Improving Low-Dose Pediatric Abdominal CT by Using Convolutional Neural Networks

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Low-dose pediatric CT scans are crucial for minimizing radiation exposure.
• Traditional image reconstruction methods like filtered back projection (FBP) can result in suboptimal image quality.
• Iterative reconstruction (IR) algorithms improve image quality but are not always available on all scanners.

Purpose of the Study:

• To evaluate the efficacy of convolutional neural networks (CNNs) in enhancing the image quality of low-dose pediatric abdominal CT scans.
• To determine if CNN-based postprocessing can simulate the results of iterative reconstruction (IR) on images reconstructed with filtered back projection (FBP).

Main Methods:

• A residual CNN was trained to predict image noise reduction by comparing FBP and IR reconstructed images.
• CNN-based postprocessing was applied to low-dose pediatric CT datasets acquired on a scanner limited to FBP reconstruction.
• Objective noise measurements and subjective image reviews by two pediatric radiologists were performed to assess image quality.

Main Results:

• CNN-enhanced images showed a 31% reduction in image noise compared to FBP images (P < .001).
• Radiologists preferred CNN images for overall image quality, citing improvements in low contrast, image noise, and artifacts.
• While spatial resolution was comparable, CNNs effectively improved other key image quality metrics.

Conclusions:

• Well-trained CNNs can significantly improve image quality in the image space for FBP reconstructed CT scans.
• This AI-driven approach may enable radiation dose reduction or enhanced image quality on scanners limited to FBP reconstruction.
• CNNs offer a promising solution for improving diagnostic confidence in pediatric CT imaging.