Fig 1.
Patient inclusion and exclusion schema for this study.
Fig 2.
Receiver operator characteristic (ROC) curves for diagnostic performance of lesion-level treatment response prediction in lymphoma using transfer learning (on test data sets for 5 input scenarios and 3 image modalities using 40 epochs and batch size 5).
TP = true positive fraction, FP = false positive fraction, AUC = area under the curve, dCT = diagnostic computed tomography, lCT = low-dose computed tomography, PET = positron emission tomography.
Table 1.
Diagnostic performance of lesion-level treatment response prediction in lymphoma using transfer learning (for 5 input scenarios and 3 image modalities using 40 epochs and batch size 5).
Mean and standard deviation values are displayed. VOI = volume of interest, dCT = diagnostic computed tomography, lCT = low-dose computed tomography, PET = positron emission tomography, Acc = accuracy, Sens = sensitivity, Spec = specificity, AUC = area under the curve.
Table 2.
Diagnostic performance of patient-level treatment response prediction in lymphoma using rule-based reasoning approach (from lesion-level response predictions using 1 whole-slice input scenario, 3 image modalities, and transfer learning) compared to International Prognostic Index risk factors for diffuse large B-cell lymphoma (DLBCL) patients.
Note that results are shown for entire subject cohort (All) and for DLBCL subject cohort. dCT = diagnostic computed tomography, lCT = low-dose computed tomography, PET = positron emission tomography, IPI = International Prognostic Index, Acc = accuracy, Sens = sensitivity, Spec = specificity.