A real-world decision tree model based on ¹⁸F-FDG PET/CT Z scores and Mini-Mental State Examination for differentiating mild cognitive impairment from clinically diagnosed Alzheimer disease

Abstract

Objective

To evaluate the ability of regional metabolic Z scores derived from ¹⁸F-fluorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG PET/CT), medial temporal atrophy (MTA) scores, and Mini-Mental State Examination (MMSE) results to distinguish mild cognitive impairment (MCI) from clinically diagnosed Alzheimer disease (AD), and to develop an interpretable decision-tree model based on these parameters.

Methods

This retrospective single-center study included 124 patients who underwent ¹⁸F-FDG PET/CT for suspected cognitive impairment between 2023 and 2025. Patients were classified as AD or MCI according to final clinical diagnoses established by experienced neurologists using National Institute on Aging–Alzheimer’s Association criteria. Regional metabolic Z scores were generated with CortexID Suite by comparison with an age-matched healthy normative database. MTA was graded on structural magnetic resonance imaging using the Scheltens scale by two blinded nuclear medicine physicians. MMSE scores were retrieved from clinical records. A random forest algorithm was used to identify discriminative variables, which were then incorporated into a transparent decision-tree classifier. The dataset was divided into a training cohort (n = 100) and a split-sample test cohort (n = 24).

Results

The final decision tree integrated regional metabolic Z scores and MMSE values in a hierarchical classification structure. In the split-sample test cohort, the model achieved an accuracy of 95.8% and a balanced accuracy of 96.4%, with 100% sensitivity and 92.9% specificity for AD classification. The area under the receiver operating characteristic curve was 0.964 (95% confidence interval, 0.894–1.000).

Conclusions

An interpretable decision-tree approach integrating quantitative ¹⁸F-FDG PET/CT metrics with cognitive assessment showed promising performance for distinguishing clinically diagnosed AD from MCI in a real-world clinical cohort.