Department of Pediatrics, Nagoya University Graduate School of Medicine, Nagoya, Japan.
Department of Pediatrics, Nagoya University Graduate School of Medicine, Nagoya, Japan.
Magn Reson Imaging. 2021 Jun;79:38-44. doi: 10.1016/j.mri.2021.03.004. Epub 2021 Mar 12.
The accuracy of brain age estimates from magnetic resonance (MR) images has improved with the advent of deep learning artificial intelligence (AI) models. However, most previous studies on predicting age emphasized aging from childhood to adulthood and old age, and few studies have focused on early brain development in children younger than 2 years of age. Here, we performed brain age estimates based on MR images in children younger than 2 years of age using deep learning. Our AI model, developed with one slice each of raw T1- and T2-weighted images from each subject, estimated brain age with a mean absolute error of 8.2 weeks (1.9 months). The estimates of our AI model were close to those of human specialists. The AI model also estimated the brain age of subjects with a myelination delay as significantly younger than the chronological age. These results indicate that the prediction accuracy of our AI model approached that of human specialists and that our simple method requiring less data and preprocessing facilitates a radiological assessment of brain development, such as monitoring maturational changes in myelination.
基于磁共振（MR）图像的脑龄估计的准确性随着深度学习人工智能（AI）模型的出现而提高。然而，大多数以前关于预测年龄的研究都强调了从儿童期到成年期和老年期的衰老，很少有研究关注 2 岁以下儿童的早期大脑发育。在这里，我们使用深度学习对 2 岁以下儿童的 MR 图像进行脑龄估计。我们的 AI 模型使用每个受试者的原始 T1-和 T2 加权图像的一个切片开发，其脑龄估计的平均绝对误差为 8.2 周（1.9 个月）。我们的 AI 模型的估计值与人类专家的估计值非常接近。该 AI 模型还估计了髓鞘化延迟受试者的脑龄，明显低于其实际年龄。这些结果表明，我们的 AI 模型的预测准确性接近人类专家，并且我们的简单方法需要更少的数据和预处理，有助于对大脑发育进行放射学评估，例如监测髓鞘化的成熟变化。
