Abstract:The crossMoDA2023 challenge aims to segment the vestibular schwannoma (sub-divided into intra- and extra-meatal components) and cochlea regions of unlabeled hrT2 scans by leveraging labeled ceT1 scans. In this work, we proposed a 3D multi-style cross-modality segmentation framework for the crossMoDA2023 challenge, including the multi-style translation and self-training segmentation phases. Considering heterogeneous distributions and various image sizes in multi-institutional scans, we first utilize the min-max normalization, voxel size resampling, and center cropping to obtain fixed-size sub-volumes from ceT1 and hrT2 scans for training. Then, we perform the multi-style image translation phase to overcome the intensity distribution discrepancy between unpaired multi-modal scans. Specifically, we design three different translation networks with 2D or 2.5D inputs to generate multi-style and realistic target-like volumes from labeled ceT1 volumes. Finally, we perform the self-training volumetric segmentation phase in the target domain, which employs the nnU-Net framework and iterative self-training method using pseudo-labels for training accurate segmentation models in the unlabeled target domain. On the crossMoDA2023 validation dataset, our method produces promising results and achieves the mean DSC values of 72.78% and 80.64% and ASSD values of 5.85 mm and 0.25 mm for VS tumor and cochlea regions, respectively. Moreover, for intra- and extra-meatal regions, our method achieves the DSC values of 59.77% and 77.14%, respectively.
Abstract:The crossMoDA challenge aims to automatically segment the vestibular schwannoma (VS) tumor and cochlea regions of unlabeled high-resolution T2 scans by leveraging labeled contrast-enhanced T1 scans. The 2022 edition extends the segmentation task by including multi-institutional scans. In this work, we proposed an unpaired cross-modality segmentation framework using data augmentation and hybrid convolutional networks. Considering heterogeneous distributions and various image sizes for multi-institutional scans, we apply the min-max normalization for scaling the intensities of all scans between -1 and 1, and use the voxel size resampling and center cropping to obtain fixed-size sub-volumes for training. We adopt two data augmentation methods for effectively learning the semantic information and generating realistic target domain scans: generative and online data augmentation. For generative data augmentation, we use CUT and CycleGAN to generate two groups of realistic T2 volumes with different details and appearances for supervised segmentation training. For online data augmentation, we design a random tumor signal reducing method for simulating the heterogeneity of VS tumor signals. Furthermore, we utilize an advanced hybrid convolutional network with multi-dimensional convolutions to adaptively learn sparse inter-slice information and dense intra-slice information for accurate volumetric segmentation of VS tumor and cochlea regions in anisotropic scans. On the crossMoDA2022 validation dataset, our method produces promising results and achieves the mean DSC values of 72.47% and 76.48% and ASSD values of 3.42 mm and 0.53 mm for VS tumor and cochlea regions, respectively.