Recovering Image Quality in Low-Dose Pediatric Renal Scintigraphy Using Deep Learning.

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Pediatric renal scintigraphy often requires significant radiation exposure.
• Reducing radiation dose is crucial for patient safety, especially in pediatric populations.
• Image noise is a major challenge in low-dose scintigraphy.

Purpose of the Study:

• To develop and evaluate an advanced image enhancement strategy for pediatric renal scintigraphy.
• To assess the feasibility of reducing radiation doses while maintaining diagnostic image quality.
• To investigate the effectiveness of deep learning-based denoising neural networks.

Main Methods:

• Utilized a public dynamic renal scintigraphy database.
• Evaluated four denoising neural networks: DnCNN, UDnCNN, DUDnCNN, and AttnGAN.
• Assessed image quality using kidney signal-to-noise ratio (SNR) and multiscale structural similarity (MS-SSIM).
• Simulated radiation dose reduction by using 50% of the acquired data.

Main Results:

• All evaluated neural networks demonstrated noise reduction capabilities.
• UDnCNN achieved the optimal balance between SNR and MS-SSIM, showing the most significant image quality improvements.
• Deep learning enhancement allowed 50% of acquired frames to yield results comparable to the full dataset.
• The proposed method suggests feasibility in reducing patient radiation exposure.

Conclusions:

• Deep learning-based neural networks can significantly enhance renal scintigraphic image quality.
• The UDnCNN network shows promise for improving image quality in low-dose pediatric renal scintigraphy.
• This approach facilitates high-quality imaging with reduced radiation doses, benefiting pediatric patients.