Bone age determination using only the index finger: a novel approach using a convolutional neural network compared...

Area of Science:

• Radiology
• Artificial Intelligence
• Pediatric Imaging

Background:

• Convolutional neural network (CNN) models show superior accuracy in bone age determination compared to human radiologists.
• Current methods often require analysis of the entire hand radiograph.

Purpose of the Study:

• To evaluate the accuracy of CNN models and radiologists in predicting bone age using only the index finger.
• To compare index finger-only bone age assessment with whole-hand analysis.

Main Methods:

• Trained CNN models on whole-hand and index-finger cropped radiographs from the RSNA pediatric bone age challenge dataset.
• Compared CNN model performance against ground truth and consensus bone age determined by three pediatric radiologists.

Main Results:

• CNN models achieved similar mean absolute differences for whole-hand (4.7 months) and index-finger (5.1 months) bone age.
• Both CNN models significantly outperformed radiologists using single-finger radiographs (8.0 months).

Conclusions:

• CNN-based bone age determination from the index finger is comparable to whole-hand analysis by both CNNs and radiologists.
• The index finger alone provides sufficient information for accurate bone age assessment using CNNs, potentially improving radiologist workflow.