Skull stripping represents an essential pre-processing step in a broad range of neuroimaging processing applications. It is crucial as a result of the complex brain structure as well as variations in MRI intensity. Concerning brain segmentation activities and clinical analysis, accurate removal of the skull region is crucial as it allows for more accurate diagnosis. However, the process of skull stripping is rather difficult, as the brain regions should be detached from the skull carefully. In order to have a complete resolution of this problem, a new technique known as the end-to-end 3D attention residual recurrent U-Net (3D-ARR-U-NET) is proposed. This architecture effectively incorporates recurrent, attention, and residual mechanism into the traditional U-Net model. Thus, incorporating the standard UNIT with such mechanisms is expected to enhance the performance efficacy and speed of the U-NET. Fine tuning of the degree of skull stripping specifically regarding MRI scans. The proposed approach is trained in the dataset called NFBS which can be considered as a benchmark. For the evaluation measures to assess the performance, intersection over union (IOU), dice scores (DSC), specificity, sensitivity, precision, and accuracy were employed. The results of the experiment confirm that in comparison with other approaches in such a field of study, the suggested strategy is more useful and effective. Apparently, the proposed method demonstrates high effectiveness of the section’s separation of the brain from the skull with a high DSC of 1.0000, sensitivity of 0.9811, an IOU of 0.9811, specificity of 0.9971, precision of 0.9971, and accuracy of 0.9979.