Centre de recherche CERVO, 2601 de la Canardière, Québec, G1J 2G3, Canada.
Centre de recherche CERVO, 2601 de la Canardière, Québec, G1J 2G3, Canada.
Neuroimage. 2019 Aug 15;197:618-624. doi: 10.1016/j.neuroimage.2019.05.025. Epub 2019 May 11.
Aging is associated with structural alterations in many regions of the brain. Monitoring these changes contributes to increasing our understanding of the brain's morphological alterations across its lifespan, and could allow the identification of departures from canonical trajectories. Here, we introduce a novel and unique patch-based grading procedure for estimating a synthetic estimate of cortical aging in cognitively intact individuals. The cortical age metric is computed based on image similarity between an unknown (test) cortical label and known (training) cortical labels using machine learning algorithms. The proposed method was trained on a dataset of 100 cognitively intact individuals aged 19-61 years, within the 31 bilateral cortical labels of the Desikan-Killiany-Tourville parcellation, then tested on an independent test set of 78 cognitively intact individuals spanning a similar age range. The proposed patch-based framework yielded a R = 0.94, as well as a mean absolute error of 1.66 years, which compared favorably to the literature. These experimental results demonstrate that the proposed patch-based grading framework is a reliable and robust method to estimate brain age from image data, even with a limited training size.
衰老是与大脑许多区域的结构改变相关的。监测这些变化有助于增加我们对大脑在其生命周期内形态改变的理解，并可以识别偏离典型轨迹的情况。在这里，我们引入了一种新颖独特的基于斑块的分级程序，用于估计认知正常个体的皮质老化的综合估计。皮质年龄指标是基于机器学习算法，在未知（测试）皮质标签和已知（训练）皮质标签之间的图像相似性计算得出的。该方法在一个由 100 名认知正常的个体组成的数据集上进行了训练，这些个体的年龄在 19 至 61 岁之间，涉及 Desikan-Killiany-Tourville 分区的 31 个双侧皮质标签，然后在一个跨越相似年龄范围的 78 名认知正常个体的独立测试集上进行了测试。该基于斑块的框架产生了 R=0.94，以及 1.66 年的平均绝对误差，这与文献相比表现良好。这些实验结果表明，即使在训练规模有限的情况下，所提出的基于斑块的分级框架也是一种可靠且稳健的方法，可以从图像数据估计大脑年龄。
