Skip to main content
Fig. 6 | Acta Neuropathologica Communications

Fig. 6

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

Fig. 6

Deep histopathology features are partially associated with several known clinicopathologic features and partially independent. a Correlation analysis of deep histopathology results and clinicopathologic features: age, Braak score, evidence of cerebrovascular pathology (coded as 0 = not present and 1 = present), ARTAG positivity in the hippocampus (coded as 0 = not present and 1 = present), cognitive label (coded as 0 = not cognitively impaired and 1 = cognitively impaired), probability of cognitive impairment as predicted by the top-performing model trained on the hippocampal data, and median LFB staining intensity in the top attention tiles in the hippocampus data set. Upper right: rank correlation values and associated p-values (*p < 0.05, **p < 0.01, ***p < 0.001). Diagonal: histograms of variables. Lower left: scatterplots with linear model trend lines for the variable pairs (red lines) and 95% confidence intervals (blue envelopes). This plot was made using the R package GGally (v. 2.1.2). b, c Scatter plots for probability of cognitive impairment estimated in the frontal cortex and hippocampus with Braak stage (b) and AT8 staining positive pixel counts in the medial temporal lobe (MTL) (c). Trend lines show predictions via a linear model and grey envelopes show associated 95% confidence intervals. CI = Cognitive impairment; ARTAG = Aging-related tau astrogliopathy; LFB = Luxol Fast Blue

Back to article page