Lakatos Lehel, Bolognese Manuel, Müller Martin, Österreich Mareike, von Hessling Alexander
Department of Neurology and Neurorehabilitation, Lucerne Kantonsspital, Lucerne, Switzerland.
Department of Radiology (Section Neuroradiology), Lucerne Kantonsspital, Lucerne, Switzerland.
Front Neurol. 2022 Jul 8;13:907151. doi: 10.3389/fneur.2022.907151. eCollection 2022.
The present computerized techniques have limits to estimate the ischemic lesion volume especially in vertebrobasilar ischemia (VBI) automatically. We investigated the ability of the RAPID AI (RAPID) software on diffusion-weighted imaging (DWI) to estimate the infarct size in VBI in comparison to supratentorial ischemia (STI).
Among 123 stroke patients (39 women, 84 men, mean age 66 ± 11 years) having undergone DWI, 41 had had a VBI and 82 a STI. The infarct volume calculation by RAPID was compared to volume calculations by 2 neurologists using the ABC/2 method. For inter-reader and between-method analysis intraclass correlation coefficient (ICC), area under the curve (AUC) estimations, and Bland-Altman plots were used.
ICC between the two neurologists and each neurologist and RAPID were >0.946 (largest 95% boundaries 0.917-0.988) in the STI group, and > 0.757 (95% boundaries between 0.544 and 0.982) in the VBI group. In the STI group, AUC values ranged between 0.982 and 0.999 (95% 0.971-1) between the 2 neurologists and between 0.875 and 1 (95% 0.787-1) between the neurologists and RAPID; in the VBI group, they ranged between 0.925 and 0.965 (95% 0.801-1) between the neurologists, and between 0.788 and 0.931 (95% 0.663-1) between RAPID and the neurologists. Compared to the visual DWI interpretation by the neurologists, RAPID did not recognize a substantial number of infarct volumes of ≤ 2 ml.
The ability of the RAPID software to depict strokes in the vertebrobasilar artery system seems close to its ability in the supratentorial brain tissue. However, small lesion volumes ≤ 2 ml remain still undetected in both brain areas.
当前的计算机技术在自动估计缺血性病变体积方面存在局限性，尤其是在椎基底动脉缺血（VBI）中。我们研究了RAPID人工智能（RAPID）软件在扩散加权成像（DWI）上估计VBI梗死面积的能力，并与幕上缺血（STI）进行比较。
在123例接受DWI检查的中风患者（39例女性，84例男性，平均年龄66±11岁）中，41例患有VBI，82例患有STI。将RAPID计算的梗死体积与2名神经科医生使用ABC/2方法计算的体积进行比较。对于阅片者间和方法间分析，使用组内相关系数（ICC）、曲线下面积（AUC）估计值和Bland-Altman图。
在STI组中，两名神经科医生之间以及每名神经科医生与RAPID之间的ICC>0.946（最大95%界限为0.917 - 0.988），在VBI组中>0.757（95%界限在0.544和0.982之间）。在STI组中，两名神经科医生之间的AUC值在0.982至0.999（95%为0.971 - 1）之间，神经科医生与RAPID之间的AUC值在0.875至1（95%为0.787 - 1）之间；在VBI组中，神经科医生之间的AUC值在0.925至0.965（95%为0.801 - 1）之间，RAPID与神经科医生之间的AUC值在0.788至0.931（95%为0.663 - 1）之间。与神经科医生的视觉DWI解读相比，RAPID未识别出大量体积≤2 ml的梗死灶。
RAPID软件描绘椎基底动脉系统中风的能力似乎与其在幕上脑组织中的能力相近。然而，在两个脑区中，体积≤2 ml的小病变仍未被检测到。
