FDG-PET Image Classification in Alzheimer's Disease: from Traditional Visual Analysis to Advanced Transfer Learning.

Abstract

Purpose: Alzheimer's disease (AD) often coexists with other brain pathologies, and we aimed to classify people with AD using 18 F- Flouro-Deoxy-Glucose-Positron Emission Tomography (FDG-PET).

Method: Baseline FDG-PET data were collected as part of two large scale Phase III clinical trials of a novel tau aggregation inhibitor drug, Leuco-Methylthioninium (LMTX®). A total of 794, well-characterised probable AD subjects were included in the study and the images were classified into "typical AD"(temporoparietal hypometabolism) and "mixed" (patchy hypo-metabolism in other vascular territories of brain such as frontal and cerebellar regions along with temporo-parietal hypo-metabolism) patterns based on visual interpretation. The differences in the two groups were further assessed with region-of-interest based analysis of Standardized Uptake Value Ratio (SUVR) and automated classification using transfer learning with visual classification as the gold standard.

Results: Of the total of 794 (438 female) participants, 533 (284 female) were classified as typical AD and 261 (154 female) participants classified as mixed. A subset of 50 images each from typical and mixed subtypes were used for transfer learning and sensitivity, specificity and accuracy for one of the cross-validation loops was 94.73%, 95.23% and 95% respectively. The average accuracy to distinguish the two subtypes after 5-fold cross validation was found to be 97.5%.

Conclusions: This study is first of its kind to distinguish two subtypes of AD through visual interpretation of FDG-PET images and exploring the findings with a semi-quantitative method followed by transfer learning, which has been used to predict the two subtypes with high accuracy, sensitivity and specificity.