The extent of resection after surgery is a critical prognostic factor for patients with glioblastoma and accurate segmentation of residual tumors from post-operative MRI is essential to assess this. An automated segmentation tool for identifying residual tumor in post-operative MRI shows potential for improving the accuracy of assessing resection extent and tumor regrowth. Building on the success of transformer models, recent studies have explored their application in 3D medical segmentation tasks. In these models, the self-attention mechanism serves as a key component, aiming to capture long-range dependencies more effectively than traditional convolution-based approaches, which are typically limited to local contexts. In this paper, we present a channel-aware encoder-decoder Swin transformer architecture, called CAST, designed for the efficient segmentation of post-operative residual tumors in MRI scans. CAST (i.e. Channel Aware Swin Transformers) features a robust parallel integration of the Swin transformer module and the 3D convolution module at every stage of the network. CAST also has the channel-aware transformer module positioned following the parallel combination of Swin transformer and convolutional blocks at each stage of the network for efficient computing of global multi-head self-attentions (MSAs) along the channel dimension. CAST is specifically designed to leverage channel-specific information through global MSAs and spatial information through local MSAs. We have compared CAST with four CNN-based and five transformer-based networks for the Uppsala University post-operative glioblastoma dataset. CAST has achieved state-of-the-art performance with a mean dice score of \(71.72\%\) .