Dose Reduction in 4-Dimensional Computed Tomography Imaging: Breathing Signal-Guided Deep Learning-Driven Data...

Area of Science:

• Medical Imaging
• Radiation Oncology
• Artificial Intelligence in Medicine

Background:

• Four-dimensional computed tomography (4D CT) is crucial for thoracic tumor radiation therapy planning.
• Current 4D CT protocols often acquire excess data, increasing radiation exposure and deviating from the ALARA principle.
• Optimizing data acquisition is needed to reduce patient dose.

Purpose of the Study:

• To develop and validate a deep learning (DL) model for guided 4D CT data acquisition.
• To minimize radiation exposure by acquiring only necessary projection data.
• To ensure diagnostic image quality is preserved for radiation therapy planning.

Main Methods:

• Retrospective analysis of 1415 breathing signals from 294 patients for DL model training and validation.
• A DL model was trained to predict optimal beam-on events for projection acquisition.
• Independent testing on 104 clinical 4D CT scans, comparing dose-reduced images with reference images using segmentation, registration, and artifact analysis.

Main Results:

• The DL approach reduced beam-on time and imaging dose by a median of 29%, with an average dose reduction of 11.6 mGy.
• Reconstructed dose-reduced images showed minimal differences compared to reference images (high Dice coefficients, small displacement fields).
• Artifact frequency remained unchanged, and tumor segmentation/motion ranges showed only minor deviations, indicating preserved diagnostic quality.

Conclusions:

• The proposed DL-driven data acquisition method effectively reduces radiation exposure in 4D CT.
• The approach maintains diagnostic image quality, crucial for radiation therapy planning.
• This represents a clinically viable and ALARA-compliant solution for 4D CT imaging.