A system-level DOI discrimination method based on SSDA for a brain-dedicated DOI-PET scanner.

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Instrumentation

Background:

• Depth-of-interaction (DOI) information is crucial for enhancing spatial resolution in positron emission tomography (PET) imaging.
• Current deep learning DOI methods necessitate extensive detector-specific labeled data, hindering practical system-level calibration.

Purpose of the Study:

• To develop and evaluate a system-level DOI discrimination method for brain-dedicated DOI-PET scanners using semi-supervised domain adaptation (SSDA).
• To reduce the calibration burden by requiring labeled data from only a small subset of detectors.

Main Methods:

• A novel SSDA approach was applied to a brain-dedicated DOI-PET scanner.
• The method was validated on a two-layer lutetium yttrium oxyorthosilicate (LYSO) detector array and a 72-detector PET prototype.
• Performance was assessed using flood image quality and spatial resolution measurements with a 22Na point source.

Main Results:

• The SSDA method achieved 98.21% accuracy, comparable to fully supervised methods, using labeled data from just one detector.
• High flood image quality (k values of 4.82 ± 0.55 and 6.65 ± 0.65) and minimal misclassified clusters (0.80%) were observed.
• Excellent spatial resolution was achieved, with average radial resolutions of 1.83 mm (2-level DOI) and 1.39 mm (4-level DOI).

Conclusions:

• The proposed SSDA method offers an efficient, accurate, and scalable solution for system-level DOI calibration.
• This approach is practical for next-generation, high-performance brain-dedicated PET scanners.
• It significantly reduces calibration costs and labor while maintaining high performance.