The study of non-complete-ring positron emission tomography (PET) detection method

Positron Emission Tomography (PET) is a vital molecular imaging tool widely used in medical diagnosis and treatment evaluation. Traditional PET systems typically rely on complete detector rings to achieve full angular coverage for uniform and statistically robust sampling of coincidence events. However, incomplete-ring PET scanners have emerged in various scenarios due to hardware failures, cost constraints, or specific clinical needs. In such cases, conventional reconstruction algorithms often suffer from performance degradation due to reduced data completeness and geometric inconsistencies. This thesis proposes a coarse-to-fine reconstruction framework for incomplete-ring PET scanners. The framework first employs an Attention U-Net model to recover complete sinograms from incomplete ones, then uses the OSEM algorithm for preliminary reconstruction, and finally applies a two-stage architecture comprising a Coarse Prediction Module (CPM) and an Iterative Refinement Module (IRM) for fine reconstruction. Our approach utilizes neighboring axial slices and spectral transform features as auxiliary guidance at the input level to ensure spatial and frequency domain consistency, and integrates a contrastive diffusion strategy at the output level to improve correspondence between low-quality PET inputs and refined PET outputs. Experimental results on public and in-house brain PET datasets demonstrate that the proposed method significantly outperforms existing approaches in metrics such as PSNR (35.6421 dB) and SSIM (0.9588), successfully preserving key anatomical structures and tracer distribution features, thus providing an effective solution for incomplete-ring PET imaging.