The RSNA Pediatric Bone Age Machine Learning Challenge.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• The Radiological Society of North America (RSNA) initiated a challenge to explore machine learning (ML) and artificial intelligence (AI) applications in medical imaging.
• Accurate bone age assessment is crucial for pediatric growth and development evaluation.

Purpose of the Study:

• To foster collaboration and identify innovators in medical imaging AI.
• To develop and evaluate ML models for accurate bone age determination from pediatric hand radiographs.

Main Methods:

• A curated dataset of 14,236 pediatric hand radiographs was provided for model development and testing.
• Participants utilized ML techniques, predominantly deep neural networks and convolutional neural networks (CNNs).
• The primary evaluation metric was the mean absolute distance (MAD) in months between estimated and reference bone age.

Main Results:

• 260 individuals/teams registered, with 105 submissions uploaded by 48 unique users.
• The top-performing models achieved a mean absolute distance (MAD) as low as 4.2 months.
• Deep neural network-based approaches, particularly CNNs, were widely adopted and successful.

Conclusions:

• The RSNA Pediatric Bone Age Machine Learning Challenge demonstrated the effectiveness of a coordinated approach to solving complex medical imaging problems.
• Future ML challenges are expected to accelerate the development of AI tools, enhancing diagnostic accuracy and patient care.
• The challenge successfully catalyzed collaboration and innovation in AI for pediatric bone age assessment.