ESPE Abstracts (2024) 98 RFC2.6

1CHUV, Lausanne, Switzerland. 2UNIL, Lausanne, Switzerland


Purpose: Artificial intelligence (AI) is increasingly used for medical imaging procedures, including automated bone age (BA) evaluation. We aimed to evaluate the agreement between two different BA methods, the IB lab PANDA® and BoneXpert® artificial intelligence methods vs. the manual reference rating in children for BA determination, and to compare the IB lab PANDA performance to the BoneXpert method.

Patients and methods: 219 left hand and wrist radiographs of patients aged 1-19 years were rated with BoneXpert and IB lab PANDA methods and compared to the mean of two experts Greulich-Pyle (GP) based manual rating. We analyzed all patients and divided them into gender, diagnosis, and puberty categories. Accuracy was calculated as Mean Absolute deviation (MAD) and Root Mean Square error (RSME).

Results: Both IB lab PANDA and BoneXpert showed a strong positive correlation to the manual rating (IB lab PANDA to manual, r =0.98, R2 0.96, P <0.001; BoneXpert to manual r =0.98, R2 0.96, P <0.001). Both automated methods had a significant positive bias in all patients. Both methods overestimated BA in boys (BoneXpert 0.28±0.84 years, P <0.001; PANDA 0.49±0.77 years, P <0.001). PANDA also underestimated BA in all girls (-0.22±0.75 years, P = 0.015) exclusively because of post pubertal girls (-0.57±0.86 years, P = 0.003). We found a MAD of 0.64±0.55 years and an RMSE of 0.85 years for IB lab PANDA and a MAD of 0.62±0.52 years, and an RMSE of 0.82 years for BoneXpert compared to the manual assessment. In the subgroup analysis, IB lab PANDA had a significantly better MAD in children during puberty (0.49±0.42 years vs. 0.62±0.49 years; P = 0.012) and inversely, BoneXpert had a significantly better accuracy than IB lab PANDA in children before puberty (0.65±0.57 years vs. 0.78±0.53 years; P = 0.009)

Conclusion: IB lab PANDA® and BoneXpert® have an equivalent agreement to the reference manual BA determination. In boys, the results of both methods should be supervised by a manual rater, as well as in post pubertal girls rated with IB lab PANDA. Both AI systems are valuable AI assist tools in helping clinicians achieve a faster BA determination, reduce inter-rater variability, and leave time for manual assessment of eventual skeletal disorders indicating an underlying disease.

Volume 98

62nd Annual ESPE (ESPE 2024)

Liverpool, UK
16 Nov 2024 - 18 Nov 2024

European Society for Paediatric Endocrinology 

Browse other volumes

Article tools

My recent searches

No recent searches.