Automated Cell Counting System Using Improved Implicit Activation Based U-Net (IA-U-Net)

Abstract

Cell counting refers to any of several techniques used in life sciences, including medical diagnosis and treatment, to count or quantify cells. This is vital for various disease detection, treatment, and other medical research purposes. In general, one can manually count the number of cells in a digital image. However, the manual counting method takes a long time and labor and is costly. Hence, we require an automated cell counting system to boost efficiency, reduce labor expenses, and reduce mistake rates in order to overcome the limitations of human counting. Over the decade, various machine learning and deep learning methods have been proposed for counting cells automatically. However, a handful of algorithms are robust enough to determine the cell area with accuracy due to the tremendous density distribution of the cell in any image. In order to solve the issue of inaccurate approximation, we suggest an enhanced version of U-net. Implicit activation (IA) block is added to the extended U-net to extract more characteristics than regular U-net and improve the accuracy of cell counting. In terms of cell counting accuracy, the simulation results show that our suggested IA-based U-net (IA-U-net) is much better than the original U-net architecture.