Accurate segmentation and classification of nuclei in histology images of bladder cancer (BC) is crucial for diagnosis and treatment. Hematological analysis gives valuable information about the tumor. However, issues like nuclei overlap and staining inconsistencies make the segmentation and classification challenging. To overcome these challenges, we propose a dual attention-based framework for segmenting and classifying nuclei using a bottom-up tri-decoder approach. Our framework employs three different attention decoder heads to produce semantic, edge, and classification maps. This multi-head attention mechanism enables the model to capture the features and the variations of the images. Additional measures, such as controlled watershed and pixel grouping, are applied to enhance segmentation and classification results. The proposed framework shows better segmentation accuracy and classification reliability with a Dice score of 0.8621 and provides better discrimination between nuclei with an mPQ value of 0.6273. The improved results increase the possibilities of the application of the new AI-based approaches in the diagnosis of bladder cancer and may serve as a basis for further developments in the field of digital pathology.