Skip to main content
Fig. 4 | Brain Informatics

Fig. 4

From: Explainability of random survival forests in predicting conversion risk from mild cognitive impairment to Alzheimer’s disease

Fig. 4

Local explanations of Random Survival Forests (RSF). A. On the left: histograms of sMCI and pMCI patients’ risk distribution predicted by RSF. pMCI subjects were stratified by risk grade: low (in green, between 1 and 3.5), medium (in orange, between 3.5 and 5), high (in red, between 5 and 7.1). On the right: RSF survival functions of pMCI patients per risk score: pMCI#1 high risk (score 7.088, converted to AD after 12 months), pMCI#2 medium risk (score 3.876, converted to AD after 24 months), pMCI#3 low risk (score 2.12, converted to AD after 36 months). SHAP waterfall plot (top) and force plot (bottom) of B. patient pMCI#1, C. patient pMCI#2, D. patient pMCI#3, E. stable MCI patient who does not convert to AD within 48 months (sMCI#1, risk score 0.233). Blue and red arrows represent those features that, respectively, decrease and increase the conversion-to-AD risk within 48 months. Average predicted risk E[f(x)] = 2.968. Actual value of feature in gray

Back to article page