Developing Next-Generation CapsNet Models for Enhanced White Blood Cell Classification in Medical Diagnostics

Abstract

Correct classification of white blood cells (WBCs) is very crucial for diagnosis and monitoring of haematological disorders. Although conventional convolutional neural networks (CNNs) perform, they can struggle with translational invariance and the precise spatial structures required for more complicated medical image categorisation. We aim to address these issues by developing next-generation Capsule Network (CapsNet) models that will provide more accurate and dependability in WBC categorisation. CapsNet models offer significant potential as a development over conventional CNNs because of their special designs that retain the internal spatial linkages of WBC components without compromising the purity of the found features. This work fast separates and sorts WBCs from microscopic images using a modified version of the CapsNet architecture along with dynamic routing and reconstruction regularisation. Thousands of annotated blood smear images from several public medical databases were included in the collection to exhibit a broad spectrum of disorders. We first improved and standardised the pictures using pre-processing methods. We then classified the WBCs using the CapsNet model in their proper groups. With a clear increase in precision and recall, the results of classification accuracy revealed that the CapsNet models outperformed conventional CNN models. The models may be extensively used in real-life medical diagnostic environments as they also operated well with shifting pictures and noise. This work demonstrates how CapsNet models might alter WBC classification, therefore improving diagnosis accuracy and resulting in improved patient outcomes.