Big congratulations to HandOsteoVision for taking 1st place in the Launchpad Track at Health Tech Hackathon 2025.
Bone age assessment is essential for diagnosing pediatric growth disorders, but manual reading can vary between clinicians, and many automated tools miss key factors like gender-specific maturation and real-world generalizability.
Their solution, HandOsteoNet, is a deep-learning approach that combines segmentation + attention, supports unilateral & bilateral hand X-rays, and explicitly encodes patient gender.
Multi-source training (RSNA + RHPE), external validation (DHA), explainability with Grad-CAM, and stress testing for imaging artifacts.
Reported MAE reached 3.86 ± 0.37 months, 4.27 ± 1.10, 5.32 ± 0.86, and 4.70 ± 0.50 months.