Structural MRI Morphometry Across the Alzheimer’s Disease Spectrum: Classification and Generalization

Structural magnetic resonance imaging (MRI) provides quantitative measures of brain changes associated with Alzheimer's disease (AD), but reliable classification also depends on the evaluation protocol and performance across cohorts. This study assessed a morphometry-based approach for distinguishing cognitively normal (CN) participants, those with mild cognitive impairment (MCI), and those with AD. The primary cohort included 815 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI), with one screening MRI per participant and predefined training, validation, and test sets. MRI records were linked to FreeSurfer measurements using exact participant and image identifiers. A multinomial L2-regularized logistic regression classifier used 42 predefined structural features and five-fold cross-validation grouped by acquisition site. In the held-out three-class test (n = 121), balanced accuracy was 0.6119 (95% confidence interval [CI]: 0.5291-0.6873) and the macro-averaged area under the receiver operating characteristic curve (ROC-AUC) was 0.7525. Recall was 0.8529 for CN, 0.2931 for MCI, and 0.6897 for AD. Additional development analyses examined demographic and clinical variables separately from the primary MRI-only result. For the binary transfer experiment, a separate ADNI-trained CN-versus-AD model using 32 shared morphometric features achieved balanced accuracy of 0.8458 and ROC-AUC of 0.9016 in site-grouped out-of-fold evaluation. The fixed model was then tested without target-cohort tuning on 181 older participants from the Open Access Series of Imaging Studies (OASIS-1), grouped by Clinical Dementia Rating (CDR = 0 versus CDR > 0). External balanced accuracy was 0.6492 (95% CI: 0.5799-0.7158) and ROC-AUC was 0.7502. Because ADNI diagnoses and OASIS-1 CDR groups are not identical clinical endpoints, the external experiment was interpreted separately from the primary three-class analysis. The results demonstrate the value of reporting class-specific performance and independent-cohort evaluation when assessing structural MRI morphometry.

Authors

Institutions

Publication Details

Journal
Biomedical Physics & Engineering Express
Published
2026-11-17
DOI
https://doi.org/10.1088/2057-1976/aec2f4
Primary Topic
Dementia and Cognitive Impairment Research
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Structural MRI Morphometry Across the Alzheimer’s Disease Spectrum: Classification and Generalization

Ahmed Benyahia, Imad Eddine Araar, Abdessalem Benammar, Ahmed bahi azzououm
1 links
Biomedical Physics & Engineering Express
Dementia and Cognitive Impairment Research
article

Structural MRI Morphometry Across the Alzheimer’s Disease Spectrum: Classification and Generalization

Ahmed Benyahia, Imad Eddine Araar, Abdessalem Benammar, Ahmed bahi azzououm
article en

Abstract

Structural magnetic resonance imaging (MRI) provides quantitative measures of brain changes associated with Alzheimer's disease (AD), but reliable classification also depends on the evaluation protocol and performance across cohorts. This study assessed a morphometry-based approach for distinguishing cognitively normal (CN) participants, those with mild cognitive impairment (MCI), and those with AD. The primary cohort included 815 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI), with one screening MRI per participant and predefined training, validation, and test sets. MRI records were linked to FreeSurfer measurements using exact participant and image identifiers. A multinomial L2-regularized logistic regression classifier used 42 predefined structural features and five-fold cross-validation grouped by acquisition site. In the held-out three-class test (n = 121), balanced accuracy was 0.6119 (95% confidence interval [CI]: 0.5291-0.6873) and the macro-averaged area under the receiver operating characteristic curve (ROC-AUC) was 0.7525. Recall was 0.8529 for CN, 0.2931 for MCI, and 0.6897 for AD. Additional development analyses examined demographic and clinical variables separately from the primary MRI-only result. For the binary transfer experiment, a separate ADNI-trained CN-versus-AD model using 32 shared morphometric features achieved balanced accuracy of 0.8458 and ROC-AUC of 0.9016 in site-grouped out-of-fold evaluation. The fixed model was then tested without target-cohort tuning on 181 older participants from the Open Access Series of Imaging Studies (OASIS-1), grouped by Clinical Dementia Rating (CDR = 0 versus CDR > 0). External balanced accuracy was 0.6492 (95% CI: 0.5799-0.7158) and ROC-AUC was 0.7502. Because ADNI diagnoses and OASIS-1 CDR groups are not identical clinical endpoints, the external experiment was interpreted separately from the primary three-class analysis. The results demonstrate the value of reporting class-specific performance and independent-cohort evaluation when assessing structural MRI morphometry.

Biomedical Physics & Engineering Express
Research Center in Industrial Technologies (DZ), Hassiba Benbouali University of Chlef (DZ)
Foundation for the National Institutes of Health, Alzheimer's Association, Biogen, Alzheimer's Disease Neuroimaging Initiative, BioClinica, Northern California Institute for Research and Education, National Institutes of Health, Canadian Institutes of Health Research
Good health and well-being
Dementia and Cognitive Impairment Research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon