What is a 2×2 table?
A 2×2 table summarizes two binary variables. In diagnostics, one axis is the true condition (disease present/absent) and the other is the test result (positive/negative).
| Condition + | Condition − | |
|---|---|---|
| Test + | True Positive (TP) | False Positive (FP) |
| Test − | False Negative (FN) | True Negative (TN) |
Real Dental Scenario
AI Caries Detection vs Clinical Examination
Scenario: A dental clinic tested 200 patients for caries using a new AI detection tool. Results were compared against clinical examination (gold standard).
1 Building the 2×2 Table
| Caries Present (Gold Standard +) |
Caries Absent (Gold Standard −) |
Total | |
|---|---|---|---|
| AI Positive | — | — | — |
| AI Negative | — | — | — |
| Total | — | — | — |
2 Sensitivity (True Positive Rate)
3 Specificity (True Negative Rate)
4 Positive Predictive Value (PPV)
5 Negative Predictive Value (NPV)
Clinical Summary: AI Caries Detection Performance
Recommendation: This AI tool shows strong diagnostic performance. However, with a sensitivity of 85.7%, approximately 1 in 7 caries cases may be missed. For screening purposes this is acceptable, but for definitive diagnosis, clinical examination should remain the primary method with AI as a supplementary tool.
Prevalence in sample = 84/200 = 42%. PPV/NPV will shift if prevalence changes in the population.
Interactive calculator
Enter TP/FP/FN/TN (e.g., caries detection test vs reference standard) and compute metrics.
Metric definitions
- Sensitivity = TP / (TP + FN) — among condition+, how often test is +
- Specificity = TN / (TN + FP) — among condition−, how often test is −
- PPV = TP / (TP + FP) — among test+, probability condition is present
- NPV = TN / (TN + FN) — among test−, probability condition is absent
Visual: sensitivity & specificity
Dental example (quick interpretation)
For a caries detection device, high sensitivity reduces missed lesions (low FN), while high specificity reduces false alarms (low FP). The best balance depends on clinical context and downstream costs of errors.