Direct PET-to-CT Generation for Attenuation Correction: A Slice-to-Slice Continual Transformer Segmentation-Aware...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiological Physics

Background:

• Direct synthetic computed tomography (CT) generation from positron emission tomography (PET) is vital for PET attenuation correction, providing structural information for functional imaging.
• Current methods like PET/CT and indirect PET/MR-CT involve intermediate steps and supplementary equipment, introducing errors and complexity.
• Direct PET-to-CT translation offers advantages by bypassing intermediate processes and eliminating the need for additional hardware, reducing scan duration and complexity.

Purpose of the Study:

• To address challenges in direct PET-to-CT translation, including spatial resolution mismatches and semantic differences between functional PET and structural CT data.
• To propose a novel 2D hierarchical method, the Slice-to-Slice Continual Transformer-Segmentation-aware (S2SCT-SA) Network, for accurate CT generation from PET data.
• To enhance the spatial resolution of generated CT images and improve the accuracy of PET attenuation correction.

Main Methods:

• A slice-continual network within the S2SCT-SA architecture learns semantic transformation knowledge from PET slices to CT slices.
• A segmentation-aware network component captures spatial correlations both within and between slices to improve CT spatial resolution.
• The proposed 2D hierarchical method facilitates domain conversion between functional (PET) and structural (CT) imaging.

Main Results:

• The S2SCT-SA Network demonstrates superior performance in CT generation compared to mainstream methods.
• The method achieves improved PET attenuation correction accuracy, validated by visual inspection and quantitative metrics.
• Experimental results show enhanced spatial resolution in the synthetically generated CT images.

Conclusions:

• The proposed S2SCT-SA Network effectively overcomes spatial and semantic challenges in direct PET-to-CT translation.
• Direct PET-to-CT generation using the S2SCT-SA method shows significant promise for clinical applications in PET attenuation correction.
• This approach offers a more efficient and accurate alternative to existing multimodal imaging techniques for PET attenuation correction.