Fig. 1: Example cases—Examples in representative test set cases showing the segmentation predictions from the DeepLab V3 network and the ILD-method. | npj Digital Medicine

Fig. 1: Example cases—Examples in representative test set cases showing the segmentation predictions from the DeepLab V3 network and the ILD-method.

From: Handling missing MRI sequences in deep learning segmentation of brain metastases: a multicenter study

Fig. 1

The image mosaic shows the post-Gd 3D T1-weighted image-series (a), and the predictions as probability maps (voxel-wise ranging from 0.5 to 1 as indicated by the color bar) and performance maps (classified as true negative, false positive, and false negative as specified by the color code) from the DeepLab V3 network (b) and the ILD-method (c). All maps are shown as overlays on a post-Gd 3D T1-weighted image-series. The cases shown here are [first row] a 65-year-old female with malignant melanoma, [second row], 73-year-old male with non-small cell lung cancer (NSCLC), [third row] 66-year-old male with NSCLC, [fourth row] 44-year-old female with NSCLC, [fifth row] 64-year-old female with NSCLC, and [sixth row] 73-year-old male with NSCLC. The blue arrows indicate true positive lesions, while yellow arrows indicate false positive lesions. Note that in the bottom three cases, the DeepLab V3 returns several false positive lesions which are not reported by the ILD-method, thus reflecting the results indicating a superior performance on false positive rate by the ILD-method.

Back to article page