A statistical method for predicting amyloid-β deposits from severity, extend, and ratio indices of the 99mTc-ECD SPECT.

Abstract

BackgroundAmyloid-β (Aβ) deposit prediction accuracy is necessary for clinicians treating patients desiring Alzheimer's disease (AD) modifying drugs. Aβ-PET imaging is useful for diagnosis, but high in cost compared to brain perfusion SPECT. However, SPECT displays regional cerebral blood flow (rCBF) and does not detect Aβ deposits, therefore requiring additional clinical information.ObjectiveThis article describes a novel statistical method to predict amyloid deposits via PET images (Aβ+ or Aβ-) using the three indices of the ^99mTc-ECD SPECT - severity, extend, and ratio - for the three ROIs.MethodsCandidate patients (N = 114 patients [55% male], 81 Aβ+ 33 Aβ-, mean age 74.2 ± 6.6 years, mean MMSE score 23.7 ± 2.8) underwent MRI and ^99mTc-ECD SPECT scanning. After examining SPECT index, demographic, and age data, age and sex were treated as confounders in one, two, and three-index logistic additive models with severity, extend, and ratio as explanatory variables. Area under curve (AUC), sensitivity and specificity were used as statistical indices for model fitness and accuracies. Three-hold cross validation analyses were conducted to evaluate error rates.ResultsAccording to ROC analysis, best scores for fitness and accuracy were obtained from the three-index model with patients' age and sex for the configured ROIs including precuneus, posterior cingulate and temporal-parietal region of SPECT (AUC: 0.818, Sensitivity: 0.803, Specificity: 0.727).ConclusionsThis technique using ^99mTc-ECD SPECT data can predict amyloid deposits with acceptable accuracy. To confirm the reliability and validity, a multicenter SPECT study is needed.