Accurate delineation of residual tumor tissue from post-surgical MR images is crucial for assessing the prognosis of patients with glioblastoma, with the extent of surgical removal being a key prognostic factor. Though nearly accurate post-surgical residual tumor segmentation can be achieved with deep neural architectures, interactive refinement can improve the segmentation further. In this study, we implemented the novel network named Class Attention Augmented Transformers (CAAT) for quantifying the post-operative residual enhanced brain tumors. The 3D segmentation output from the post-operative baseline scan was further corrected with level-set method to reduce the over-segmentation and under-segmentation. The dataset comprises the post-operative glioblastoma data from Uppsala University Hospital, encompassing the baseline and follow-up MRI for each patient. The corrected baseline scan was used to fine-tune the network further, which finally resulted in the improvement of dice scores of the follow-up. The mean dice score with CAAT is 0.6536 and it increased to 0.6601 after level-set correction.