How Do You Know if an AI is Good?

Interactive guide to AI model performance metrics in healthcare

Select a Use Case

Different clinical situations require different performance metrics. Explore how priorities change:

Sepsis Prediction

Early detection of sepsis can be life-saving. High sensitivity is critical to catch all potential cases, while maintaining reasonable specificity to avoid alarm fatigue.

Most Important: Balance between all metrics

Patient Data

Click the circles to change the true diagnosis or model prediction

Patient
Actual Diagnosis
Model Prediction
Outcome

Confusion Matrix

True Positive (TP): 0
False Negative (FN): 0
False Positive (FP): 0
True Negative (TN): 0

Performance Metrics

Accuracy

-
= (TP+TN) / Total

Precision

-
= TP / (TP+FP)

Recall (Sensitivity)

-
= TP / (TP+FN)

Specificity

-
= TN / (TN+FP)

NPV

-
= TN / (TN+FN)

F1-Score

-
= 2×(P×R)/(P+R)