Skip to main content
Fig. 4 | Acta Neuropathologica Communications

Fig. 4

From: Interpretable deep learning of myelin histopathology in age-related cognitive impairment

Fig. 4

Interpretation of tissue-level attention maps and tile-level staining intensity in the hippocampus suggests myelin loss. a, b Representative WSIs labeled and predicted to be in the non-cognitive impaired (upper) or cognitively impaired (lower) groups (a) and corresponding representative attention heatmaps (b). In these heatmaps, dark red indicates relatively high attention values, while dark blue indicates relatively low attention values. c Top 5 highest attention tiles (upper) and blue hue range positive pixel annotations (lower) from the matching WSIs as shown in sub-figures A/B. Scale bar = 20 μm. d Median z-transformed attention score values in the grey matter and white matter. Each data point is a median attention score from the white matter or the grey matter from one WSI. e, f Median dark blue range pixel counts as a measure of LFB staining intensity (e) and ratio of the dark blue to light blue pixel counts in the top attention tiles of WSIs predicted and labeled to have cognitive impairment or not (f). g Scatter plot and contour lines showing the relationship between dark blue range pixel counts and the ratio of the dark blue to light blue pixel counts in the top attention tiles of WSIs. Orange dots indicate that the WSI was predicted to come from a CI donor, while blue dots indicate NCI. *p < 0.05, ***p < 0.001. GM = Grey Matter; WM = White Matter; CI = Cognitively Impaired; NCI = Not Cognitively Impaired

Back to article page