The inaugural RSNA machine learning challenge targeted bone age assessment from pediatric hand radiographs, a routine task that determines an important developmental indicator, but one that is “considered tedious, time consuming, and limited by considerable interrater and intrarater variability”. The algorithm submitted by Alex and Mark earned first place, achieving a mean absolute difference (MAD) from the ground truth (a weighted mean of 6 reads) of 4.3 months. As a comparison, the MAD of one of the six pediatric radiologist reviewers to the ground truth, unaided by an algorithm, was 7.0 months. The algorithm led to the development of our software-as-a-medical device product, Physis.
