Table 2 Summary of evaluation metrics used for segmentation performance assessment.
From: A hybrid attention network for accurate breast tumor segmentation in ultrasound images
Metric | Formula | Interpretation |
|---|---|---|
Jaccard Index (IoU) | \(\displaystyle \frac{TP}{TP + FP + FN}\) | Measures overlap between predicted and ground-truth masks. Higher values indicate stronger agreement. |
Accuracy | \(\displaystyle \frac{TP + TN}{TP + TN + FP + FN}\) | Represents the overall proportion of correctly classified pixels. |
Recall (Sensitivity) | \(\displaystyle \frac{TP}{TP + FN}\) | Evaluates the ability to correctly identify lesion pixels. High recall reduces the risk of missed detections. |
Precision | \(\displaystyle \frac{TP}{TP + FP}\) | Reflects the proportion of correctly predicted lesion pixels among all pixels labeled as lesion. |
Dice Coefficient (DSC) | \(\displaystyle \frac{2TP}{2TP + FP + FN}\) | Balances recall and precision, providing an intuitive measure of segmentation overlap. |
Specificity | \(\displaystyle \frac{TN}{TN + FP}\) | Assesses the ability to correctly classify background pixels, minimizing false alarms. |
