Neuro Notes
This is the #1 risk factor in developing dementia.
Aging
This is the radioactive tracer used in the PET images analyzed in this study
Fludeoxyglucose-18 (FDG)
This is the full name for the type of deep learning model used in this paper.
Convolutional Neural Network
This metric evaluates True Positives vs False Negatives
Sensitivity
This is the name of the dataset that the authors used to train their model.
ADNI
Abbreviated as MCI, this syndrome can progress to Alzheimer’s disease but doesn’t always.
Mild Cognitive Impairment
Abbreviated as CSF, this is the substance used to test for biomarkers associated with dementia
Cerebrospinal fluid
This is the purpose of the convolution layers of a CNN.
The convolution layers use filters to extract features from input data (e.g., edges, shapes, textures)
This is the AUC value predicted by a completely random model
0.5
Abbreviated as PET, this technique quantifies brain metabolism
Positron Emission Tomography
These are the three most common causes of dementia.
AD, Dementia with Lewy Bodies, Frontotemporal Degeneration
Other than detecting neurodegeneration, this is one other use of FDG PET.
Diagnosing cancer, monitoring heart disease
These are the two CNN layers set up in repeating blocks to give the learning model more depth.
The convolution and pooling layers
This was the visualization method determined most reliable by the authors.
SmoothGRAD
This is the first author of the study.
Prats-Climent
These are two neuron-level problems that occur in dementia.
Synaptic dysfunction and neuronal cell death
This kind of event causes a positron-electron pair to rapidly vanish, with their energy being released as gamma-rays
Annihilation event
Participants needed to meet these two requirments in order to be part of the CNN testing group.
Had to meet the definition of a MCI, and have had an FDG PET scan
This was the overall accuracy determined by the CNN in either of the two data sets.
79% or 80%
On the CDR rating scale (which varies from 0-3), this is the score that constitutes a MCI diagnosis.
0.5
These are the two aggregates associated with AD.
Ab42 plaques and hyperphosphorylated Tau
These are four of the six categories on the CDR rating scale, which quantifies severity of dementia.
Memory, orientation, judgment/problem solving, community affairs, home & hobby, personal care
This is the type of deep learning model the authors began with for the basis of their CNN, before tweaking the parameters to their needs.
Visual Geometry Group Network (VGG)
These were the two areas highlighted by the two most reliable visualization maps.
Posterior cingulate and superior parietal areas
This was the importance of the time-of-flight technology used by the Philips Gemini PET/CT scanner.
TOF technology gives better spatial resolution in FDG PET images, because it accounts for the difference in time for the arrival of the gamma-rays at the PET scanner detectors