Biologically informed dual deep learning for skeletal maturity prediction in pediatrics

Area of Science:

• Artificial Intelligence in Medicine
• Radiology and Imaging Analysis
• Pediatric Endocrinology

Background:

• Accurate bone age estimation is crucial for clinical diagnostics, forensic science, and growth research.
• Traditional radiographic interpretation is subjective and time-consuming.
• Existing AI models lack biological grounding, potentially limiting accuracy.

Purpose of the Study:

• To introduce a biologically informed dual deep learning framework for bone age prediction.
• To leverage published physiological data for enhanced AI-driven bone age estimation.
• To improve the accuracy, reproducibility, and efficiency of bone age assessment.

Main Methods:

• Developed a dual deep learning framework integrating anatomical and developmental knowledge.
• Utilized one neural network for morphological feature extraction from public datasets.
• Employed a second network for supervised learning of age-related growth patterns.
• Conducted simulations and conceptual analyses without collecting new human or animal data.

Main Results:

• Biologically informed priors significantly improve bone age estimation accuracy.
• Incorporating physiological knowledge leads to more stable AI model training.
• The proposed framework reduces prediction variability and aligns better with normative growth.
• Outputs demonstrate superior performance compared to AI models without biological priors.

Conclusions:

• A theoretically grounded, AI-driven concept for bone age estimation using published data is presented.
• Combining biological knowledge with dual deep learning enhances reproducibility, interpretability, and efficiency.
• Future validation on real-world imaging data and clinical integration are recommended.