The aim of this study was to evaluate imaging-based predictive validity to Alzheimer's disease in amnestic Mild Cognitive Impairment (MCI) patients using 18F-fluorodeoxyglucose Positron Emission Tomography (FDG PET) by an automatic computer-assisted system and visual evaluation rating. Baseline FDG PET images were compared between MCI converter to Alzheimer's disease group (MCI-C) and MCI non-converter group (MCI-NC) to get predictive validity by automatic computer-assisted system and two blinded human nuclear medicine professionals. To obtain the optimal cut-off point to discriminate MCI-C from MCI-NC, Minimal distance method from Receiver Operating Characteristic (ROC) curve was used. Fifty-nine patients with MCI received baseline FDG PET examinations when enrolled and were followed up for two years. After two years, 22 MCI patients (37.29%) were converted to Alzheimer's disease. When using automatic computer-assisted system, distance from ROC is the shortest in parietal lobe and at this point, sensitivity is 0.63 and specificity is 0.66. When using visual evaluation rating, distance from ROC is the shortest in parietal lobe and at this point, sensitivity is 0.64 and specificity is 0.70. Kappa value in inter-rater reliability is high (0.84). In the predictive validity from MCI to Alzheimer's disease, computer-assisted system showed similar accuracy to visual imaging rating and cannot replace visual imaging evaluation.