A study investigated the efficacy of focusing on localized PET-active areas for classification purposes. In this work, they propose a pipeline using learned features from semantically labeled PET images to perform group classification. Early identification of dementia in the early or late stages of mild cognitive impairment (MCI) is crucial for a timely diagnosis and slowing down the progression of Alzheimer's disease (AD).

Positron emission tomography (PET) is considered a highly powerful diagnostic biomarker, but few approaches investigated the efficacy of focusing on localized PET-active areas for classification purposes. In this work, we propose a pipeline using learned features from semantically labeled PET images to perform group classification.

There is no definitive cure for the AD, whereas active research areas seek treatments which are more effective for early MCI to slow down the progression of the disease. This implies a great urgency to develop sensitive biomarkers to detect and monitor early brain changes. The ability to diagnose and classify AD and MCI at an early stage allows clinicians to make more informed decisions at later stages for clinical intervention and treatment planning, thus having a great impact on reducing the cost of longtime care.

From a neuroimaging perspective, positron emission tomography (PET) of fluorodeoxyglucose (FDG) for cerebral glucose metabolism and β amyloid can provide complementary information for the diagnosis of the AD.

PET and MRI Datasets

FDG-PET is a metabolic neuroimaging modality frequently used in the AD, producing a distribution map of glucose uptake. The reduction of glucose uptake in the temporal and parietal areas is often seen as an indication of the AD. Foster et al. demonstrated that FDG-PET improves the accuracy of differentiating AD from frontotemporal dementia, especially when the symptoms and clinical tests are equivocal. However other types of tracers exist such as florbetapir which helps to image amyloid accumulation in the brain.

To the best of our knowledge, this is one of the few studies to focus on PET classification for cognitive stage identification in the AD, comparing learned features from both AV45-PET and FDG-PET images, while using a multiregional approach based on segmented cortical and subcortical areas.

In this work, they compared both whole-brain and multicortical region approaches to identify cognitive stages of the AD, comparing both FDG-PET and AV45-PET in classification accuracy. They observed that the classification accuracy of AD versus NC was improved using longitudinal images, as well as for other pairs of cognitive classes.

The objective was to assess how a feature-learning approach focused on predefined anatomical regions with a known decline in uptake for AD patients can help achieve better accuracy and minimize the errors of automated classification of different stages of Alzheimer's, especially in the early stages of the disease.

Either FDG-PET or AV45-PET enabled discriminating early and late MCI from the AD, as well as NC, with a slight improvement using FDG-PET. The methodologies used in this work can contribute to improving the classification accuracy between different stages of the AD by using the combination of two-time points. Our results confirm that we can rely on PET images as a single biomarker, although the inclusion of additional biomarkers can also improve the accuracy of classification.