Development of Amyloid PET Analysis Pipeline Using Deep Learning-Based Brain MRI Segmentation-A Comparative...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine

Background:

• Amyloid PET scans are crucial for diagnosing Alzheimer's disease by detecting amyloid-beta plaques.
• Current quantitative analysis of PET scans is complex and time-consuming, limiting clinical application.

Purpose of the Study:

• To introduce a novel deep learning-based method for Standardized Uptake Value Ratio (SUVR) quantification in amyloid PET scans.
• To simplify preprocessing and reduce analysis time while maintaining accuracy.

Main Methods:

• A deep learning approach was developed for SUVR quantification using two amyloid ligands.
• The method's performance was evaluated against established tools like PETSurfer and PMOD.
• Key metrics included intra-class correlation coefficients, global SUVR differences, AUC-ROC, processing time, and registration failure rate.

Main Results:

• The deep learning method demonstrated high consistency with PETSurfer (ICC=0.97) and PMOD (ICC=0.99).
• Clinically acceptable differences in global SUVRs were observed (-0.02 to 0.04).
• The method achieved an AUC-ROC > 0.95 for amyloid-positive assessment, with rapid processing and a low failure rate (1%).

Conclusions:

• The proposed deep learning method provides accurate and reliable SUVR quantification for amyloid PET scans.
• Its efficiency in processing time and low failure rate make it suitable for clinical practice.
• This tool can enhance the diagnostic capabilities for cognitive impairment suspected to be Alzheimer's disease.