MMD-Net: Image domain multi-material decomposition network for dual-energy CT imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Science

Background:

• Multi-material decomposition is crucial for dual-energy CT (DECT) imaging.
• Conventional algorithms often face limitations in accuracy and performance.

Purpose of the Study:

• To introduce a novel deep neural network, MMD-Net, for enhanced multi-material decomposition in DECT.
• To improve the accuracy and performance of DECT imaging through advanced computational methods.

Main Methods:

• Developed MMD-Net, a deep neural network comprising Net-I for material triangle distinction and Net-II for predicting effective attenuation coefficients.
• Validated MMD-Net using benchtop and clinical DECT imaging experiments.
• Quantitatively evaluated decomposition accuracy, edge spreading function, and noise power spectrum.

Main Results:

• MMD-Net effectively suppresses image noise compared to conventional multiple material decomposition (MMD) algorithms.
• Outperformed iterative MMD approaches in maintaining decomposition accuracy, image sharpness, and high-frequency content.
• Generated high-quality material decomposition images.

Conclusions:

• A high-performance MMD-Net has been developed for DECT imaging.
• The proposed network offers superior results for multi-material decomposition tasks.