摘要
目的探讨头颈部计算机体层扫描血管成像(CTA)应用人工智能(AI)对脑卒中患者的图像质量研究。方法选择2021年6月至2022年6月在乐山市人民医院接受治疗的脑卒中患者60例,其中男性38例,女性22例;年龄46~92岁,平均年龄68.26岁(标准差5.25岁);单侧肢体无力19例,口角歪斜14例,语言不清13例,昏迷8例,肢体运动不协调6例。所有患者头颈部CTA原始图像均上传星云工作站,并由两名主治医师对图像进行手动后处理及上传AI工作站,分别重建容积再现(VR)、最大密度投影(MIP)、曲面重建(CPR)。对比AI重建与手动重建的图像质量及时效性。结果头颈部CTA对脑卒中患者诊断中,AI重建的VR图像3分为75%,2分为23.3%,1分为1.7%;MIP图像3分为76.7%,2分为23.3%;CPR图像3分为85%;2分为15%;手动重建VR图像3分为80%;2分为20%;MIP图像3分为81.7%,2分为18.3%;CPR图像3分为95%,2分为5%。手动后处理与AI重建方法VR、MIP、CPR图像比较,差异均无统计学意义(P>0.05)。AI处理图像所需时间为(2.33±0.15)min,手动处理图像的时间为(12.33±2.30)min,AI的处理时间明显少于手动处理时间(P<0.05),AI较手动时间增益率为83.3%。结论AI后处理图像质量与手动后处理图像质量相近,但AI后处理图像时效性更好,有利于为脑卒中患者治疗赢得更多时间,值得临床推广。
Objective To discuss the image quality of cranio-cervical computed tomography angiography(CTA) combined with artificial intelligence(AI) in patients with stroke. Methods From June 2021 to June 2022, a total of 60 patients with stroke were enrolled, which included 38 males and 22 females, aged 46-92 years old with mean age of 68.26 years old(standard deviation 5.25 years old);clinical manifestations of unilateral limb weakness in 19 cases(31.67 %), angular deviation in14(23.33 %), unclear language in 13(21.67 %), coma in 8(13.33 %), and uncoordinated limb movements in 6(10.00 %). The cranio-cervical CTA original images of all patients were uploaded to the nebula workstation and the images were manually post-processed by two attending physicians, and uploaded to AI workstation. The volume rendering(VR), maximum density projection(MIP) and curved plannar reconstruction(CPR) were respectively reconstructed. The image quality and timeliness of AI reconstruction and manual reconstruction were compared. Results In diagnosis of stroke patients by cranio-cervical CTA,AI reconstructed VR images 3-score was 75 %, 2-score was 23.3 %, 1-score was 1.7 %;MIP images 3-score was 76.7 %, 2-score was 23.3 %;CPR images 3-score was 85 %, 2-score was 15 %. Meantime, manual reconstruction of VR images 3-score was80 %, 2-score was 20 %;MIP images 3-score was 81.7 %, 2-score was 18.3 %;CPR images 3-score was 95 %, 2-score was 5 %.There was no statistically significant difference between manual post-processing and AI reconstruction methods for VR, MIP, and CPR images(P>0.05). The time required for AI image processing was(2.33 ± 0.15) minutes, while the time for manual image processing was(12.33 ± 2.30) minutes. The processing time of AI was significantly shorter than that of manual processing(P<0.05),and the gain rate of AI was 83.3 % than that of manual. Conclusion It is demonstrated that the quality of AI post-processing image is similar to that of manual post-processing image, but the timeliness of AI post-processing image is better,
作者
彭新壹
余佳强
刘玉凯
张健
余东
PENG Xin-yi;YU Jia-qiang;LIU Yu-kai;ZHANG Jian;YU Dong(Department of Radiology,Leshan People's Hospital,Leshan 614000,Sichua,China)
出处
《生物医学工程与临床》
CAS
2023年第4期471-475,共5页
Biomedical Engineering and Clinical Medicine
基金
乐山市级科研基金项目(20SZD152)。
