Department of Internal Medicine, People's Hospital of Yanshan County, Cangzhou, Hebei, China.
Eur Rev Med Pharmacol Sci. 2023 Jun;27(12):5338-5355. doi: 10.26355/eurrev_202306_32768.
This work aimed to explore the application of lightweight artificial intelligence algorithms in magnetic resonance imaging (MRI) image processing of patients with acute ischemic stroke (AIS) to clarify the effect and mechanism of early rehabilitation training on the mobilization of circulating endothelial progenitor cells (EPCs) in AIS.
A total of 98 AIS patients undergoing MRI examination were selected as the research objects and were randomly divided into a rehabilitation group (early rehabilitation training, 50 cases) and a routine group (conventional treatment, 48 cases) by random number table method and lottery method. In this work, based on the convolutional neural network (CNN) algorithm, a low-rank decomposition algorithm was introduced to optimize it, and a lightweight MRI image computer intelligent segmentation model (LT-RCNN) was established. The LT-RCNN model was used in the MRI image processing of AIS patients, and the role of the model in AIS image segmentation and lesion localization was analyzed. Furthermore, flow cytometry was used to detect the number of peripheral circulating EPCs and CD34+KDR+ cells in the two groups of patients before and after treatment. The serum levels of vascular endothelial growth factor (VEGF), tumor necrosis factor-α (TNF-α), interleukin 10 (IL-10), and stromal cell-derived factor-1α (SDF-1α) content were detected by Enzyme-Linked Immunosorbent Assay (ELISA). In addition, the correlation between each factor and CD34+KDR+ was analyzed by Pearson linear correlation.
The diffusion-weighted imaging (DWI) signal of MRI images of AIS patients under the LT-RCNN model was high. The location of the lesion could be accurately detected, and the contour of the lesion could be displayed and segmented, and the segmentation accuracy and sensitivity were significantly better than before optimization. The number of EPCs and CD34+KDR+ cells in the rehabilitation group was increased compared with the control group (p<0.01); the expression levels of VEGF, IL-10, and SDF-1α were higher than those of the control group (p<0.001), and TNF-α content was lower than the control group (p<0.001). The number of CD34+KDR+ cells was positively correlated with VEGF, IL-10, and TNF-α contents (p<0.01).
The results showed that the computer-intelligent segmentation model LT-RCNN could accurately locate, and segment AIS lesions and the early rehabilitation training could change the expression level of inflammatory factors and further promote the mobilization of AIS circulation EPCs.
本研究旨在探讨轻量级人工智能算法在急性缺血性脑卒中（AIS）患者磁共振成像（MRI）图像处理中的应用，阐明早期康复训练对 AIS 患者循环内皮祖细胞（EPCs）动员的作用及机制。
选取 98 例行 MRI 检查的 AIS 患者作为研究对象，采用随机数字表法和抽签法将其分为康复组（早期康复训练，50 例）和常规组（常规治疗，48 例）。本研究基于卷积神经网络（CNN）算法，引入低秩分解算法对其进行优化，建立了一种轻量级 MRI 图像计算机智能分割模型（LT-RCNN）。应用 LT-RCNN 模型对 AIS 患者的 MRI 图像进行处理，分析模型在 AIS 图像分割和病灶定位中的作用。采用流式细胞术检测两组患者治疗前后外周血循环 EPCs 及 CD34+KDR+细胞数量，采用酶联免疫吸附试验（ELISA）检测血清血管内皮生长因子（VEGF）、肿瘤坏死因子-α（TNF-α）、白细胞介素 10（IL-10）及基质细胞衍生因子-1α（SDF-1α）含量。采用 Pearson 线性相关分析各因素与 CD34+KDR+的相关性。
LT-RCNN 模型下 AIS 患者磁共振成像弥散加权成像（DWI）信号高，病灶位置可准确检出，病灶轮廓可清晰显示并分割，分割准确性和灵敏度明显优于优化前。康复组 EPCs 及 CD34+KDR+细胞数量较对照组增加（p<0.01）；VEGF、IL-10、SDF-1α 表达水平高于对照组（p<0.001），TNF-α 含量低于对照组（p<0.001）。CD34+KDR+细胞数量与 VEGF、IL-10、TNF-α 含量呈正相关（p<0.01）。
研究结果表明，计算机智能分割模型 LT-RCNN 能准确定位、分割 AIS 病灶，早期康复训练可改变炎症因子表达水平，进一步促进 AIS 循环 EPCs 的动员。
