Technical Note: Deep learning based MRAC using rapid ultrashort echo time imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Accurate attenuation correction is crucial for quantitative PET/MR imaging.
• Traditional CT-based attenuation correction in PET/MR requires separate scans and introduces registration challenges.
• Deep learning offers a potential solution for automated and rapid attenuation correction.

Purpose of the Study:

• To evaluate the feasibility of a novel deep learning framework for MR-based attenuation correction (MRAC) in PET/MR imaging.
• To develop a fully automated method for generating a pseudo CT image using rapid MR acquisitions.
• To enable robust and accurate PET quantitation without a separate CT scan.

Main Methods:

• A novel framework utilizing convolutional neural networks for MR-based attenuation correction was developed.
• Rapid MR acquisition using dual echo ramped hybrid encoding (dRHE) provided ultrashort echo time (UTE), fat, and water images within 35 seconds.
• Deep learning models, trained on CT-derived labels, estimated tissue labels (air, soft tissue, bone) from UTE images, refined by conditional random fields and further separated into fat/water components.

Main Results:

• The deep learning model achieved high Dice similarity coefficients for tissue labels: air (0.76), soft tissue (0.96), and bone (0.88).
• The proposed MRAC method resulted in relative PET errors of less than 1% in most brain regions.
• The generated pseudo CT images were suitable for PET attenuation correction.

Conclusions:

• The developed MRAC method, leveraging deep learning and efficient dRHE acquisition, provides reliable PET quantitation.
• The framework enables accurate and rapid pseudo CT generation, enhancing the efficiency of PET/MR imaging.
• This approach demonstrates the potential of AI in advancing quantitative medical imaging techniques.