Predicting conversion from mild cognitive impairment to Alzheimer's disease: A systematic review of deep learning models for early-stage disease classification.
Mozhgan Ghasabi, Hadi Aghdasi, Mohammad-Reza Feizi-Derakhshi
✦ AI-curated · Sources linked
In 30 seconds
This PRISMA-guided systematic review assessed deep learning approaches for classifying Alzheimer’s disease stages and predicting conversion from mild cognitive impairment (MCI) to Alzheimer’s disease. It included 60 studies published between January 1, 2019, and February 20, 2026. Cross-sectional designs predominated, and 42 studies relied on the ADNI dataset. For the challenging binary classification of progressive versus stable MCI, reported accuracy ranged from 71.71-96.3%; results varied across studies and are not directly comparable because of methodological and dataset heterogeneity.
Key findings
- Of 60 included studies, 47 used cross-sectional designs and 13 used longitudinal designs.
- The ADNI dataset was used in 42 studies, raising concerns about generalizability.
- For progressive versus stable MCI classification, reported accuracy ranged from 71.71-96.3%.
- Performance declined as the number of disease-stage classes increased; multimodal models often reported strong results, particularly for MCI-related tasks.
Why it matters
The review highlights both the promise and the current limits of deep learning for anticipating MCI-to-Alzheimer’s conversion. Dataset concentration, inconsistent validation and prediction tasks, and limited interpretability make reported performance difficult to translate into dependable clinical tools. Longitudinal, multimodal, and explainable approaches may help address these gaps, but the evidence remains heterogeneous.
What to keep in mind
The review reports substantial heterogeneity across datasets, prediction tasks, validation strategies, and methodological quality. Heavy reliance on ADNI and recurring concerns about overfitting and poor interpretability limit confidence in generalizability.