Deep learning-based material decomposition of iodine and calcium in mobile photon counting detector CT

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Photon-counting detector (PCD)-based computed tomography (CT) offers advantages over conventional CT.
• Energy discrimination in PCD-CT provides material-specific information, enabling material decomposition (MD).

Purpose of the Study:

• To develop and evaluate a deep learning-based material decomposition method using live animal data.
• To improve the accuracy and reduce noise in material decomposition imaging.

Main Methods:

• A deep learning strategy, MD-Unet, based on a Unet architecture was developed for material decomposition.
• The model was trained using data from three energy bins, incorporating a pretrained model with simulation data and augmentation to address data insufficiency.
• The trained network was applied to live animal data for evaluation.

Main Results:

• MD-Unet demonstrated more accurate material decomposition imaging compared to conventional methods.
• The network achieved improved material decomposition ability and significantly reduced image noise.
• The method showed potential for enhancing image quality similar to contrast agents, possibly reducing required doses.

Conclusions:

• Deep learning-based material decomposition using MD-Unet is effective for live animal data.
• MD-Unet offers enhanced precision, noise reduction, and potential for reduced contrast agent administration in clinical settings.
• This technology holds significant clinical value for improving CT imaging quality and patient safety.