Few shots transfer learning for universal SPECT denoising across diverse acquisition protocols.
Authors
Affiliations (6)
Affiliations (6)
- Institute of Magnetic Resonance and Molecular Imaging in Medicine, East China Normal University, Shanghai, China, No. 3617 North Zhongshan Road, Putuo District, Shanghai, China, Shanghai, 200062, China.
- Nuclear Medicine Department, Southwest Hospital (the First Affiliated Hospital), Third Military Medical University (Army Medical University), Chongqing, China, No.29 Gaoyanzheng street, Shapingba District, Chongqing, China, Chongqing, 400038, China.
- RadioDynamic Medical, Shanghai, China, No. 339 Tongpu Road, Putuo District, Shanghai, China, Shanghai, 200062, China.
- Laboratory for Intelligent Medical Imaging, Tsinghua Cross-Strait Research Institute, Xiamen, China, No.516 North Qishan Road, Huli Distruct, Xiamen, China, Xiamen, 361000, China.
- RadioDynamic Medical, Shanghai, China, No.602 Tongpu Road, Putuo District, Shanghai, China, Shanghai, 200062, China.
- Institute of Magnetic Resonance and Molecular Imaging in Medicine, East China Normal University, Shanghai, China, No.516 North Qishan Road, Huli Distruct, Xiamen, China, Shanghai, 200062, China.
Abstract

Accelerated Single Photon Emission Computed Tomography (SPECT) imaging, achieved by reducing either the number of projection angles or the acquisition time per angle, enhances clinical workflow efficiency but introduces elevated noise. Although deep learning-based methods promises to overcome this limitation, their practical application is hindered by the reliance on large datasets makes developing separate models for each specific acceleration protocol infeasible.
Method: 
SPECT bone scans from 103 patients were acquired under a standard scan (60 views, 12 seconds/view, 60v12s) followed by a fast scan using one of five acceleration protocols (60v6s, 60v3s, 30v12s, 30v6s, 16v12s). A U-Net-based reconstruction framework was implemented using three strategies: 1) single model, trained on individual acceleration protocol), 2) base model, trained on aggregated datasets from five acceleration protocols, and 3) transfer model, fine-tuned from the base model for protocol-specific optimization. Pixel-level accuracy and structural similarity were assessed using Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM) metrics, and maximum pixel value of lesions. Clinical evaluation of image quality, radionuclide detail, artifacts, and diagnostic confidence was conducted using a 5-point system.
Results: 
Quantitative evaluation showed the transfer model achieved better PSNR and SSIM across all protocols ( highest 48.02 PSNR and 0.9918 SSIM in 30v6s protocol). Qualitative analysis confirmed enhanced structural fidelity. Clinical evaluations rated the transfer model highest across metrics, with scores of 4.667±0.508 (image quality, 30v12s), 4.800±0.250 (radionuclide detail, 60v6s), 1.150±0.173 (artifact reduction, 60v6s), and 4.800±0.250 (diagnostic confidence, 60v6s), surpassing full-scan results in most cases.
Conclusion: 
The proposed transfer learning framework effectively addressed data scarcity and improved reconstruction performance across diverse SPECT acceleration scenarios. The adoption of a transfer learning strategy mitigates data scarcity by utilizing shared features and fine-tuning for specific protocols. The framework demonstrates potential for integration into fast SPECT workflows, facilitating reliable use across diverse imaging scenarios.