Three-dimensional structure tensor based PET/CT fusion in gradient domain.

Area of Science:

• Medical Imaging
• Computer Vision
• Oncology

Background:

• Gradient-based image fusion effectively uses structure tensors for 2D image fusion.
• Existing methods may not optimally capture 3D details in medical imaging.

Purpose of the Study:

• To generalize and apply 3D gradient-based image fusion for non-small cell lung cancer PET/CT imaging.
• To develop a novel 3D structure tensor feature for enhanced detail preservation.

Main Methods:

• Proposed a novel 3D structure tensor based feature for weighted structure tensor construction.
• Deduced the fusion gradient domain from a rank one tensor approximation.
• Obtained the final PET/CT fusion image by solving a Poisson equation.

Main Results:

• The proposed method significantly increased average information entropy (13.5%) and average gradient measure (42.3%) compared to wavelet transform fusion.
• Effectively preserved lung vessel structures and sphere-like lesion details.
• Produced clear, stable, and well-consistent fusion images.

Conclusions:

• The novel 3D structure tensor based fusion method offers superior performance for lung cancer PET/CT image fusion.
• This technique enhances diagnostic accuracy by preserving critical anatomical and pathological details.
• The method provides a robust and effective approach for 3D medical image fusion.