Pediatric Bone Age Prediction Using Deep Learning

This paper presents a deep-learning approach for pediatric bone-age prediction from hand X-rays using more than 12,000 images from the RSNA bone age dataset. It compares EfficientNetB0, EfficientNetB4, and EfficientNetB4 enhanced with an additive-attention mechanism. According to the abstract, the B4 and attention-enhanced variants generally track the original ages more closely than B0. The work describes preprocessing X-rays into three-channel images and reports training and validation learning curves. However, the supplied abstract does not provide exact performance metrics, uncertainty estimates, subgroup analysis, or external validation, so the clinical significance of the reported improvement remains difficult to assess.