Hybrid Residual Blocks - Transformers Improving Cardiac MRI Segmentation in Deep Learning

Cardiovascular disease is one of the primary causes of mortality and disability. By assisting in early diagnosis, segmentation techniques based on deep learning might help to lessen their severity, although extremely high levels of accuracy are required. Introduce a novel and flexible architecture that raises the bar for semantic segmentation accuracy in this study. Our contributions focus on the merging of strong self-attention mechanisms obtained from Transformer with traditional convolutional neural network (CNN) architecture. This hybrid model demonstrates improved feature representation, contextual understanding, and pixel-wise predictions, setting new standards in semantic segmentation tasks. the average dice coefficients for ACDC dataset by our model, AVG DICE 92.52%, RV 90.77%, MYO 90.00%, LV 96.2S%