Deep learning-based triple-tracer brain PET scanning in a single session: A simulation study using clinical data.

Area of Science:

• Neuroimaging
• Radiochemistry
• Artificial Intelligence

Background:

• Multiplexed Positron Emission Tomography (PET) enables simultaneous multi-tracer imaging, improving diagnostics and patient comfort.
• Traditional multi-tracer PET protocols involve delays, causing physiological changes and noise.
• Deep learning (DL) and multi-tracer compartment modeling offer advanced solutions for PET imaging.

Purpose of the Study:

• To explore a DL-based single-session triple-tracer brain PET imaging protocol.
• To simplify multi-tracer PET imaging and reduce radiation exposure.
• To evaluate the DL model's ability to separate signals from simultaneous PET tracers.

Main Methods:

• Utilized the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset with amyloid, FDG, and tau PET scans.
• Generated synthetic dual- and triple-tracer images by summing individual tracer scans.
• Developed a DL model (Swin Transformer) to separate signals, validated using cross-validation and image quality metrics.

Main Results:

• The DL model successfully synthesized realistic amyloid and FDG images from triple-tracer scans.
• Clinical evaluation showed high sensitivity for amyloid (92-93%) and moderate for tau (67%) status.
• Quantitative metrics and voxel-wise correlation analysis confirmed strong agreement between synthetic and reference amyloid images.

Conclusions:

• The DL model effectively separates signals from simultaneous triple-tracer PET scans.
• This approach can make multiplex PET scanning feasible in clinical settings.
• The method has the potential to reduce scan time, radiation dose, and enhance patient comfort.