Brain tumours are unwanted masses of abnormally grown cells in the human brain. Originate in the brain, which would be primary brain tumours, or migrate from other parts of the body, which would be secondary (metastatic) brain tumours. There are different types of brain tumours such as noncancerous (benign) and cancerous (metastatic) ones. In the era of medical image processing and health technology to perform early detection and estimation of such primary or secondary tumours image segmentation models are much popular. In this paper/work the previous traditional models are replaced with new machine learning and deep learning techniques. The proposed U-Net-based architectures are prevalent marked a trend in the tumour detections. A new module is proposed as DCA with skip connections such as U-Net and its variants with 3D models like pre-trained models with capability of highest degree of integration into any encoder-decoder architectures its performance is intended. The overall performance of 3D net models not only limited to the images of medical image processing but also for electrochemical sensors, nano, and micro-chip level material-layer deposition areas too.