Table 6. Performance comparison of state-of-the-art models with the proposed model.

Reference	Dataset	Method	Features	Limitations
Seok et al. (2012)	24 GP female images	scale-invariant feature transform, singular value decomposition Fully connected neural network	SVD-based fixed-sized features vectors from the SIFT description	Lack of robustness to real photos; sparse training and validation data
Somkantha, Theera-Umpon & Auephanwiriyakul (2011)	180 images Cao et al. (2000)	Canny edge detection Fuzzy classification	Morphological features regarding carpal bones	Not applicable to children above the age of seven
Zhang, Gertych & Liu (2007)	205 images Cao et al. (2000)	Canny edge detection Fuzzy classification	Morphological features regarding carpal bones (Capitate Hamate)	Children older than 5 years old for girls and 7 years old for boys are not eligible.
Thodberg et al. (2008)	1,559 images from multiple sources	Active appearance model	Features regarding shapes, intensity, and texture of RUS bones	Prone to excessive noise in photos when input is based on chronological age
Our work	images	Customized CNN	Data-driven, automatically extracted features