Introduction
Material and methods
Overview of the feature selection approach
Neuropathology features in the CFAS cohort
No. | Feature | Feature description | Type | Control | ||
---|---|---|---|---|---|---|
Dementia (n=107) | No dementia (n=70) | Missing (n=9) | ||||
1 | Braak NFT stage | Nominal | 107 | 70 | 0 | |
2 | Thal phase | Nominal | 107 | 70 | 0 | |
3 | Aβ stage typical | Aβ stage typical indicates the Aβ stage typical and atypical [18] | Nominal | 107 | 70 | 0 |
4 | PART-definite | PART relates to the new primary age-related-tauopathy concept. PARTdefinite as cases having no Aβ pathology (Thal 0) and with Braak NFT stages I–IV [23] | Nominal | 50 | 47 | 80 (45.2%) |
5 | PART-all | Those cases with mild Aβ pathology (Thal I–II) and with Braak NFT stages I–IV [23] | Nominal | 71 | 63 | 43 (24.3%) |
6 | CAA areas | The number of brain areas examined that have CAA (number of areas out of 9 maximum) [24] | Numeric | 107 | 70 | 0 |
7 | CAA type | Nominal | 107 | 70 | 0 | |
8 | CAA parenchymal | Nominal | 107 | 70 | 0 | |
9 | CAA meningeal | CAA severity meningeal has the same scoring system as CAA parenchymal with the score ranging from 0 to 12 [18] | Nominal | 107 | 70 | 0 |
10 | CAA total severity | The scores for parenchymal and leptomeningeal amyloid were summed in four areas, and scores range from 0 (minimum) to 24 (maximum) for severity in cortical areas [24] | Numeric | 107 | 70 | 0 |
11 | CAA frontal | CAA in the frontal cortex (present or absent) [26] | Nominal | 107 | 70 | 0 |
12 | CAA temporal | CAA in the temporal cortex (present or absent) [26] | Nominal | 107 | 70 | 0 |
13 | CAA parietal | CAA in the parietal cortex (present or absent) [26] | Nominal | 107 | 70 | 0 |
14 | CAA occipital | CAA in the occipital cortex (present or absent) [26] | Nominal | 107 | 70 | 0 |
15 | CAA hippocampus | CAA in the hippocampus and occipitotemporal gyrus (present or absent) [26] | Nominal | 107 | 70 | 0 |
16 | CAA cerebellum | CAA in the cerebellum (present or absent) [26] | Nominal | 106 | 69 | 2 (1.13%) |
17 | BrainNet tau stage | BrainNet tau stage refers to BrainNet Europe protocol for tau pathology, a six-stage scheme that uses neuropil threads and is proposed by the BrainNet Europe Consortium [22] | Nominal | 107 | 69 | 1 (0.6%) |
18 | Hippocampal tau NFT stage | Hippocampal tau neurofibrillary tangles (NFT) stage [26] | Nominal | 56 | 35 | 86 (48%) |
19 | Subpial TSA in the expanded cortex | The subpial thorn-shaped astrocytes (TSA) in the expanded cortex | Nominal | 107 | 69 | 1 (0.6%) |
20 | Subpial TSA in the mesial temporal lobe | The subpial thorn-shaped astrocytes (TSA) in the mesial temporal lobe | Nominal | 107 | 69 | 1 (0.6%) |
21 | Subpial TSA in the brainstem | The subpial thorn-shaped astrocytes (TSA) in the brainstem | Nominal | 107 | 67 | 3 (1.7%) |
22 | TSA-any | Thorn-shaped astrocytes (TSA) in any brain area (present or absent). | Nominal | 107 | 69 | 1 (0.6%) |
23 | TSA-total | Numeric | 107 | 69 | 1 (0.6%) | |
24 | Tufted astrocytes | The tufted parenchymal astrocytes in any brain area | Nominal | 107 | 69 | 1 (0.6%) |
25 | Subpial mesial temporal | The subpial tau neurites in the mesial temporal lobe | Nominal | 107 | 69 | 1 (0.6%) |
26 | Subpial brainstem | The subpial tau neurites in the brainstem/subcortical region | Nominal | 107 | 67 | 3 (1.7%) |
27 | Argyrophilic grains | The argyrophilic grains disease | Nominal | 107 | 69 | 1 (0.6%) |
28 | Cortical stage | The cortical microinfarcts stage which distinguishes the number of cortical areas that have microinfarcts | Numeric | 106 | 70 | 1 (0.6%) |
29 | Subcortical stage | Subcortical lacune stage which distinguishes the number of subcortical areas that have microinfarcts | Numeric | 106 | 70 | 1 (0.6%) |
30 | Microinfarct stage | The total microinfarct stage which differentiates the number of total areas that have microinfarcts | Numeric | 106 | 70 | 1 (0.6%) |
31 | Frontal microinfarct | Frontal microinfarct [31] | Nominal | 106 | 70 | 1 (0.6%) |
32 | Temporal microinfarct | Temporal microinfarct [31] | Nominal | 106 | 70 | 1 (0.6%) |
33 | Parietal microinfarct | Parietal microinfarct [31] | Nominal | 106 | 70 | 1 (0.6%) |
34 | Occipital microinfarct | Occipital microinfarct [31] | Nominal | 106 | 70 | 1 (0.6%) |
35 | Age | Patient’s age at death | Numeric | 107 | 70 | 0 |
36 | Brain weight | Patient’s brain weight | Numeric | 91 | 59 | 27 (15%) |
37 | Gender | Sex | Nominal | 107 | 70 | 0 |
38 | Virchow-Robin space expansion | Virchow-Robin spaces (VRS) are cavities filled with cerebrospinal fluid surrounding small penetrating cerebral arterioles with extensions of the subarachnoid space | Nominal | 106 | 70 | 1 (0.6%) |
39 | Lewy bodies in substantia nigra | The Lewy body is a distinguishing neuronal inclusion. This is always found in the substantia nigra and brain regions in Parkinson’s disease, which occurs wherever there is excessive loss of neurons | Nominal | 105 | 68 | 4 (2.3%) |
40 | Neuronal loss in the hippocampus | Neuronal loss in the hippocampus | Nominal | 106 | 70 | 1 (0.6%) |
41 | Neuronal loss in substantia nigra | Neuronal loss in substantia nigra | Nominal | 105 | 68 | 4 (2.3%) |
42 | Tangles in the temporal lobe | Tangles in the temporal lobe | Nominal | 106 | 70 | 1 (0.6%) |
43 | Parenchymal CAA in the frontal lobe | Parenchymal CAA in the frontal lobe | Nominal | 106 | 70 | 1 (0.6%) |
44 | Gliosis in the hippocampus | Gliosis in the hippocampus | Nominal | 106 | 70 | 1 (0.6%) |
45 | Dementia status | Class label (dementia or no dementia) status of a patient | Binary | 107 | 70 | 0 |
Dementia status
Ranking neuropathology features
Dementia classification
Classification with multiple feature sets
Evaluation of classification performance
-
True positives (TP): number of dementia cases that were correctly classified.
-
False positives (FP): number of healthy subjects incorrectly classified as dementia cases.
-
True negatives (TN): number of healthy subjects correctly classified.
-
False negatives (FN): number of dementia cases incorrectly classified as healthy subjects.
-
Accuracy (%): the proportion of correct classifications among total classifications:
-
Sensitivity (%): The proportion of correctly classified dementia cases.
-
Specificity (%): The proportion of correctly classified healthy subjects.
-
Precision: The proportion of subjects classified as dementia cases who have dementia.
-
F1-score (F-measure) (%): Harmonic mean of precision and sensitivity.