缺血型中風患者的缺血區和核心梗塞區，及在影像上同樣呈現高信號的白質病變區域，進行準確的分割與量化。在本研究中，我們使用了Mask-Region convolutional neural network (Mask R-CNN)來進行實質腦遮罩的提取，去除頭骨、眼球等易影響後續神經網路誤判為病灶的區域，再透過影像正規化及重採樣的方式來加強影像的特徵，接著再使用三維卷積神經網路 DeepMedic，分別對DWI和FLAIR進行缺血區、梗塞核心區與白質病變的分割，透過兩者分割的結果比較，量化出缺血但尚未到達梗塞壞死的缺血半影的腦體素體積與比例，最後，我們也開發了一套以利醫師更直觀、更好上手的圖形使用者介面(Graphical User Interface, GUI)，更有助於輔助醫師做決策。本次研究中，我們採用聯新國際醫院申請於2020年1月1日至2023年4月30日有採取針對腦中風之動脈血栓溶解術或血栓切除術治療的病歷資料，將所有收集的影像取得之後將被去除辨識，最終患有急性缺血型中風病總計有267位病人，我們採用了其中216筆取栓治療前以及89筆取栓治療後追蹤的影像。最終研究結果顯示，我們使用的Mask R-CNN進行實質腦提取的Dices Score達到0.966，使用三維卷積神經網路 DeepMedic進行缺血區自動分割的Dice Score為0.801，梗塞核心區的Dice Score為0.745，白質病變自動分割的Dice Score為0.656。關鍵詞:急性缺血型中風腦缺血區、腦梗塞區、白質病變、缺血半影、自動化分割、深度學習、實質腦遮罩提取、影像對位;In this study, we employed a Mask-Region Convolutional Neural Network (Mask R-CNN) to extract brain masks, removing regions such as the skull and eyeballs that might lead neural networks to misinterpret them as lesions. Following this, we enhanced the image features through normalization and resampling. Subsequently, we utilized a three-dimensional convolutional neural network, DeepMedic, to segment ischemia, infarct core, and white matter hyperintensities on DWI and FLAIR images. By comparing the segmentation results of the two modalities, we quantified the ischemic penumbra and differentiated the brain parenchymal volume affected by white matter hyperintensities. Additionally, we developed a user-friendly Graphical User Interface (GUI) to assist physicians in decision-making. For this study, we collected medical records from Landseed International Hospital between January 1, 2020, and April 30, 2023, involving patients who underwent thrombolysis or thrombectomy for acute ischemic stroke. All identifiable information was removed from the collected images. Ultimately, data from a total of 267 patients with acute ischemic stroke were included, comprising 216 pre-thrombolytic treatment images and 89 post-thrombolytic treatment follow-up images. The study results indicated a Dice Score of 0.966 for brain extraction using Mask R-CNN, 0.801 for automatic segmentation of ischemia, 0.745 for infarct core segmentation, and 0.656 for white matter hyperintensities segmentation using the three-dimensional convolutional neural network, DeepMedic. Keywords: Acute ischemic stroke brain ischemic region, Acute ischemic stroke brain infarct area, Penumbra, White matter hyperintensities, Automated segmentation, Deep learning, Brain mask extraction, Image registration
Appears in Collections:	[Graduate Institute of Electrical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	15	View/Open

社群 sharing

Loading...