Fig. 1: The ARDA of LCR images.

The ARDA is generated from LCR images and a pre-trained deep learning ageing feature extractor. a Instances in LCR images. The parietal and frontal instances are not labeled because the ARDA is barely distributed on them. b Illustration of the quantitative distribution of ARDA on instances in LCR images. The three dimensions are age, instance and quantified ARDA. The contours and position of each instance are shown in a. The quantified ARDA of each instance is the mean value of the average ageing salience of the ageing-significant regions it contains. From the bottom to the top are the cases in which the threshold of ageing-significant region is set to the median, 75-th percentile, and 90-th percentile of corresponding ARDA, respectively. In the dimension of instances, Cx V and Cx S represent the x-th vertebral body and the x-th spinous processes of the cervical spine, respectively. c The ARDA map of LCR images. To visualize the distribution of ARDA and ageing regions, we mapped the average ageing salience to a color ranging from blue to red. The closer the color of the pixel in the ARDA map is to red, the larger the average ageing salience of the pixel. We display the LCR image of a randomly selected 28-year-old subject overlaid with its corresponding ageing salience map to show the relative position between the ARDA map and the LCR image. Source data are provided as a Source Data file.