Premium
P3‐428: A BIOMARKER FOR ARTERIOLAR SCLEROSIS BASED ON MRI‐DERIVED FEATURES
Author(s) -
Makkinejad Nazanin,
Evia Arnold M.,
Tamhane Ashish A.,
Schneider Julie A.,
Bennett David A.,
Arfanakis Konstantinos
Publication year - 2018
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2018.06.1791
Subject(s) - medicine , biomarker , hyperintensity , imaging biomarker , multiple sclerosis , cohort , magnetic resonance imaging , white matter , dementia , neuroimaging , cardiology , pathology , radiology , psychiatry , disease , biochemistry , chemistry
using the AUC. For thickness and volume, the most important features (occurrence in CV-Lasso > 5) were used to select a subset of significant features in combined models. Results: Figure 1 reports the results obtained when each feature group is taken individually. Adding APOE4 status to BF only yields a marginal gain (Fig-1B). Adding either the Ab status, volume or thickness leads to a larger AUC increase. After feature selections, the combined volume and thickness (#features1⁄417) yields the highest AUC1⁄40.8260.13 comparable to the BF+Ab (Fig-1C). Figure 2 reports the results of the BF + features combined. BF+Ab+APOE4 model leads to no increase in the AUC compared to BF+Ab. The highest accuracy is obtained when combining MRI volume, thickness and Ab status (AUC1⁄40.8660.11). Conclusions: Combining regional brain volumes, cortical thickness and basic baseline characteristics appears to be a powerful tool for predicting MCI conversion to AD. Determination of thresholds based on these baseline values may help identify rapid decliners, who would be the ideal population for a clinical trial aiming at delaying disease onset, as it would help decrease sample size and study cost. P3-428 A BIOMARKER FOR ARTERIOLAR