[The Effects of Different Adaptive Statistical Iterative Reconstruction-V and Convolution Kernel Parameters on...

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Deep learning auto-segmentation is crucial for radiotherapy planning.
• Image reconstruction parameters can influence segmentation accuracy.
• Understanding these influences is key to reliable automated contouring.

Purpose of the Study:

• To evaluate the impact of adaptive statistical iterative reconstruction-V (ASiR-V) and convolution kernel parameters on CT auto-segmentation stability.
• To assess the performance of deep learning-based segmentation across different reconstruction settings.
• To identify optimal parameters for robust auto-segmentation in pelvic radiotherapy.

Main Methods:

• Twenty patients undergoing pelvic radiotherapy were included.
• CT datasets were generated using varying ASiR-V and kernel parameters.
• Deep learning models automatically segmented soft tissues (bladder, bowelbag, small intestine) and bone structures (femoral heads, femurs, pelvis).
• Segmentation performance was quantified using Dice Similarity Coefficient (DSC) and Hausdorff distance against Filtered Back Projection (FBP).

Main Results:

• Deep learning auto-segmentation demonstrated significant sensitivity to ASiR-V parameter variations.
• Convolution kernel parameters had a less pronounced effect on segmentation stability, particularly for soft tissues.
• Filtered Back Projection (FBP) served as a stable reference for performance evaluation.

Conclusions:

• Parameter selection in CT reconstruction algorithms critically affects auto-segmentation stability.
• Balancing image quality and segmentation quality is essential for clinical application.
• Enhancing deep learning segmentation networks may further improve auto-segmentation robustness.