High speed detection of breast masses from mammography images at an affordable cost is a problem of practical significance in large volume real-time processing and diagnostic assessments. In this paper, we present a new approach to real-time detection of breast masses by introducing the concept of cognitive cells that has a fully parallel high speed computing architecture realised in a low cost hardware. The prototype system was tested using the Compute Unified Device Architecture (CUDA) that achieved an average speed of 6 ms for processing a single 1024x1024 pixels mammography image. Initial results shows feasibility of using cognitive cells for suspicious breast cancer mass detection in mammograms with superior performances in speed in comparison to other standard methods. We report specificity of 95.25% and the cancer false positives per image as 2.275 for MISC, ASYM, CIRC and SPIC cases, while a relatively lower specificity of 70% and the false positives per image as 2.25 is reported for CALC and ARCH cases of abnormalities.