White Blood Cell Segmentation Using Kolmogorov-Arnold Network Coupled with UNet

Leukemia, a cancer of white blood cells (WBC), is dangerous. Disordered WBC detection is critical in Leukemia diagnosis and treatment. Aided computer approach for extracting WBC(s) from the blood cell images effectively assists WBC analysis. One of the popular and powerful methods is to use deep learning to segment WBC(s). Although deep learning models in segmentation tasks achieve considerable performance, recent advanced methods can be applied to enhance the accuracy of the models. The emerging Kolmogorov-Arnold Network (KAN) has recently been an innovative approach to designing deep learning networks. KAN can take the role of a Multi-layer Perceptron (MLP) and can achieve a comparable or better accuracy with a smaller number of parameters. Inspired by the effectiveness of KAN, we propose a novel and effective network named KAN4UNet for WBC segmentation. KAN4UNet is based on the U-shape of UNet and employs KAN for enhanced performance. We evaluate KAN4UNet on three public WBC datasets LISC, KRD-WBC, and Raabin-WBC, and achieve accuracies (Dice score) of 92.78%, 94.96%, and 89.09% respectively, which outperform the performance of representative WBC segmentation methods.