Deep Learning Model Predicts Cognitive Decline in Non-Demented Elderly

Strategic Insights -

A recent study published in Radiology Advances has demonstrated the potential of a deep learning model in predicting cognitive decline in non-demented elderly individuals across the Alzheimer's disease clinical spectrum. The research, conducted by researchers from Yonsei University College of Medicine, utilized an F-fluorodeoxyglucose-positron emission tomography (F-FDG-PET)-based deep learning model to classify patients with mild cognitive impairment (MCI) and normal cognition (NC) into dementia and non-dementia groups.

Key Takeaways:

  • The F-FDG-PET-based deep learning model demonstrated a high degree of accuracy in predicting cognitive decline in non-demented elderly individuals, with an area under the curve (AUC) of 0.85.
  • The model performed better than conventional measures from F-FDG-PET and amyloid PET in predicting cognitive decline.
  • Subgroup analysis in the amyloid-positive ADNI MCI participants showed that the DL output remained independently prognostic among other factors at 4-year follow-up.
  • The model has the potential to improve cognitive decline prediction beyond clinical information and conventional measures.
  • The study was funded by the Samsung Research Funding Center of Samsung Electronics and the National Research Foundation of Korea.
  • The research involved a large dataset of 756 patients from the ADNI and J-ADNI datasets.

Statistics:

  • The F-FDG-PET-based deep learning model demonstrated an AUC of 0.85 in predicting cognitive decline.
  • The model performed better than conventional measures from F-FDG-PET and amyloid PET, with an AUC of 0.75 and 0.65, respectively.
  • The subgroup analysis in the amyloid-positive ADNI MCI participants showed that the DL output remained independently prognostic among other factors at 4-year follow-up (tdAUC = 0.85).

Sources:

  • NewsRx. Researchers from Yonsei University College of Medicine Report on Findings in Alzheimer Disease (18F-FDG-PET-based deep learning for predicting cognitive decline in non-demented elderly across the Alzheimer's disease clinical spectrum). Mental Health Weekly Digest. October 20, 2025; p 1002.
  • Seok Jong Chung, et al. 18F-FDG-PET-based deep learning for predicting cognitive decline in non-demented elderly across the Alzheimer's disease clinical spectrum. Radiology Advances, 2024;1(3).