Biologically informed dual deep learning for skeletal maturity prediction in pediatrics

Background and objective Accurate bone age estimation is critical for clinical diagnostics, forensic assessments, and growth research. Traditional methods rely on manual interpretation of radiographs, which is subjective and time-consuming. This Perspective introduces a biologically informed dual deep learning framework that leverages published physiological data to conceptually support bone age prediction. Methods The framework integrates anatomical and developmental knowledge with a dual neural network architecture. One network extracts morphological features from publicly available datasets, while the second captures age-related growth patterns via supervised learning. No new human or animal data were collected. Illustrative simulations and conceptual analyses explore expected model behavior. Results Simulations indicate that incorporating biologically informed priors can improve bone age estimation. Embedding physiological knowledge promotes more stable training, reduces prediction variability, and produces outputs that better align with normative growth trajectories compared to conventional AI models without biological priors. Conclusions We present a theoretically grounded, AI-driven concept for bone age estimation using only published data. Combining biological knowledge with a dual deep learning approach may enhance reproducibility, interpretability, and efficiency. Future work should validate this framework on real-world imaging datasets and assess its integration into automated clinical assessment pipelines.