Event-based White Blood Cell Classification using Convolutional Spiking Neural Networks

Advances in mixed reality and blockchain technology in Metaverse are transforming the digital healthcare by enabling real-time health monitoring and remote medical treatments. This significantly enhances patient care and medical interactions, permitting virtual consultations without the need for physical hospital visits. This study utilized a dynamic vision sensor camera to track white blood cells and employed spiking neural networks to classify and analyze event-based data from nail capillaries. Additionally, the soft-reset integrate-and-fire neuron model, which adjusts the membrane potential by subtracting the threshold post-spike, was used to optimize the training of spiking neural networks. Surrogate gradient learning, which approximates gradients for non-differentiable spiking functions, was also applied to enhance the training process. The results confirmed that the classification performance of white blood cells achieved an accuracy of 80.48%, with precision and recall at 99.65% and 78.47% respectively, and F1-score of 86.62%.