Novelty Detection for Bone Age Anomaly Identification Using Self-Supervised Learning.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Radiology
• Pediatric Skeletal Development

Background:

• Early detection of pediatric skeletal anomalies is vital for diagnosing growth disorders.
• Traditional bone age assessment models overlook skeletal anomalies present in X-rays.
• Rare anomalies create imbalanced datasets, hindering traditional supervised learning.

Purpose of the Study:

• To develop a novel self-supervised learning (SSL) framework for robust bone age anomaly detection in pediatric X-rays.
• To leverage Vision Transformers (ViTs) for feature extraction from unlabeled X-ray data.
• To enhance diagnostic accuracy for rare skeletal disorders through improved anomaly detection.

Main Methods:

• Proposed a self-supervised learning (SSL) framework utilizing Vision Transformers (ViTs).
• Employed SSL pretraining on unlabeled X-ray data for feature extraction.
• Utilized novelty detection techniques for identifying skeletal abnormalities, validated on an expert-curated dataset.

Main Results:

• SSL-based models significantly outperformed non-SSL methods in anomaly detection.
• Achieved state-of-the-art classification accuracy (98.2%) and Area Under the Curve (AUC) (99.8%).
• Demonstrated improved sensitivity for detecting rare skeletal abnormalities.

Conclusions:

• The proposed SSL framework offers a scalable and data-efficient solution for pediatric radiology.
• This approach enhances diagnostic accuracy for rare skeletal disorders, facilitating early intervention.
• ViT-based SSL effectively addresses data imbalance issues in detecting rare anomalies.