Summary of classification scores of Persistence Images (PIs) after cross-validation on the discovery and validation datasets.

We constructed classifiers using logistic Regression (LR) and Support Vector Machines (SVM) on the discovery and validation datasets, using PIs obtained from PH analysis of biomarkers CD10, CD20, CD38, and CD45. The accuracy of the SVM method depends on the dimension of the barcode, the spread of the Gaussian and grid size used to generate the PIs, as well as the parameters C and γ for the SVM method. The accuracy (Acc.) of the model predictions after 6-fold cross-validation is greatest for Dimension 0 and 1. Additional summaries are provided in Tables C-E in S1 Text. An expanded analysis, using fixed hyperparameters and other classification scores (AUC, mean accuracy, standard deviation of the accuracy, recall, precision, F1-score and confusion matrices) can be found in Tables F and G in S1 Text.