Skip to main content

Table 2 Examples of reference standards, common performance metrics, and model selection for various tasks

From: Deep learning workflow in radiology: a primer

 

Detection

Segmentation

Classification

Prediction

Features

-Bounding boxes

-Masks

-Lesion patch

-Full image at max diameter

-Radiomics features

-Masks

-Lesion patch

-Radiomic features

-Lesion patch

-Time to recurrence

-Survival time

-TRG

Model architectures

-CNN

-U-Net

-Fully connected

-CNN

Performance metrics

-Intersection over union (IOU)

-Mean average precision (mAP)

-Dice score

-IOU

-Receiver operating characteristic (ROC)

-Accuracy

-ROC curve

-Accuracy

-R2