Table 2 Stratified performance metrics of the ROS1 fusion prediction model on the test and holdout sets

From: Predicting ROS1 and ALK fusions in NSCLC from H&E slides with a two-step vision transformer approach

 

Test

Holdout

 

AUC

Acc

PPA

PPV

F1

NPA

AUC

Acc

PPA

PPV

F1

NPA

ALL

0.85

0.92

0.50

0.06

0.10

0.92

0.85

0.92

0.48

0.06

0.10

0.93

Bx

0.86

0.92

0.45

0.05

0.09

0.92

0.86

0.92

0.46

0.05

0.09

0.93

Rx

0.85

0.91

0.58

0.07

0.12

0.92

0.84

0.92

0.50

0.08

0.13

0.92

  1. Results are reported for all cases (ALL), biopsy samples (Bx), and resection samples (Rx). Metrics include area under the ROC curve (AUC), accuracy (Acc), positive percent agreement (PPA), positive predictive value (PPV), F1-score (F1), and negative percent agreement (NPA). Values represent the average performance across five cross-validation folds. A positive class weight factor of 5 was used in the loss function.
  2. Bx biopsy, Rx resection.
Back to article page