Deep learning-based amyloid PET harmonization to predict cognitive decline in non-demented elderly.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Gerontology

Background:

• Conventional amyloid PET harmonization across tracers lacks robustness.
• Accurate prediction of cognitive decline is crucial for early Alzheimer's disease (AD) intervention.

Purpose of the Study:

• To evaluate deep learning-based harmonization of amyloid PET in predicting conversion from cognitively unimpaired (CU) to mild cognitive impairment (MCI) and MCI to AD.
• To assess the prognostic performance of a deep learning-derived amyloid PET probability score (DL-ADprob).

Main Methods:

• Developed a deep learning model to classify AD-dementia vs CU across multiple amyloid PET tracers using data from ADNI, Japanese ADNI, and AIBL cohorts (n=1050).
• Evaluated DL-ADprob's prognostic value in predicting cognitive decline in ADNI-MCI (n=451) and HABS-CU (n=271) participants over 4 years.
• Calculated intraclass correlation coefficients (ICCs) for DL-ADprob across tracers in the GAIN dataset (n=155).

Main Results:

• DL-ADprob was independently prognostic in both ADNI-MCI (P<.001) and HABS-CU (P=.048) cohorts.
• Incorporating DL-ADprob improved prognostic performance in both ADNI-MCI (tdAUC 0.758 to 0.782) and HABS-CU (tdAUC 0.846 to 0.870) groups.
• DL-ADprob demonstrated high consistency across tracers (ICCs 0.913-0.935).

Conclusions:

• Deep learning-based harmonization of amyloid PET significantly enhances the prediction of cognitive decline in non-demented individuals.
• DL-ADprob shows potential as a complementary tool to conventional amyloid PET measures for improved prognostic accuracy in early AD detection.