Skip to main content
Fig. 7 | BMC Medical Imaging

Fig. 7

From: Deep learning-based segmentation of the lung in MR-images acquired by a stack-of-spirals trajectory at ultra-short echo-times

Fig. 7

Results of the substudy investigating generalizability are presented. The left column shows the anatomical image while on the right the lung labels of both segmentations can be depicted (yellow—manual, blue—automatic, green—consensus). Numbers are the DSC for the respective slice. First row: image of a patient with pneumonia after stem cell transplantation.The consolidations on both lungs are segmented correctly by the network. Second row: example of another patient with pneumonia. Consolidations are segmented correctly. Third row: different slice of the same patient as in the row above. Due to very high signal intensity in the consolidations, the network failed to segment the lung correctly in this slice. Fourth row: patient with large tumor in the lung (indicated by the red circle on the anatomical image). The network did not include the tumor in the lung label. In this case, this is the intended behavior

Back to article page