From the Department of Clinical Neurosciences and Hotchkiss Brain Institute (K.Z., F. Bala, J.Z., F. Benali, P.C., R.M., N.S., M.D.H., M.G., A.D., B.K.M.).
College of Electronic Engineering (K.Z.), Xi'an Shiyou University, Xi'an, Shaanxi, China.
AJNR Am J Neuroradiol. 2023 Jun;44(6):641-648. doi: 10.3174/ajnr.A7878. Epub 2023 May 18.
Identifying the presence and extent of intracranial thrombi is crucial in selecting patients with acute ischemic stroke for treatment. This article aims to develop an automated approach to quantify thrombus on NCCT and CTA in patients with stroke.
A total of 499 patients with large-vessel occlusion from the Safety and Efficacy of Nerinetide in Subjects Undergoing Endovascular Thrombectomy for Stroke (ESCAPE-NA1) trial were included. All patients had thin-section NCCT and CTA images. Thrombi contoured manually were used as reference standard. A deep learning approach was developed to segment thrombi automatically. Of 499 patients, 263 and 66 patients were randomly selected to train and validate the deep learning model, respectively; the remaining 170 patients were independently used for testing. The deep learning model was quantitatively compared with the reference standard using the Dice coefficient and volumetric error. The proposed deep learning model was externally tested on 83 patients with and without large-vessel occlusion from another independent trial.
The developed deep learning approach obtained a Dice coefficient of 70.7% (interquartile range, 58.0%-77.8%) in the internal cohort. The predicted thrombi length and volume were correlated with those of expert-contoured thrombi ( = 0.88 and 0.87, respectively; < .001). When the derived deep learning model was applied to the external data set, the model obtained similar results in patients with large-vessel occlusion regarding the Dice coefficient (66.8%; interquartile range, 58.5%-74.6%), thrombus length ( = 0.73), and volume ( = 0.80). The model also obtained a sensitivity of 94.12% (32/34) and a specificity of 97.96% (48/49) in classifying large-vessel occlusion versus non-large-vessel occlusion.
The proposed deep learning method can reliably detect and measure thrombi on NCCT and CTA in patients with acute ischemic stroke.
在选择接受急性缺血性脑卒中治疗的患者时，识别颅内血栓的存在和范围至关重要。本文旨在开发一种自动量化脑卒中患者 NCCT 和 CTA 上血栓的方法。
共纳入 Safety and Efficacy of Nerinetide in Subjects Undergoing Endovascular Thrombectomy for Stroke（ESCAPE-NA1）试验中 499 例大血管闭塞的患者。所有患者均行薄层 NCCT 和 CTA 检查。手动勾画的血栓作为参考标准。开发了一种深度学习方法自动分割血栓。499 例患者中，263 例和 66 例患者被随机选择用于训练和验证深度学习模型，其余 170 例患者用于独立测试。采用 Dice 系数和体积误差对深度学习模型与参考标准进行定量比较。将提出的深度学习模型在另一个独立试验的 83 例有或无大血管闭塞的患者中进行外部测试。
在内部队列中，开发的深度学习方法获得了 70.7%的 Dice 系数（四分位间距，58.0%77.8%）。预测的血栓长度和体积与专家勾画的血栓长度和体积具有相关性（ = 0.88 和 0.87，均<0.001）。当应用于外部数据集时，该模型在大血管闭塞患者中获得了类似的 Dice 系数（66.8%；四分位间距，58.5%74.6%）、血栓长度（ = 0.73）和体积（ = 0.80）结果。该模型还在区分大血管闭塞与非大血管闭塞方面获得了 94.12%（32/34）的敏感性和 97.96%（48/49）的特异性。
该深度学习方法可可靠地检测和测量急性缺血性脑卒中患者 NCCT 和 CTA 上的血栓。
