Fully-automated cartilage segmentation using deep learning - data from the osteoarthritis initiative

Segmentation of anatomical tissues from medical images is extremely time consuming and is prone to human error. Historically, automated methods have produced insufficient accuracy and have been computationally expensive. This investigation tested a custom two-stage deep learning framework that achieved the best-reported segmentation accuracies, for all knee cartilage regions, in less than 3 mins.