Deep learning for contrast-enhanced MRI in pediatric brain imaging

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology
• Pediatric Neuroradiology

Background:

• Deep learning algorithms are increasingly used for medical image analysis.
• Contrast amplification in brain MRI can improve lesion detection.
• Generalizability of AI models across different patient populations is crucial.

Purpose of the Study:

• To evaluate the cross-population generalization of a deep learning algorithm for contrast amplification in pediatric brain MRI.
• To assess the performance of an adult-trained algorithm on pediatric subjects, including infants (0-2 years).

Main Methods:

• A retrospective study of 22 pediatric brain tumor cases (0-17 years).
• Input: T1-weighted pre- and post-contrast MRI images.
• Output: Contrast-amplified images processed with HDR algorithm.
• Evaluation: Quantitative (CNR, CEP, LBR) and qualitative (Likert scale) assessments by neuroradiologists.
• Anatomical similarity assessed using SSIM and log-Jacobian range.

Main Results:

• Contrast-amplified images showed significant increases in CNR (+186.5%), LBR (+61.9%), and CEP (+110.4%).
• Qualitative assessment revealed comparable lesion visualization, with majority preference for amplified images by neuroradiologists (54.5%–81.8%).
• High anatomical similarity (average SSIM: 0.98) with no significant anatomical differences.

Conclusions:

• The deep learning algorithm effectively enhances quantitative contrast metrics in pediatric brain MRI.
• The algorithm demonstrates cross-population applicability, performing well on pediatric patients despite being trained on adult data.
• Contrast amplification shows promise for improving diagnostic accuracy in pediatric brain MRI.