A Comparative Analysis of ResNet-Based White Blood Cell Classification Across Multi-Scale Datasets for Enhanced Hematological Diagnostics

Abstract

In this proposed research work, we used the power of deep learning models, ResNetwork to correctly explore WBC. For implementation, six different datasets are collected. In Each datasets images are clicked from different angle and also a combined dataset is made. Some challenges are noticed while creating a new dataset of all the images. The disturbances in the images like overlapping cells and artifacts. ResNet can train the deep layers correctly, it lowers the problem of vanishing gradient. Combined data can enhance the generalization but the performance can be degraded due to versatility. Some classes work good on individual dataset but some typos are shown in rare classes like basophils or blasts. Domain adaptation, data augmentation and explainable AI are the possible suggestions to improve the model. A model alone is not enough for accurate WBC classification. Data variability, class imbalance, and interpretablity must also be handled. Only then can AI-based diagnostics become reliable and scalable.