DTI-based AI for Alzheimer's: lessons from 98 studies on algorithm choice and clinical generalizability

Review of 98 DTI-based AI studies for Alzheimer's (2010–2026): algorithm choice tied to sample size (SVM for small, DL for large), external validation degraded performance, and certain tracts (hippocampal, callosal, fornix, cingulum) were robust.

• Modality selection (single-modality DTI vs. multimodal fusion) should be driven by the specific diagnostic task.
• Conventional machine learning (especially SVM) dominated small-sample studies, while deep learning (CNNs, graph convolutional networks, Transformers) was used with larger datasets.
• Data heterogeneity (source, acquisition, processing) and lack of external validation remain major barriers to clinical translation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.