|
AI model
|
Aβ-model
|
Iron model
|
Fibrin model
|
GFAP model
|
CD68 model
|
Calcium model
|
|---|
|
Stain
|
IHC for Aβ
|
Perls’ Prussian blue
|
IHC for Fibrin(ogen)
|
IHC for GFAP
|
IHC for CD68
|
von Kossa
|
|
No. of training sections/total no. of sections in dataset
|
13/146
|
16/144
|
13/142
|
11/144
|
8/39*
|
9/44*
|
|
Layer 1
|
Area
|
Area
|
Area
|
Area
|
Area
|
Area
|
| |
Leptomeningeal vessels
|
Tissue
|
Non-vascular tissue
|
Tissue
|
Tissue
|
Tissue
|
| |
Tissue
| |
Vascular fibrin positive tissue
| | | |
|
Layer 2
|
Area
|
Object
|
Object
|
Object
|
Object
|
Area (Calcium positive)
|
| |
CAA
|
Iron positive cells
|
Fibrin positive cells
|
Reactive astrocytes
|
CD68 positive cells
|
Neuronal
|
| |
Amyloid-β plaques
| | | | |
Vascular
|
| | | | | | |
Extracellular
|
|
Layer 3
| | | | | |
Object
|
| | | | | | |
Calcium positive cells
|
|
No. of sections excluded after QC
|
4
|
5
|
3
|
0
|
0
|
0
|
|
No. of sections for validation
|
13
|
14
|
14
|
14
|
4
|
4
|
- The table summarizes the six models described in this study, listing the kind of staining for each set of sections, the number of sections on which the model was trained, and the layers by which the model was built. For each layer it is reported whether the convolutional neural network was trained to recognize an area or an object. Finally, the total number of sections on which the model was applied, those excluded after QC, and those on which the model was validated is reported. Key: IHC: immunohistochemistry; AD: Alzheimer’s disease; CAA: cerebral amyloid angiopathy; QC: quality control
- *Note that the CD68 and calcium models were trained on and applied to sections derived only from cohorts 2 and 3 (AD cases), whereas the other four models used data from all 3 cohorts (AD and CAA cases)
