Magnetic resonance imaging-based pseudo computed tomography using anatomic signature and joint dictionary learning

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiotherapy Physics

Background:

• Magnetic Resonance Imaging (MRI) offers advantages for radiation therapy planning but lacks electron density data.
• Accurate electron density information is essential for precise dose calculations in radiotherapy.
• Computed Tomography (CT) provides electron density but has limitations compared to MRI.

Purpose of the Study:

• To develop and validate a dictionary-learning-based method for deriving electron density information from MRI scans.
• To enable accurate CT image prediction from MRI data for improved radiotherapy treatment planning.
• To assess the feasibility of using this technique for PET attenuation correction in PET/MRI scanners.

Main Methods:

• A joint dictionary learning approach was employed to predict CT patches from corresponding MR image patches.
• The method partitions MR images into patches and predicts CT patches as structured outputs.
• Feature selection was utilized to enhance the robustness of the CT prediction process.
• Predicted CT patches were combined to reconstruct a full CT image prediction from the MRI.

Main Results:

• The method was validated on 14 brain MRI and CT patient datasets.
• Quantitative metrics including Peak Signal-to-Noise Ratio (PSNR), Mean Absolute Error (MAE), Normalized Cross-Correlation (NCC), and Similarity Index (SI) were used.
• High accuracy was demonstrated with mean ± std values for PSNR, MAE, and NCC reported.
• Excellent SIs were achieved for air, soft-tissue, and bone regions, indicating reliable CT prediction.

Conclusions:

• The proposed dictionary-learning method effectively predicts CT images from MRI data, providing essential electron density information.
• This technique holds significant potential for enhancing MRI-based radiation therapy treatment planning.
• The CT image prediction capability can also be applied to improve attenuation correction in PET/MRI systems.