Nuclei are very useful in understanding the micro-environment. The pathologist's effort may be reduced and precise micro-environment observations can be obtained by using an automatic nuclei segmentation approach in clinical trials. Even though the latest deep learning (DL) based segmentation algorithms are generally outperforming more conventional approaches, however, this segmentation task is difficult in situations where nuclei are coherent and overlap. In the current study, we propose a novel nuclei segmentation method that combines the U-Net model based on a convolution neural network (CNN) with an unsharp masking technique for the pre-processing of the source images. The unsharp masking technique enhances the high-frequency information of the source images. An objective function is proposed for the optimization algorithm to further improve the segmentation performance. On the same dataset, the efficacy of the suggested method to segment nuclei is compared with existing methods. The suggested technique achieved IOU(JI), Accuracy, Precision, and F1Score values as 0.8172, 96.05 %, 0.8672, and 0.8994 respectively. The empirical results reveal that the discussed technique is performing better and is superior to the existing techniques.