Fig. 20
From: Trustworthy deep learning for malaria diagnosis using explainable artificial intelligence
Misdiagnosed samples explained using SHAP (GradientExplainer). Each example shows an uninfected cell (true class 0) misclassified as parasitized (pred class 1). The SHAP overlays highlight per-pixel contributions that misled the classifier, primarily faint staining and boundary irregularities rather than parasite structures. These fine-grained attributions reveal how minor cytoplasmic color variations can induce false positives, underscoring SHAP’s value in identifying subtle but influential artifacts in microscopy-based diagnostics.
