What ROC shows
- Each threshold produces a (FPR, TPR) point.
- TPR = Sensitivity, FPR = 1 - Specificity.
- AUC ~ 0.5 is chance; AUC ~ 1.0 is excellent separation.
Interactive: generate an ROC curve
We simulate scores for "disease" and "no disease" groups and sweep thresholds to plot ROC.
01.23
50200600
Approx AUC (demo): -
ROC
Diagonal line is chance performance.
Real Dental Scenario
AI Caries Detection System Evaluation
A dental AI system scores each tooth surface from 0 (definitely healthy) to 10 (definitely carious). You need to choose a detection threshold: any score at or above this threshold is flagged as caries. The challenge? Lower thresholds catch more cavities but also trigger more false alarms.
Diagnostic Challenge
0 (flag everything)
5.0
10 (flag nothing)
Sensitivity
-
Specificity
-
30 Tooth Surfaces (15 truly carious, 15 healthy)
Correctly detected
Missed (false negative)
False alarm
Correctly ruled out
Live ROC Curve
Move the threshold slider to explore the sensitivity-specificity trade-off.
Dental example
ROC curves help choose thresholds for caries detection scores (from radiographs/AI), balancing missed lesions vs false alarms.