A Multi-Class Nuclei Segmentation Model using Attention Modules

Accurate and precise segmentation of nuclei is one of the main challenges in clinical pathology due to the multiplex structure of nuclei which includes shape, appearance, and tortuosity pattern. In this study, we proposed a nuclei segmentation model by introducing attention modules within U-Net blocks to automate the digital pathology workflow. We extracted segmented patches from digital pathology images to train our model. We used channel and spatial attention modules to extract and identify the exhaustive relationship within nuclei regions and learn the micro-structure (complex) attributes of the nuclei from the different body organs. Traditional models use simple convolutional layers with kernels for extracting features information but we used a drop-out block within each layer to randomly discard the local block area in the learning phase and enhance the learning performance of the digital pathology image segmentation tasks. We evaluated the performance of our model on the MoNuSAC dataset. Experimental results of our model have shown superior performance compared to state-of-the-art (SoTA) models.