Purpose: The purpose of this study was to design a fully automatic computer-assisted diagnostic system for early- and late-onset mild Alzheimer's disease (AD).
Methods: Glucose metabolic images were obtained from mild AD patients and normal controls using positron emission tomography (PET) and( 18)F-fluorodeoxyglucose (FDG). Two groups of 20 mild AD patients with different ages of onset were examined. A fully automatic diagnostic system using the statistical brain mapping method was established from the early-onset (EO) and late-onset (LO) groups, with mean ages of 59.1 and 70.9 years and mean MMSE scores of 23.3 and 22.8, respectively. Aged-matched normal subjects were used as controls. We compared the diagnostic performance of visual inspection of conventional axial FDG PET images by experts and beginners with that of our fully automatic diagnostic system in another 15 EO and 15 LO AD patients (mean age 58.4 and 71.7, mean MMSE 23.6 and 23.1, respectively) and 30 age-matched normal controls. A receiver operating characteristic (ROC) analysis was performed to compare data.
Results: The diagnostic performance of the automatic diagnostic system was comparable with that of visual inspection by experts. The area under the ROC curve for the automatic diagnostic system was 0.967 for EO AD patients and 0.878 for LO AD patients. The mean area under the ROC curve for visual inspection by experts was 0.863 and 0.881 for the EO and LO AD patients, respectively. The mean area under the ROC curve for visual inspection by beginners was 0.828 and 0.717, respectively.
Conclusion: The fully automatic diagnostic system for EO and LO AD was able to perform at a similar diagnostic level to visual inspection of conventional axial images by experts.