Performance of ML

Performance Metrics

Sensitivity

Sensitivity is a metric that tells us among ALL the positive cases in the dataset, how many of them are successfully identified by the algorithm, i.e. the true positive. In other words, it measures the proportion of accurately-identified positive cases.

You can think of highly sensitive tests as being good for ruling out disease. If someone has a negative result on a highly sensitive algorithm, it is extremely likely that they don’t have the disease since a high sensitive algorithm has low false negative.

Specificity

Specificity measures ALL the negative cases in the dataset, how many of them are successfully identified by the algorithm, i.e. the true negatives. In other words, it measures the proportion of accurately-identified negative cases.

You can think of highly specific tests as being good for ruling in disease. If someone has a positive result on a highly specific test, it is extremely likely that they have the disease since a high specific algorithm has low false positive.

Dice coefficient

The dice coefficient measures the overlap of algorithm output and true labels. It is used to assess the performance of segmentation and localization.