An accurate pediatric bone age prediction model using deep learning and contrast conversion

Area of Science:

• Radiology
• Artificial Intelligence
• Pediatric Endocrinology

Background:

• Accurate bone age assessment is crucial for pediatric growth monitoring and clinical decision-making.
• Traditional methods for bone age determination can be subjective and time-consuming.

Purpose of the Study:

• To develop an accurate pediatric bone age prediction model using deep learning and contrast conversion techniques.
• To enhance clinical decision-making in pediatric growth assessment.

Main Methods:

• Utilized various deep learning models (CNN, ResNet50, VGG19, Inception V3, Xception) trained on pediatric hand X-ray images.
• Applied contrast conversion techniques (fuzzy contrast enhancement, CLAHE, HE) for image preprocessing.
• Evaluated model performance using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).

Main Results:

• The Xception model achieved the best performance with an MAE of 41.12.
• Histogram equalization (HE) improved image quality, enhancing SNR and contrast.
• Bone age prediction accuracy improved, with MAE decreasing from 2.11 to 0.24 and RMSE from 0.21 to 0.02 after preprocessing.

Conclusions:

• Deep learning models, particularly Xception, show promise for accurate pediatric bone age prediction.
• Image preprocessing techniques like HE can enhance image quality but do not significantly alter prediction performance.
• The findings support the integration of AI-powered tools for improved pediatric growth assessment.