LAUSR

T his study 1 applied nnU-Net, developed by Isensee et al, 2 to as many as 659 meningioma patient images in total after contrast enhancement for automatic segmentation. Both training and validation data images were acquired on 1.0T, 1.5T and 3.0T MRI scanners of various MR venders including those of other hospitals. The performances of 2D, 3D and attention nnU-Nets were compared to those of corresponding U-Net versions, and the 2D nnU-Net resulted in the highest Dice similarity coef fi cients (DSCs) of 0.922 and 0.893 for internal and external validation sets, respectively. This study also clari fi ed the limitation of this method for small meningio-mas, especially for those smaller than 1 cm 3 in addition to their locations, although the results outperformed previous methods. Some of the results (and methods) are detailed in supplementary materials, which should be referenced.