Skip to main content
Fig. 5 | BMC Medical Informatics and Decision Making

Fig. 5

From: Learning rich features with hybrid loss for brain tumor segmentation

Fig. 5

Visualized comparison between our proposed method and U-net on BRATS 2015 validation set (best view in color). There are three groups of experiments to verify the effectiveness of our proposed model. The first column shows images generated by T1c, T2 and FLAIR with a merge operation. The second column shows the ground truth. The first group (the third and fourth column) shows the segmentation results generated by U-net and the proposed method with cross-entropy loss function. The second group (the fifth and sixth column) shows the segmentation results generated by U-net and the proposed method with sliced Dice loss function. The last group (the last two columns) shows the segmentation results generated by U-net and the proposed method with combined Dice loss function. Pixels labeled in black are background in the last five columns. Each of the other colors represents a tumor region: necrosis (blue), edema (green), non-enhancing (white) and enhancing (red)

Back to article page