Artificial Neural Networks Based Cell Counting Techniques Using Microscopic Images: A Review

Counting microscopic cells is a time-consuming task and exhausting, but it is important to assess the different experimental conditions and effects on biological structures of interest. Although such objects are able to be identified, the process of manual counting cells sometimes errors because the borderlines are difficult to define precisely. In this review, the challenges and possible neural network-based solutions of cell counting based on microscopic images were examined. Depending on the type of cells and the microscopic methodology used, different neural network architectures have proven to be effective. Fluorescence micrographs of cancer cells, a focal point in our work, pose challenges for neural network-based processing due to the clustered and crowded location of the cells, the blurring of cell borders, and the overlapping cells. The limited size of available databases also hinders artificial intelligence-based automation. Two widely used types of neural network architectures have been highlighted for cell counting: U-Net and YOLO. Based on the comparison, it is advisable to use neural networks based on U-N et. Additionally, considering the combined application of several architectures to create a new framework may offer enhanced solutions.