Fig. 3: Performance of radiomic prediction of malaria using selected features.

a Feature importance of the top ten significant radiomic features, excluding shape-related features due to the lack of potential diagnostic information. The bar graph displays the mean decrease in impurity for each feature. b Pairwise correlation coefficients among the selected radiomic features, illustrating complex interactions. The absence of unity among the pairwise correlation coefficients supports the uniqueness of each selected feature. c Receiver operating characteristic (ROC) curves of the malaria prediction model plotted. The 95% confidence intervals for the ROC curves were calculated using the Wald method. d Precision-recall curve reflecting the relationship between positive predictive values and sensitivity. e Box plot of Pearson residuals comparing the two smartphone models. f Box plot of Pearson residuals between the left and right inner eyelids. g Box plot of Pearson residuals by sex. The box plots indicate the minimum, maximum, median, and the first and third quartiles of the data.