Artificial intelligence-driven three-dimensional modeling of biliary strictures using magnetic resonance cholangiopancreatography for quantitative assessment and stent planning

Table 3 Projection-based classification performance for stricture presence and severity

Experiment	Evaluation level/aggregation	Accuracy (95%CI)	Balanced accuracy (95%CI)	Macro F1 (95%CI)	ROC AUC (95%CI)
Binary presence, view OOF	Individual projected view; patient-cluster bootstrap	0.9478 (0.9302-0.9636)	0.9399 (0.9187-0.9590)	0.9430 (0.9236-0.9603)	0.9768 (0.9649-0.9866)
Binary presence, patient OOF	Case-level aggregation across projected views	0.9770 (0.9605-0.9934)	0.9721 (0.9494-0.9909)	0.9749 (0.9564-0.9928)	0.9953 (0.9903-0.9988)
Binary presence, hold-out ensemble	Internal 20% hold-out ensemble; soft voting	0.9737 (0.9342-1.0000)	0.9643 (0.9107-1.0000)	0.9713 (0.9262-1.0000)	0.9970 (0.9888-1.0000)
3-class severity, patient OOF	Case-level severity classification	0.8711 (0.8395-0.9000)	0.7833 (0.7302-0.8355)	0.7817 (0.7287-0.8330)	0.9328 (0.9070-0.9557)

Confidence intervals were obtained from 10000 patient-level bootstrap resamples; projected views were clustered by patient for the view-level analysis. Multiclass receiver operating characteristic area under the curve used macro-averaged one-vs-rest evaluation. The hold-out ensemble used soft-voting aggregation. OOF: Out-of-fold; ROC AUC: Receiver operating characteristic area under the curve.