Multimodal imaging-based material mass density estimation for proton therapy using supervised deep learning.

Area of Science:

• Medical Imaging
• Deep Learning
• Physics-Informed AI

Background:

• Accurate patient mass density mapping is crucial for reducing proton range uncertainty in radiotherapy.
• Current methods for deriving mass density from CT scans have limitations.

Purpose of the Study:

• To develop a physics-constrained deep learning-based multimodal imaging (PDMI) framework.
• To integrate physics, deep learning, MRI, and dual-energy CT (DECT) for accurate mass density map generation.

Main Methods:

• Developed a PDMI framework incorporating physics insights and deep learning.
• Utilized MRI and DECT imaging data for training and validation.
• Compared physics-constrained (PRN) and non-constrained (RN) deep learning models.

Main Results:

• The PDMI framework accurately generated mass density maps for various tissue surrogates.
• Physics-constrained models (PRN-MR-DE) demonstrated superior accuracy compared to non-constrained models (RN-MR-DE).
• Patient data showed PRN-MR-DE predicted soft tissue and bone densities within expected ranges.

Conclusions:

• The PDMI framework effectively generates accurate mass density maps using multimodal imaging.
• Physics-informed deep learning enhances model performance for improved accuracy.
• Accurate mass density maps hold potential for improving proton range uncertainty.