Transfer Learning From Hand-Trained Deep Learning Models to Estimate Bone Age From Knee Radiographs

The authors aimed to develop a deep learning model for estimating bone age from knee radiographs, addressing the limitations of the traditional Greulich and Pyle atlas which requires additional imaging and radiation exposure. Their model demonstrated a mean absolute error of 5.02 months, significantly outperforming existing methods, and offers a promising tool for orthopaedic surgeons and radiologists in assessing skeletal maturity in young patients. The study highlights the potential for automated bone age estimation to enhance clinical decision-making, pending further validation and refinement.