CT-free attenuation and Monte-Carlo based scatter correction-guided quantitative ⁹⁰Y-SPECT imaging for improved dose...

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Artificial Intelligence in Medicine

Background:

• Quantitative 90Y SPECT imaging is crucial for accurate dose calculation in selective internal radiation therapy (SIRT).
• Traditional methods for attenuation correction (AC) and scatter correction (SC) often rely on CT, which may be unavailable or unreliable.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for CT-free AC and Monte Carlo-based SC in 90Y SPECT imaging.
• To improve voxel-level dose calculation accuracy for SIRT patients.

Main Methods:

• A modified 3D shifted-window UNet Transformer (Swin UNETR) architecture was used to train DL models for AC, SC, and joint AC/SC (ASC).
• Data from 190 patients undergoing 90Y SIRT was utilized, with models trained and validated using cross-validation and an unseen test set.
• Performance was assessed by comparing DL-generated dose maps against reference dose maps using quantitative metrics and Gamma analysis.

Main Results:

• DL models achieved high performance in AC, SC, and ASC tasks, with excellent quantitative metrics (e.g., high SSIM and PSNR) and around 98% agreement in voxel-level Gamma evaluations across different criteria.
• Mean absolute errors for tumor and liver dosimetry were within acceptable ranges, demonstrating the models' effectiveness.

Conclusions:

• Task-specific DL models for CT-free AC, SC, and ASC were successfully developed for 90Y SPECT imaging.
• These models show promise for enhancing dosimetry in clinical scenarios where CT data is compromised, potentially improving treatment planning and delivery.