Biomarkers.

Area of Science:

• Neuroimaging and Artificial Intelligence
• Neurology and Geriatrics

Background:

• Mild cognitive impairment (MCI) affects millions of elderly individuals, with a significant percentage progressing to Alzheimer's disease (AD) annually.
• Accurate prediction of MCI to AD progression is critical for timely interventions, clinical trial design, and understanding disease mechanisms.

Purpose of the Study:

• To develop and validate a 3D convolutional neural network (CNN) for predicting cognitive decline in individuals with MCI.
• To forecast changes in the Clinical Dementia Rating scale sum of boxes (sobCDR) score over a two-year period.

Main Methods:

• Analysis of three independent cohorts (ADNI, OASIS-3, NACC) comprising participants with 3D T1-weighted brain MRI and longitudinal sobCDR scores.
• A 3D DenseNet-121 CNN was employed to predict future sobCDR scores using MRI data.
• The CNN-derived prediction was integrated into linear mixed-effects models to assess its added value in forecasting cognitive decline.

Main Results:

• The predictive models, incorporating image-derived predictions, accurately forecasted sobCDR scores at two years with a mean absolute error of approximately 1 point.
• The inclusion of neuroimaging data significantly improved the accuracy of cognitive decline prediction.

Conclusions:

• Deep learning approaches, particularly CNNs utilizing neuroimaging, demonstrate substantial potential for enhancing prognostic models in neurodegenerative diseases.
• Future research should explore advanced deep learning techniques, multimodal neuroimaging data (e.g., PET, diffusion MRI), and novel data fusion strategies.