Department of Medical Imaging Centre, The First People's Hospital of Xianyang, Xianyang 712000, Shannxi, China.
Department of Ultrasound Medicine, The First People's Hospital of Xianyang, Xianyang 712000, Shannxi, China.
Contrast Media Mol Imaging. 2022 May 9;2022:5373585. doi: 10.1155/2022/5373585. eCollection 2022.
The purpose of the research was to discuss the application values of deep learning algorithm-based computed tomography perfusion (CTP) imaging combined with head and neck computed tomography angiography (CTA) in the diagnosis of ultra-early acute ischemic stroke. Firstly, 88 patients with acute ischemic stroke were selected as the research objects and performed with cerebral CTP and CTA examinations. In order to improve the effect of image diagnosis, a new deconvolution network model AD-CNNnet based on deep learning was proposed and used in patient CTP image evaluation. The results showed that the peak signal-to-noise ratio (PSNR) and feature similarity (FSIM) of the AD-CNNnet method were significantly higher than those of traditional methods, while the normalized mean square error (NMSE) was significantly lower than that of traditional algorithms ( < 0.05). 80 cases were positive by CTP-CTA, including 16 cases of hyperacute ischemic stroke and 64 cases of acute ischemic stroke. The diagnostic sensitivity was 93.66%, and the specificity was 96.18%. The cerebral blood flow (CBF), cerebral blood volume (CBV), and the mean transit time (MTT) in the infarcted area were significantly greater than those in the corresponding healthy side area, and the time to peak (TTP) was significantly less than that in the corresponding healthy side area ( < 0.05). The cerebral perfusion parameters CBF, TTP, and MTT in the penumbra were significantly different from those in the infarct central area and the corresponding contralateral area, and TTP was the most sensitive ( < 0.05). To sum up, deep learning algorithm-based CTP combined with CTA could find the location of cerebral infarction lesions as early as possible to provide a reliable diagnostic result for the diagnosis of ultra-early acute ischemic stroke.
本研究旨在探讨基于深度学习算法的计算机断层灌注（CTP）成像与头颈部计算机断层血管造影（CTA）联合应用于超早期急性缺血性脑卒中的诊断价值。首先，选取 88 例急性缺血性脑卒中患者进行脑 CTP 和 CTA 检查。为提高图像诊断效果，提出了一种新的基于深度学习的去卷积网络模型 AD-CNNnet，并应用于患者 CTP 图像评价。结果表明，AD-CNNnet 方法的峰值信噪比（PSNR）和特征相似性（FSIM）显著高于传统方法，而归一化均方误差（NMSE）显著低于传统算法（<0.05）。80 例 CTP-CTA 检查阳性，其中超急性缺血性脑卒中 16 例，急性缺血性脑卒中 64 例。诊断灵敏度为 93.66%，特异度为 96.18%。梗死区脑血流量（CBF）、脑血容量（CBV）和平均通过时间（MTT）明显大于相应健侧区，达峰时间（TTP）明显小于相应健侧区（<0.05）。半暗带的脑灌注参数 CBF、TTP 和 MTT 与梗死中心区和相应对侧区有显著差异，其中 TTP 最敏感（<0.05）。综上所述，基于深度学习算法的 CTP 联合 CTA 能尽早发现脑梗死病灶的位置，为超早期急性缺血性脑卒中的诊断提供可靠的诊断结果。
