Fig 1.
Examples of focal liver lesions.
From left to right livers with an adenoma, cyst, HCC, hemangioma, and a metastasis from colorectal carcinoma origin are shown. The top row shows the arterial phases of the DCE-MRI and the bottom row the T2-weighted images. A zoom-in of the lesions is inserted.
Table 1.
Features derived from DCE-MR and T2-weighted images and risk factors.
Fig 2.
Time-to-peak feature maps of a liver with a hemangioma (left) and a HCC (right).
The lesions correspond with the lesions in Fig 1. The red contours show the lesion segmentations.
Table 2.
Classification results for the four different feature sets (sensitivity/specificity), with 19 features selected.
Table 3.
Selected features with the highest ANOVA F-scores.
Table 4.
Confusion matrix of the five class problem, including the sensitivity, specificity and one-versus-other accuracy per lesion class.
Fig 3.
ROC curves of all lesion classes in a one-versus-other approach.
The rest class is calculated as the outcome probabilities of the other four lesions.
Table 5.
Areas under the ROC curve (AUC) for each class including the optimal cut-off value and the corresponding true positive rate (TPR), false positive rate (FPR) and false negative rate (FNR).
Fig 4.
ROC curve of benign-versus-malignant classification problem.
The area under the ROC curve (AUC) is 0.94.