Skip to main content

Table 2 Confusion matrices for the three proposed cut-offs for the model predictions regarding the presence of definite radiographic sacroiliitis on the validation dataset

From: Deep learning for detection of radiographic sacroiliitis: achieving expert-level performance

  nr-axSpA r-axSpA  
Cut-off 1, favouring sensitivity over specificity
 Model predicts nr-axSpA 15 1 16
 Model predicts r-axSpA 64 149 227
  79 150 229
 Cohen’s kappa 0.22 (95% CI 0.11–0.33) Accuracy: n = 164/229 (71.6%)
Cut-off 2, favouring specificity over sensitivity
 Model predicts nr-axSpA 78 38 116
 Model predicts r-axSpA 1 112 113
  79 150 229
 Cohen’s kappa 0.66 (95% CI 0.57–0.76) Accuracy: n = 190/229 (83.0%)
Cut-off 3, optimal relationship between sensitivity and specificity
 Model predicts nr-axSpA 75 19 94
 Model predicts r-axSpA 4 131 135
  79 150 229
 Cohen’s kappa 0.79 (95% CI 0.7–0.87) Accuracy: n = 206/229 (90.0%)