Confocal Microwave Imaging algorithms for the early detection of breast cancer are straight-forward to implement using general-purpose computing platforms. However, the large number of radar signals, higher sampling rate and ever-more sophisticated reconstruction algorithms can all significantly slow down the image formation process. This paper focuses on improved computational implementations of Confocal Microwave Imaging algorithms using Graphics Processing Units, in order to accelerate the image formation process. The GPU based implementations have been evaluated by comparing them with both CPU-based sequential implementations and CPU-based parallel implementations. The computational time was reduced by a factor of 250 compared to sequential implementation and a factor of 110 compared to a parallel CPU implementation.