Using artificial intelligence models to evaluate envisaged points initially: A pilot study.

Area of Science:

• Radiology
• Biomedical Engineering
• Orthopedics

Background:

• Skeletal maturity assessment is crucial in clinical practice.
• Hand-wrist radiographs (HWRs) offer valuable indicators of skeletal development.
• Standardized classification of phalangeal morphology is needed.

Purpose of the Study:

• To validate anatomical landmarks for classifying phalangeal bone morphology in HWRs.
• To develop and evaluate classical neural network (NN) classifiers for skeletal maturity assessment.
• To assess the reliability of epiphysis-diaphysis relationship classifications.

Main Methods:

• A dataset of 136 HWRs was used to train classical neural network (NN) classifiers (NN-1 and NN-2 with 5-fold cross-validation).
• 22 anatomical landmarks were identified on phalanges (PP3, MP3, DP3, MP5) and epiphysis-diaphysis relationships were categorized.
• 18 ratios and 15 angles were extracted, and model performance was evaluated using agreement, Kappa coefficients, precision, recall, F1-score, and accuracy.

Main Results:

• Method error ranged from Cohen's Kappa (cκ) of 0.7-1.
• Overall classification performance ranged from 82.14% to 89.29%.
• Average performance for NN-1 and NN-2 models was 85.71% and 85.52%, respectively, with promising results except in regions with limited samples.

Conclusions:

• The validated anatomical landmarks are suitable for future skeletal maturity studies.
• Neural network models demonstrate effective classification of phalangeal morphology.
• This approach offers a reliable method for radiological skeletal maturity assessment.