Fig. 1. Overview of the study methodology.

a) Illustration of the FaceAge algorithm that uses a single photograph of the face as input. First, a convolutional neural network localizes the face within the photograph, and next, a second convolutional neural network quantifies face features and uses these to predict the FaceAge of the person. b) Overview of the datasets used in this study. The FaceAge algorithm was developed using a training dataset with presumed healthy individuals with the assumption that their age closely approximates biological FaceAge. This dataset was manually curated for individuals of 60 years and older to enhance the dataset quality for the age range of the clinical oncology population. The performance of the algorithm was validated across genders and ethnicities in the UTK dataset. Three independent cohorts of cancer patients covering a large spectrum of cancer patients were used to assess the clinical relevance of the algorithm. c) Overview of the clinical experiments performed in this study to assess the clinical utility of FaceAge. The website https://thispersondoesnotexist.com was used to generate the example face photo.