The study introduces a new framework called CISCA that can automatically segnemtn and classify cell instances in histological slices. The U-Net of CISCA was separated into three heads. The first head of CISCA was the classification of pixels into separations between neighboring cells, their cell bodies, and cell background. The second head of CISCA was four distance maps that were lapsed into their four respective directions. The first and second head results are put together using a step that segments each cell. At the third head, cells are able to be synchronously classified into classes. The method’s effectiveness was evaluated using the H&E stain datasets CoNIC, PanNuke, and MoNuSeg. Additionally, the dataset CytoDArk0 was introduced as the first openly annotated dataset for the cell instance slices in Nisssl-stained brain histological images. The findings showed that CISCA performed better than other methods and multiple metric rating systems. Only R churned out better results. CISCA performs far better than Hover-Net, a framework with similar abilities. CISCA’s superiority can be credited to its post-processing abilities, which helped the framework maintain good quality results at 20x and 40x resolutions.