Image reconstruction and elongation artifact reduction for a dual-panel dedicated prostate PET scanner

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Artificial Intelligence in Healthcare

Background:

• Active research in organ-specific PET scanners aims to reduce costs and improve lesion detection.
• Innovations in hardware and software drive the development of high-resolution PET imaging.

Purpose of the Study:

• Investigate and compare image reconstruction strategies for a dual-panel prostate-dedicated PET scanner (ProVision).
• Evaluate reconstruction methods to enhance angular coverage and image quality in limited-angle systems.

Main Methods:

• Developed and optimized a list-mode MLEM algorithm with multi-ray modeling.
• Compared classical MLEM, MLEM with PSF modeling, hybrid list-mode reconstruction, and a Swin-UNETR deep learning model.
• Assessed performance using contrast recovery coefficient (CRC), contrast-to-noise ratio (CNR), and contrast-to-noise consistency (CNC) on NEMA and anthropomorphic phantoms.

Main Results:

• The Swin-UNETR deep learning model achieved the highest CNR and CNC in simulations, particularly for small spheres.
• In experimental phantom studies, deep learning and hybrid methods showed improved CNR over standard MLEM for both large (10 mm) and small (8 mm) lesions.
• Swin-UNETR demonstrated superior CNC for smaller lesions, indicating its effectiveness in challenging imaging scenarios.

Conclusions:

• PET scanner-adapted reconstruction combined with deep learning refinement significantly improves image quality.
• These advanced reconstruction techniques are beneficial for dedicated, limited-angle PET systems, enhancing diagnostic capabilities.