Validation of deep learning-based CT image reconstruction for treatment planning

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Artificial Intelligence in Medicine

Background:

• Deep learning-based CT image reconstruction (DLR) is an advanced technique for generating CT images.
• Evaluating novel reconstruction methods is crucial for optimizing radiotherapy treatment planning.

Purpose of the Study:

• To assess the utility of DLR in radiotherapy applications.
• To compare DLR with hybrid iterative reconstruction (H-IR) regarding image quality metrics.

Main Methods:

• CT data acquired using a large-bore CT system and an electron density phantom.
• Comparison of CT values, image noise, and CT value-to-electron density conversion tables between DLR and H-IR across various radiation doses.
• Evaluation of three DLR reconstruction strengths: Mild, Standard, and Strong.

Main Results:

• DLR exhibited less variation in CT values and lower image noise compared to H-IR, particularly in low-dose regions.
• CT value differences between DLR and H-IR were minimal (under 10 HU) at doses of 100 mAs and above.
• Higher DLR reconstruction strengths (Standard, Strong) showed significant noise reduction, surpassing the Mild setting.

Conclusions:

• DLR provides stable CT values and low image noise for diverse materials, even at reduced radiation doses.
• The Standard and Strong DLR settings offer substantial noise reduction, indicating their clinical relevance.
• DLR is a promising tool for enhancing treatment planning in radiotherapy utilizing large-bore CT systems.