Generative AI in Orthopedics: An Explainable Deep Few-Shot Image Augmentation Pipeline for Plain Knee Radiographs and Kellgren–Lawrence Grading
This GitHub repository includes all Python codes and implementations to train, generate, and validate synthetic knee radiographs using a few-shot learning pipeline, along with code to train the downstream task of Kellgren–Lawrence (KL) grade classification using only the synthetic radiographs.
The entire dataset (00m/96m) analyzed during the current study is publicly available at the Osteoarthritis Initiative (OAI). You can apply for access here: https://nda.nih.gov/oai.
The complete synthetic knee radiography dataset containing both complete and cropped knee radiographs can be accessed here:
[1] KL01.zip [Download]
[2] KL234.zip [Download]
[3] Bilateral_Knee_XRay.zip [Download]
This work was supported in part by Oracle Cloud credits and related resources provided by Oracle for Research. Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the Oracle for Research.
This contribution is fully explained in the following manuscript, which is now under review at the Journal of the American Medical Informatics Association (JAMIA). Any publication using this work would require to cite the following paper:
[1] Nickolas Littlefield, Soheyla Amirian, Jacob Biehl, Edward G Andrews, Michael Kann, Nicole Myers, Leah Reid, Adolph J Yates, Brian J McGrory, Bambang Parmanto, Thorsten M Seyler, Johannes F Plate, Hooman H Rashidi, Ahmad P Tafti, Generative AI in orthopedics: an explainable deep few-shot image augmentation pipeline for plain knee radiographs and Kellgren-Lawrence grading, Journal of the American Medical Informatics Association, Volume 31, Issue 11, November 2024, Pages 2668–2678, https://doi.org/10.1093/jamia/ocae246