BACKGROUND AND PURPOSE: Current autosegmentation models such as UNets and nnUNets have limitations, including the inability to segment images that are not represented during training and lack of computational efficiency. 3D capsule networks have the potential to address these limitations.
MATERIALS AND METHODS:We used 3430 brain MRIs, acquired in a multi-institutional study, to train and validate our models. We compared our capsule network with standard alternatives, UNets and nnUNets, on the basis of segmentation efficacy (Dice scores), segmentation performance when the image is not well-represented in the training data, performance when the training data are limited, and computational efficiency including required memory and computational speed.
RESULTS:The capsule network segmented the third ventricle, thalamus, and hippocampus with Dice scores of 95%, 94%, and 92%, respectively, which were within 1% of the Dice scores of UNets and nnUNets. The capsule network significantly outperformed UNets in segmenting images that were not well-represented in the training data, with Dice scores 30% higher. The computational memory required for the capsule network is less than one-tenth of the memory required for UNets or nnUNets. The capsule network is also .25% faster to train compared with UNet and nnUNet.
CONCLUSIONS:We developed and validated a capsule network that is effective in segmenting brain images, can segment images that are not well-represented in the training data, and is computationally efficient compared with alternatives.ABBREVIATIONS: CapsNet ¼ capsule network; Conv1 ¼ first network layer made of convolutional operators; ConvCaps3 ¼ third network layer made of convolutional capsules; ConvCaps4 ¼ fourth network layer made of convolutional capsules; DeconvCaps8 ¼ eighth network layer made of deconvolutional capsules; FinalCaps13 ¼ final thirteenth network layer made of capsules; FinalCaps13 ¼ final layer; GPU ¼ graphics processing unit; PrimaryCaps2 ¼ second network layer made of primary capsules N euroanatomic image segmentation is an important component in the management of various neurologic disorders. [1][2][3] Accurate segmentation of anatomic structures on brain MRIs is an essential step in a variety of neurosurgical and radiation therapy procedures. 1,[3][4][5][6] Manual segmentation is time-consuming and is prone to intra-and interobserver variability. 7,8 With the advent of deep learning to automate various image-analysis tasks, 9,10 there has been increasing enthusiasm for using deep learning for brain image autosegmentation. [11][12][13][14] UNets are among the most popular and successful deep learning autosegmentation algorithms. 11,15-17 Despite the broad success of UNets in segmenting anatomic structures across various imaging modalities, they have well-described limitations. UNets perform best on images that closely resemble the images used for training but underperform on images that contain variant